Search

Imaging spectroscopy integrates traditional computer vision and spectroscopy into a single system and has gained widespread acceptance as a non-destructive scientific instrument for a wide range of applications. The current state of imaging spectroscopy spans diverse applications including but not limited to air-borne and ground-based computer vision systems. This paper presents the current state of research and industrial applications including precision agriculture, material classification, medical science, forensic science, face recognition and document image analysis, environment monitoring, and remote sensing, which can be aided through imaging spectroscopy. In this regard, we further discuss a comprehensive list of applications of imaging spectroscopy, pre-processing techniques, and spectral image acquisition systems. Likewise, publicly available databases and current software tools for spectral data analysis are also documented in this review. This review paper, therefore, could potentially serve as a reference and roadmap for people looking for literature, databases, applications, and tools to undertake additional research in imaging spectroscopy.

BACKGROUND AND PURPOSE: Resting-state functional MRI (rs-fMRI) studies in Parkinsons disease (PD) patients with freezing of gait (FOG) have implicated dysfunctional connectivity over multiple resting-state networks (RSNs). While these findings provided network-specific insights and information related to the aberrant or altered regional functional connectivity (FC), whether these alterations have any effect on topological reorganization in PD-FOG patients is incompletely understood. Understanding the higher order functional organization, which could be derived from the hub and the rich-club organization of the functional networks, could be crucial to identifying the distinct and unique pattern of the network connectivity associated with PD-FOG. METHODS: In this study, we use rs-fMRI data and graph theoretical approaches to explore the reorganization of RSN topology in PD-FOG when compared to those without FOG. We also compared the higher order functional organization derived using the hub and rich-club measures in the FC networks of these PD-FOG patients to understand whether there is a topological reorganization of these hubs in PD-FOG. RESULTS: We found that the PD-FOG patients showed a noticeable reorganization of hub regions. Regions that are part of the prefrontal cortex, primary somatosensory, motor, and visuomotor coordination areas were some of the regions exhibiting altered hub measures in PD-FOG patients. We also found a significantly altered feeder and local connectivity in PD-FOG. CONCLUSIONS: Overall, our findings demonstrate a widespread topological reorganization and disrupted higher order functional network topology in PD-FOG that may further assist in improving our understanding of functional network disturbances associated with PD-FOG.

Bowel obstruction is a common cause of acute abdominal pain. The development of algorithms for automated detection and characterization of bowel obstruction on CT has been limited by the effort required for manual annotation. Visual image annotation with an eye tracking device may mitigate that limitation. The purpose of this study is to assess the agreement between visual and manual annotations for bowel segmentation and diameter measurement, and to assess agreement with convolutional neural networks (CNNs) trained using that data. Sixty CT scans of 50 patients with bowel obstruction from March to June 2022 were retrospectively included and partitioned into training and test data sets. An eye tracking device was used to record 3-dimensional coordinates within the scans, while a radiologist cast their gaze at the centerline of the bowel, and adjusted the size of a superimposed ROI to approximate the diameter of the bowel. For each scan, 59.4 ± 15.1 segments, 847.9 ± 228.1 gaze locations, and 5.8 ± 1.2 m of bowel were recorded. 2d and 3d CNNs were trained using this data to predict bowel segmentation and diameter maps from the CT scans. For comparisons between two repetitions of visual annotation, CNN predictions, and manual annotations, Dice scores for bowel segmentation ranged from 0.69 ± 0.17 to 0.81 ± 0.04 and intraclass correlations [95% CI] for diameter measurement ranged from 0.672 [0.490-0.782] to 0.940 [0.933-0.947]. Thus, visual image annotation is a promising technique for training CNNs to perform bowel segmentation and diameter measurement in CT scans of patients with bowel obstruction.

Isthmic spondylolysis results in fracture of pars interarticularis of the lumbar spine, found in as many as half of adolescent athletes with persistent low back pain. While computed tomography (CT) is the gold standard for the diagnosis of spondylolysis, the use of ionizing radiation near reproductive organs in young subjects is undesirable. While magnetic resonance imaging (MRI) is preferable, it has lowered sensitivity for detecting the condition. Recently, it has been shown that ultrashort echo time (UTE) MRI can provide markedly improved bone contrast compared to conventional MRI. To take UTE MRI further, we developed supervised deep learning tools to generate (1) CT-like images and (2) saliency maps of fracture probability from UTE MRI, using ex vivo preparation of cadaveric spines. We further compared quantitative metrics of the contrast-to-noise ratio (CNR), mean squared error (MSE), peak signal-to-noise ratio (PSNR), and structural similarity index (SSIM) between UTE MRI (inverted to make the appearance similar to CT) and CT and between CT-like images and CT. Qualitative results demonstrated the feasibility of successfully generating CT-like images from UTE MRI to provide easier interpretability for bone fractures thanks to improved image contrast and CNR. Quantitatively, the mean CNR of bone against defect-filled tissue was 35, 97, and 146 for UTE MRI, CT-like, and CT images, respectively, being significantly higher for CT-like than UTE MRI images. For the image similarity metrics using the CT image as the reference, CT-like images provided a significantly lower mean MSE (0.038 vs. 0.0528), higher mean PSNR (28.6 vs. 16.5), and higher SSIM (0.73 vs. 0.68) compared to UTE MRI images. Additionally, the saliency maps enabled quick detection of the location with probable pars fracture by providing visual cues to the reader. This proof-of-concept study is limited to the data from ex vivo samples, and additional work in human subjects with spondylolysis would be

Pomegranate farming's enhanced predictive accuracy for nutrient deficiency and fertilizer recommendation was achieved through systematic evaluations of various machine learning algorithms on a comprehensive dataset. Remarkably, the Random Forest algorithm demonstrated exceptional accuracy of 99.85%, reaffirming its potential for accurate predictions in NPK value-based deficiency and disease outbreak prediction. In the context of climate-driven pest and disease outbreak forecasts, a meticulous exploration of machine-learning algorithms was conducted. Among these, the Random Forest Classifier stood out with an accuracy of 99.00%, aligning with the study's accuracy-driven focus and highlighting its capability to address climate-induced pest and disease dynamics. For "Pomegranate Disease Detection" using Convolutional Neural Networks (CNN), the VGG16 architecture yielded an accuracy of 98.44%, showcasing the power of automated disease identification in advancing agricultural practices. In "Pomegranate Quality Detection and Analysis," Convolutional Neural Networks (CNN) with VGG16 achieved an impressive 97% accuracy, demonstrating its potential for enhancing fruit quality assessment in pomegranate cultivation. These findings collectively underscore the significance of advanced algorithms in optimizing various facets of pomegranate farming, from nutrient management to disease detection and quality assessment.

This research paper introduces an innovative approach to improve the quality and sustainability of coconut farming and exports in Sri Lanka. It employs advanced image processing techniques to detect, classify, and grade pests and diseases early in coconut palms. This allows for swift interventions and reduces the need for harsh chemical treatments, promoting eco-friendly farming practices. Furthermore, the study goes beyond pest control to evaluate optimal conditions for coconut growth, considering factors like soil quality, water availability, and climate. It empowers farmers with insights to maximize coconut palm yield. Additionally, the system incorporates a growth prediction component using historical data and machine learning, enabling farmers to plan and allocate resources effectively. By combining early pest detection, pest management, growth classification, and predictive analysis, this research offers a comprehensive strategy to enhance Sri Lanka's coconut quality for export. This approach not only improves product quality but also safeguards the industry's sustainability by reducing economic losses and ecological impact. Leveraging cutting-edge tools like image processing and machine learning, this research aims to boost efficiency, economic viability, and international competitiveness in Sri Lanka's coconut farming sector.

To aim at the realization of a helical fusion reactor, we study multi-objective optimization of coil shapes, which satisfy various requirements. In the magnetic field configuration created by these coils, several unfavorable examples are found: some of them have magnetic islands or doublet configurations. In order to automatically and quickly exclude such cases that hinder the optimization, we have developed a new method to detect unfavorable magnetic surfaces by using image recognition. Binarization and erosion are performed as preprocessing, and then blanks of magnetic islands and doublets are extracted as recognition targets. Consequently, we have developed a classifier with high performance. Using this trained classifier, we have shown that almost all cases with unfavorable magnetic surfaces in various magnetic configurations can be excluded in a short time and with high precision.

Breakdown and rearrangement of a primary vortex street shed from a circular cylinder in the far wake are experimentally examined for 70 < R < 154 (R is the Reynolds number based on the diameter of the cylinder). According to the vorticity fields obtained using digital image processing for visualized flow fields, the primary vortex street breaks down into a nearly parallel shear flow of Gaussian profile at a certain downstream distance, before a secondary vortex street of larger scale appears further downstream. The process leading to the nearly parallel flow can be explanined as the evolution of the vortex regions of an inviscid fluid if we invoke the observation that the distance between the two rows in the primary vortex street increases with the downstream distance. Numerical computations with the discrete vortex method also support this explanation. Next, the wavelengths, a_1 and a_2, of the primary and secondary vortex streets are calculated from the above vorticity fields, and are also measured from the flow patterns obtained using the aluminium dust method. The ratio a_2/a_1 decreases with increasing R, and ranges from 1.7 to 2.6. Moreover, the wavelength a_2 is a little smaller than that of the most unstable mode in the linear stability theory applied to the above nearly parallel flow. The speeds of the vortex streets relative to the fluid at infinity are also measured, and are 0.12U-0.19U and 0.03U-0.10U for the primary and secondary ones, respectively. Here U is the speed of the cylinder.

0000-0002-0593-8810, For the automatic measurement of DNA length in the video captured by fluorescence microscope, image processing method using pix2pix is developed. To increase the accuracy of the measurement in low-resolution video, a method of inputting multiple time-series frames to pix2pix are proposed. To generate the training data, molecular dynamics simulation is employed to prepare dummy video. The trained pix2pix by multiple frames input can generate more accurate images in a scene of DNA breakage than the pix2pix trained by single frame input.

Breast cancer is the most common cancer in Indonesia. One way to detect it early is by screening using mammography. Previous trials showed that mammography screening in women aged 40-49 years could reduce breast cancer mortality by 25%. However, misdiagnosis may occur abaout breast density and the patient’s physical size due to machines. In addition, human reader errors can occur concerning the reader's experience and perception. Therefore, diagnostic aids are needed to distinguish benign and malignant cases and receive appropriate treatment. The methodology in this research consists of three stages: preprocessing, texture feature extraction, and data classification. Preprocessing consists of filtering, contrast, cropping, and resizing, while texture feature extraction consist of Histogram and (Gray Level Co-occurrence Matrix) GLCM. Data classification using Support Vector Machines (SVM), Naive Bayes, Multi-Layer Perceptron (MLP), Multiclass classifier, and Random Forest methods with Weka Machine Learning software. It produces an accuracy of 62.00%, 62.00%, 88.00%, 82.00%, and 100.00%, respectively. The results of data classification using the Random Forest method show that the accuracy, specifications, and specificity reach 100%. Random forest can be used as the most optimal classification method to distinguish benign and malignant cases based on texture features in mammography images using Weka Machine Learning software. This can help radiologists and medical professionals to diagnose cases and take further steps, such as therapy.

Facial attribute transformation, the ability to modify specific facial attributes in images and videos, has gained significant attention in computer vision and image processing. We explored in depth the state-of-the-art techniques based on multiple individual and sequential Latent Transformers to modify the latent representation of the image. In this work, we introduce a novel approach that utilizes a unified architecture capable of learning multiple facial attributes simultaneously in a single Transformer model. By training a single model for multiple attributes, we aim to enhance efficiency, particularly in terms of speed, while maintaining competitive overall quality and outperforming some of the metrics. The disentanglement of attributes and preservation of identity are essential goals. Our results demonstrate the successful learning of 20 attributes with a unified Latent Transformer, showcasing significant improvements in speed, identity preservation, realism, and close-to-original overall quality. Through subjective study, our model demonstrates more appealing and preferred results compared to the baseline. As a summary, this work contributes to the field of facial attribute transformation by providing a more elegant and scalable solution, opening avenues for further advancements in disentangled face editing in images and videos.

Sepsis is a severe condition that occurs as a result of an uncontrolled host response to infection. With the high fatality rates and morbidity associated with sepsis and septic shock, there is a pressing need to improve the accuracy of early detection in prehospital and emergency department settings. In the most severe form of sepsis, known as septic shock, the survival rate decreases by approximately 8% for every hour that the patient is untreated. Additionally, in the 30% of cases, treatment fails, making it imperative to rapidly detect and identify the type of bacteria responsible for the infection. To address this issue, it pressing to develop new platforms that will speed up the diagnosis time, and will yield rapid results on the bacteria antibiotic resistance spectrum, thus leading to accurate treatment prescription. In this thesis we investigate the possibility of using a microdevice developed for antibiotic susceptibility of urinary tract infection samples. Differently from urine samples, blood samples have high abundance of human cells and low quantity of bacteria. Therefore, the main issue comes from handling large number of cells and cell residue on chip. For this a pre-loading protocol was developed, where there was a centrifugation step and various protocols of filtering and cleaning the sample. Additionally, we investigate different concentrations of saponin and sodium cholate in the lysis solution, which was used to explode the red blood cells present in the sample. Several chip designs were evaluated to determine the most efficient one for trapping the largest number of bacteria and processing the highest volume of sample in one hour. To imitate the conditions that will be found in the clinical samples, the blood was first diluted 1:3 with blood culture media (BCM), then subjected to medium-speed spinning for two minutes, followed by low-speed spinning for ten minutes to pellet all bacteria before adding the lysis mix. The minimum concentration of ba, La sepsis es una enfermedad grave que se produce como resultado de una respuesta incontrolada del huésped a la infección. Con las altas tasas de mortalidad y morbilidad asociadas a la sepsis y el choque séptico, existe una necesidad de mejorar la precisión de la detección precoz en los entornos prehospitalarios y de urgencias. En la forma más grave de sepsis, conocida como shock séptico, la tasa de supervivencia disminuye aproximadamente un 8% por cada hora que el paciente no es tratado. Además, en el 30% de los casos, el tratamiento falla, por lo que resulta imprescindible detectar e identificar rápidamente el tipo de bacterias responsables de la infección. Para abordar este problema, es necesario desarrollar nuevas plataformas que acelerarán el tiempo de diagnóstico y darán resultados rápidos en el espectro de resistencia a los antibióticos de las bacterias, dando lugar así a una prescripción precisa del tratamiento. En esta tesis investigamos la posibilidad de utilizar un microdispositivo desarrollado para la susceptibilidad a los antibióticos de muestras de infección del tracto urinario. A diferencia de las muestras de orina, las muestras de sangre tienen gran abundancia de glóbulos rojos y una baja cantidad de bacterias. Por tanto, el problema principal proviene de la manipulación de un gran número de células y residuos celulares en el chip. Para ello se desarrolló un protocolo para tratar la muestra antes de introducirla en el chip, donde existía un paso de centrifugación y varios protocolos de filtrado y limpieza de la muestra. Además, investigamos diferentes concentraciones de saponina y colado de sodio en la solución de lisis, que se utilizó para explotar los glóbulos rojos presentes en la muestra. Se evaluaron varios diseños de chips para determinar el más eficiente para atrapar el mayor número de bacterias y procesar el mayor volumen de muestra en una hora. Para imitar las condiciones que se encontrarán en las muestras clínicas, la sangre se diluyó primer, La sèpsia és una malaltia greu que es produeix com a resultat d'una resposta incontrolada de l'hoste a la infecció. Amb les altes taxes de mortalitat i morbiditat associades a la sèpsia i el xoc sèptic, hi ha una necessitat de millorar la precisió de la detecció precoç en els entorns prehospitalaris i d'urgències. En la forma més greu de sèpsia, coneguda com a xoc sèptic, la taxa de supervivència disminueix aproximadament un 8% per cada hora que el pacient no és tractat. A més, en el 30% dels casos, el tractament falla, per la qual cosa és imprescindible detectar i identificar ràpidament el tipus de bacteris responsables de la infecció. Per abordar aquest problema, cal desenvolupar noves plataformes que acceleraran el temps de diagnòstic i donaran resultats ràpids en l'espectre de resistència als antibiòtics dels bacteris, donant lloc així a una prescripció precisa del tractament. En aquesta tesi investiguem la possibilitat d'utilitzar un microdispositiu desenvolupat per a la susceptibilitat als antibiòtics de mostres d'infecció del tracte urinari. A diferència de les mostres d'orina, les mostres de sang tenen una gran abundància de glòbuls vermells i una baixa quantitat de bacteris. Per tant, el problema principal prové de la manipulació d'un gran nombre de cèl·lules i residus cel·lulars al xip. Per a això es va desenvolupar un protocol per tractar la mostra abans de introduir-la al xip, on hi havia un pas de centrifugació i diversos protocols de filtrat i neteja de la mostra. A més, investiguem diferents concentracions de saponina i colat de sodi a la solució de lisi, que es va utilitzar per explotar els glòbuls vermells presents a la mostra. Es van avaluar diversos dissenys de xips per determinar el més eficient per atrapar el major nombre de bacteris i processar el volum més alt de mostra en una hora. Per imitar les condicions que es trobaran a les mostres clíniques, la sang es va diluir primer 1:3 amb medis de cultiu de sang (BCM), després es va sotmetre a una

El trencament prematur de membranes és una de les complicacions associades a la cirurgia fetal, concretament a la fetoscopia, que suposa un gran risc per al fetus durant l'embaràs. Des de l'Institut de Ginecologia i Obstetrícia de l'Hospital Clínic, es planteja una línia de recerca centrada en l'estudi de les imatges ultrasòniques (adquirides en dies previs a l'operació), amb la finalitat de crear un model de classificació capaç d'identificar aquelles dones amb major probabilitat de patir el trencament de la bossa amniòtica. Amb aquest motiu, es planteja iniciar l'estudi per l'extracció de les estructures membranoses uterines de les imatges, per a obtenir el conjunt de dades amb el qual entrenar el model de classificació. Amb la finalitat de donar suport a aquest propòsit, neix la idea d'aquest projecte enfocat en el desenvolupament d'un programari de segmentació d'imatges la finalitat del qual és automatitzar i agilitzar el procés d'extracció de les estructures membranoses. En aquest projecte, s'empra MATLAB com a eina de processament d'imatge i anàlisi de dades. Es valoren diferents tècniques per al tractament de les ecografies, triant aquelles que més s'adeqüin a les necessitats requerides. Com a resultat, s'obtenen dos mètodes diferents de segmentació que s'avaluen mitjançant mètriques estadístiques per a triar aquell que resulti òptim en el propòsit del projecte. Finalment, s'implementa el mètode en noves imatges ultrasòniques per a analitzar la seva eficàcia i precisió., La rotura prematura de membranas es una de las complicaciones asociadas a la cirugía fetal, concretamente en la fetoscopia, que supone un gran riesgo para el feto durante el embarazo. Desde el Instituto de Ginecología y Obstetricia del Hospital Clínic, se plantea una línea de investigación centrada en el estudio de las imágenes ultrasónicas (adquiridas en días previos a la operación), con el fin de crear un modelo de clasificación capaz de identificar aquellas mujeres con mayor probabilidad de sufrir la rotura de la bolsa amniótica. Con este motivo, se plantea iniciar el estudio por la extracción de las estructuras membranosas uterinas de las imágenes, para obtener el conjunto de datos con el que entrenar el modelo de clasificación. Con el fin de apoyar este propósito, nace la idea de este proyecto enfocado en el desarrollo de un software de segmentación de imágenes cuya finalidad es automatizar y agilizar el proceso de extracción de las estructuras membranosas. En este proyecto, se emplea MATLAB como herramienta de procesado de imagen y análisis de datos. Se valoran diferentes técnicas para el tratamiento de las ecografías, escogiendo aquellas que más se adecuen a las necesidades requeridas. Como resultado, se obtienen dos métodos diferentes de segmentación que se evalúan mediante métricas estadísticas para escoger aquel que resulte óptimo en el propósito del proyecto. Finalmente, se implementa el método en nuevas imágenes ultrasónicas para analizar su eficacia y precisión., Premature rupture of membranes is one of the complications associated with fetal surgery, specifically in fetoscopy, which poses a great risk to the fetus during pregnancy. From the Institute of Gynecology and Obstetrics of the Hospital Clínic, a line of research focused on the study of ultrasound images (acquired in the days prior to the operation) is proposed, in order to create a classification model capable of identifying those women most likely to suffer rupture of the amniotic sac. For this reason, it is proposed to start the study by extracting the uterine membranous structures from the images, to obtain the data set with which to train the classification model. In order to support this purpose, the idea of this project was born, focused on the development of an image segmentation software whose purpose is to automate and speed up the process of extraction of membranous structures. In this project, MATLAB is used as a tool for image processing and data analysis. Different techniques for the treatment of ultrasound scans are evaluated, choosing those that best suit the needs required. As a result, two different segmentation methods are obtained that are evaluated using statistical metrics to choose the one that is optimal for the purpose of the project. Finally, the method is implemented in new ultrasound images to analyze its efficiency and accuracy.

En la actualidad, es cada vez más frecuente que la población reciba información de acuerdo con sus preferencias, por lo que cada vez es más común la planificación de dispositivos que faciliten la captación de datos para mejorar el análisis de las preferencias. En los centros comerciales, hoy en día se ha visto la eficiencia en los sistemas de gestión que ofrecen información a través de sensores y cámaras de seguridad, lo que implica la posibilidad de estudiar los diferentes indicadores asociados con las actividades realizadas por los clientes, para entender mejor los movimientos dentro del centro comercial. Con base en ello, se planteó como objetivo realizar un estudio que permita la mejora del layout y tránsito de persones en comercios y centros comerciales. Lo que permitió la realización de un estudio estilo propuesta, y así se consideró que la planificación de un modelo de negocios inmerso en la creación de un sistema de dispositivos que permitan la monitorización de la información seg-mentada sobre patrones de movimiento y compra en centros comerciales y pequeños comercios. Esto, en líneas generales, podría ser beneficioso para la población y podría ser de utilidad para garantizar el crecimiento económico dentro de dichos negocios.

In astronomy and the aerospace world, determining the orientation or attitude of a system in space is of great importance, whether it's simply to determine observations or to successfully carry out space missions. One of the devices commonly used is the star tracker. These devices allow the determination of attitude based on observed stars, using a preloaded catalogue in the system that identifies which stars are captured. In this work, the catalogue preloaded in the star tracker has been specifically designed for the proposed requirements, a catalogue derived from the Hipparcos catalogue. The starting point of the design is a Lost In Space (LIS) situation, where the only reference available are the stars that this device can capture and identify correctly. Traditionally, they have been systems with limited accessibility, only for large organizations or major space projects, due to their high cost and complexity. With the rapid development and advancement of technologies in recent years, today, we can find high-performance technological elements at a reduced cost. In this case, a functional star tracker has been developed and implemented using a Raspberry Pi with the Raspberry Pi OS operating system, a camera, and an optical system. This setup allows capturing images of the celestial sphere. To capture the images, the MATLAB® package "MATLAB Support Package for Raspberry Pi Hardware" has been used, a package that allows direct connection between the Raspberry Pi operating system and MATLAB®, through a WiFi network or an ethernet cable. To determine the attitude, it's necessary to find the match between the stars of the preloaded catalogue and the captured stars. In this case, the three brightest stars captured after processing, correcting, and characterizing the image. To correct the light frame and eliminate random noise and outliers, the master frames of the dark, bias, and flat images have been obtained. Additionally, an average filter has been used to soften and

The constant transformation of our surroundings often goes unnoticed, but it occurs nonetheless. Imperceptibly, our skin undergoes slight color changes as our hearts beat, and our heads subtly move with each breath. While these alterations may escape our eyes, they are captured by cameras. We can magnify the variation in those small signals to extract important information such as heartbeat rate, which in turn can be used in many real-life applications such as polygraph tests. The polygraph, or lie detector test, is a tool that measures physiological changes in an individual to detect deception. This thesis aimed to explore the feasibility of using video information to extract heartbeat rates, replacing the traditional sensors that are commonly used to extract this information. Through a series of experiments involving videos of people standing still, and a video of a real-life suspect that was lying, it was consistently possible to enhance the color changes that occur in a person's skin when their heartbeats. Though it was not possible to develop a fully-functioning lie detector test, this thesis proves that the motion and color magnification technique is reliable even when applied to videos of individuals engaged in speech and movement. It also proved that it is an interesting solution that has the potential of revolutionizing how interviews and interrogations are conducted.

In a world of ephemeral moments, our brain diligently sieves through a cascade of experiences, like a skilled gold prospector searching for precious nuggets amidst the river's relentless flow. This study delves into the elusive "moment of memorability"---a fleeting, yet vital instant where experiences are prioritised for consolidation in our memory. By transforming subjects' encoding phase electroencephalography (EEG) signals into the visual domain using scaleograms and leveraging deep learning techniques, we investigate the neural signatures that underpin this moment, with the aim of predicting subject-specific recognition of video. Our findings not only support the involvement of theta band (4-8Hz) oscillations over the right temporal lobe in the encoding of declarative memory, but also support the existence of a distinct moment of memorability, akin to the gold nuggets that define our personal river of experiences.

In a world where sensory threads weave an endless tapestry of multi-modal data, the human brain stands as the masterful weaver of meaning. As we wade through this tempest of input, our brain spins these threads into an intelligible internal representation and holds on tight to what it deems important. But what, exactly, makes certain threads more important than others? And how can we predict their significance? Memorability is the tensile strength of the threads that tie us to the world. It is a proxy for human importance, indicating which threads the human brain will curate and retain with exceptional fidelity. This research investigates these multisensory threads by exploring the influence of audio, visual, and textual modalities on predicting video memorability, and how the interplay between them can influence the overall memorability of a given piece of content. The findings suggest that, while visual data may dominate our sensory experience, it is the underlying conceptual essence that truly holds the key to memorability. This thesis leverages state-of-the-art image synthesis techniques to distill and examine this essence, creating surrogate dreams of video scenes to facilitate the disentanglement of conceptual and perceptual elements of memorability. The work also leverages human EEG data to explore the possibility of a moment of memorability—a moment of encoding that corresponds to a remembering moment—which we expect to exist due to the temporal nature of the world and the natural encoding limits of our brains. The previously murky relationship between the two core means of remembrance---recognition and recall---are reconciled by conducting a novel video memorability drawing task. The research sheds new light on the nature of multi-modal memorability, providing a deeper understanding of how our brain processes and retains information in a complex sensory world. By uncovering the conceptual essence that lies at the heart of memorability, it opens up new avenu

Handwriting is an art form that most people learn at an early age. Each person's writing style is unique with small changes as we grow older and as our mood changes. Here we analyse handwritten text in a culturally significant personal diary. We compare changes in handwriting and relate this to the sentiment of the written material and to the topic of diary entries. We identify handwritten text from digitised images and generate a canonical form for words using shape matching to compare how the same handwritten word appears over a period of time. For determining the sentiment of diary entries, we use the Hedonometer, a dictionary-based approach to scoring sentiment. We apply these techniques to the historical diary entries of Rosamond Jacob (1888-1960), an Irish writer and political activist whose daily diary entries report on the major events in Ireland during the first half of the last century.

Auto-colourisation is the ill-posed problem of creating a plausible full-colour image from a grey-scale prior. The current state of the art utilises image-to-image Generative Adversarial Networks (GANs). The standard method for training colourisation is reformulating RGB images into a luminance prior and two-channel chrominance supervisory signal. However, progress in auto-colourisation is inherently limited by multiple prerequisite dilemmas, where unsolved problems are mutual prerequisites. This thesis advances the field of colourisation on three fronts: architecture, measures, and data. Changes are recommended to common GAN colourisation architectures. Firstly, removing batch normalisation from the discriminator to allow the discriminator to learn the primary statistics of plausible colour images. Secondly, eliminating the direct L1 loss on the generator as L1 will limit the discovery of the plausible colour manifold. The lack of an objective measure of plausible colourisation necessitates resource-intensive human evaluation and repurposed objective measures from other fields. There is no consensus on the best objective measure due to a knowledge gap regarding how well objective measures model the mean human opinion of plausible colourisation. An extensible data set of human-evaluated colourisations, the Human Evaluated Colourisation Dataset (HECD) is presented. The results from this dataset are compared to the commonly-used objective measures and uncover a poor correlation between the objective measures and mean human opinion. The HECD can assess the future appropriateness of proposed objective measures. An interactive tool supplied with the HECD allows for a first exploration of the space of plausible colourisation. Finally, it will be shown that the luminance channel is not representative of the legacy black-and-white images that will be presented to models when deployed; This leads to out-of-distribution errors in all three channels of the final colour image.

This thesis presents the research work conducted on developing algo- rithms capable of training neural networks for image classification and re- gression in low supervision settings. The research was conducted on publicly available benchmark image datasets as well as real world data with appli- cations to herbage quality estimation in an agri-tech scope at the VistaMilk SFI centre. Topics include label noise and web-crawled datasets where some images have an incorrect classification label, semi-supervised learning where only a small part of the available images have been annotated by humans and unsupervised learning where the images are not annotated. The principal contributions are summarized as follows. Label noise: a study highlighting the dual in- and out-of-distribution nature of web-noise; a noise detection metric than can independently retrieve each noise type; an observation of the linear separability of in- and out-of-distribution images in unsupervised contrastive feature spaces; two noise-robust algorithms DSOS and SNCF that iteratively improve the state-of-the-art accuracy on the mini-Webvision dataset. Semi-supervised learning: we use unsupervised features to propagate labels from a few labeled examples to the entire dataset; ReLaB an algorithm that allows to decrease the classification error up to 8% with one labeled representative image on CIFAR-10. Biomass composition estimation from images: two semi-supervised approaches that utilize unlabeled images either through an approximate annotator or by adapting semi-supervised algorithms from the image classification litterature. To scale the biomass to drone images, we use super-resolution paired with semi-supervised learning. Early results on grass biomass estimation show the feasibility of automating the process with accuracies on par or better than human experts. The conclusion of the thesis will summarize the research contributions and discuss thoughts on future research that I believe should be tackl

In this work, we approach the problem of cardiac Magnetic Resonance Imaging (MRI) image reconstruction from undersampled k-space. This is an inherently ill-posed problem leading to a variety of noise and aliasing artifacts if not appropriately addressed. We propose a two-step double-stream processing pipeline that first reconstructs a noisy sample from the undersampled k-space (frequency domain) using the inverse Fourier transform. Second, in the spatial domain we train a denoising GNA-UNET (enhanced by Group Normalization and Attention layers) on the noisy aliased and fully sampled image data using the Mean Square Error loss function. We achieve competitive results on the leaderboard and show that the algorithmic combination proposed is effective in high-quality MRI reconstruction from undersampled cardiac long-axis and short-axis complex k-space data.

Obtaining accurate body weight (BW) is crucial for management decisions yet can be a challenge for cow–calf producers. Fast-evolving technologies such as depth sensing have been identified as low-cost sensors for agricultural applications but have not been widely validated for U.S. beef cattle. This study aimed to (1) estimate the body volume of mature beef cows from depth images, (2) quantify BW and metabolic weight (MBW) from image-projected body volume, and (3) classify body condition scores (BCS) from image-obtained measurements using a machine-learning-based approach. Fifty-eight crossbred cows with a mean BW of 410.0 ± 60.3 kg and were between 4 and 6 yr of age were used for data collection between May and December 2021. A low-cost, commercially available depth sensor was used to collect top-view depth images. Images were processed to obtain cattle biometric measurements, including MBW, body length, average height, maximum body width, dorsal area, and projected body volume. The dataset was partitioned into training and testing datasets using an 80%:20% ratio. Using the training dataset, linear regression models were developed between image-projected body volume and BW measurements. Results were used to test BW predictions for the testing dataset. A machine-learning-based multivariate analysis was performed with 29 algorithms from eight classifiers to classify BCS using multiple inputs conveniently obtained from the cows and the depth images. A feature selection algorithm was performed to rank the relevance of each input to the BCS. Results demonstrated a strong positive correlation between the image-projected cow body volume and the measured BW (r = 0.9166). The regression between the cow body volume and the measured BW had a co-efficient of determination (R2) of 0.83 and a 19.2 ± 13.50 kg mean absolute error (MAE) of prediction. When applying the regression to the testing dataset, an increase in the MAE of the predicted BW (22.7 ± 13.44 kg) but a slightly imp

Handwriting is an art form that most people learn at an early age. Each person's writing style is unique with small changes as we grow older and as our mood changes. Here we analyse handwritten text in a culturally significant personal diary. We compare changes in handwriting and relate this to the sentiment of the written material and to the topic of diary entries. We identify handwritten text from digitised images and generate a canonical form for words using shape matching to compare how the same handwritten word appears over a period of time. For determining the sentiment of diary entries, we use the Hedonometer, a dictionary-based approach to scoring sentiment. We apply these techniques to the historical diary entries of Rosamond Jacob (1888-1960), an Irish writer and political activist whose daily diary entries report on the major events in Ireland during the first half of the last century.

Improving river monitoring techniques is critical given the concomitant impact of climate change, population growth, and pollution over the last years. Turbidity is one of the most significant metrics for water quality characteristics. In river basins, high turbidity values can be indicative of both organic and inorganic materials. Turbidity is often used as a proxy for transport of suspended particles and associated fluxes of hydrophobic pollutants in a wide range of hydrological conditions. However, it is demanding to estimate suspended sediment yields in rivers because of the high variability along stream of suspended sediment concentrations. Traditional methods, such as gravimetric analysis, are time-consuming, expensive, often discontinuous in space and time and influenced by human errors or instrumental limitations.Remote sensing techniques are a suitable alternative to point measurements. Satellite remote sensing allows to study the spatial and temporal variations of water status parameters, but it is limited by the spatial and temporal resolution of the satellites considered. Low range systems can help increase the resolution of the imagery used for this purpose. In particular, the use of optical cameras can significantly reduce the monitoring cost and exponentially increase the information on water bodies health and hydrological dynamics, offering a large amount of data distributed in time and space. Nonetheless, all optical sensing methods are strongly affected by many environmental constraints (light, good optical transmission, visibility, etc.), which make them currently not always suitable for regular long-term monitoring of turbidity in rivers. The main goal of the monitoring procedure identified in this work is to avoid all these constraints, by processing the camera image to use it as a real measurement data. In this work, an image processing procedure has been identified by exploiting the water surface reflectance properties to estimate wat

With the rapid development of modern computer technology, and gradually combined with the textile industry, the application of modern computer technology in the field of textile is increasingly extensive, which makes textile production gradually move towards the road of automation development. This paper proposes an automatic detection method of simple weave fabric density based on computer image vision. Computer vision and digital image processing technology are used to analyze and identify the simple weave fabric's warp and weft yarn information and calculate the fabric density. To avoid the phenomenon of warp and weft yarn skew, a method of fabric skew correction based on the Radon transform is proposed. The optimal decomposition order of these four fabrics is k = 2, k = 5, and k = 3. The decomposition series is k. It is found that the relative error of both warp and weft density is about 1.00%. Most of the data obtained by the method of correlation coefficient curve to determine the optimal decomposition series are consistent with the results of the energy curve method. The relative error of the density test results of No. 3 fabric, No. 6 fabric, and No. 7 fabric is higher than 10%, and the relative error of No. 3 fabric is the highest, reaching 66%. This shows serious errors in these three fabrics' warp and weft density. To solve the problems of simple weave fabric density detection, the corresponding algorithm is used to solve the problems. Finally, good results are obtained, which verifies the feasibility of this method. It is significant to realize the automatic measurement of fabric density in textile factories.

Coconut production and consumption rate of India being very high, contributes significantly towards the Indian economy; thus, ensuring the quality of allied products proves to be a necessity. Intelligent quality evaluation and grading system for copra images is proposed to classify copra, the dried coconut kernel into various categories using an image processing approach. The proposed method is useful in the current scenario, as quality deterioration during drying and storing of copra is a challenging task. Inferior quality of copra due to the presence of moisture, mold, fungi, bacteria and sulphur may adversely affect the shelf life thus deteriorating the quality of final products and human health. The grading system automatically segments the region of interest of copra images using SUSAN filter and classifies them into usable and unusable categories. The unusable categories of copra considered in the proposed method include copra with moisture, mold, wrinkles and sulphur. Various features of copra images were collected and analyzed. The selected features of copra images were used for training and classification using SVM classifier. The proposed method has been evaluated using a real database and the results are promising.

Drilling of composite structures is one of the most commonly used operations in the manufacturing of aircraft. Currently, most aerospace drilling operations are performed either manually or by extensive programming and fine-tuning of typical industrial robots. Developing fully autonomous robotic drilling cells powered by sensory systems can significantly improve the efficiency, reliability, and accuracy of aerospace composite manufacturing. To this end, the present study aims at developing vision-based techniques to enable two main functionalities: (1) accurate measurement of robot pose and workpiece placement for robot error correction, and (2) automated inspection of drilled holes using deep learning techniques. Off-the-shelf industrial robots have relatively low absolute accuracy (error~1 mm) and cannot readily satisfy the strict tolerance requirements of aerospace drilling. To provide an optimum resolution, the vision system is calibrated over the robot’s workspace using a laser tracker. Circular targets on the spindle are tracked through this system. Different target localization methods for robot motion are compared with laser tracker’s accuracy in distance and pose measurements. Online robot error correction using relative motion towards part fixtures is implemented. A similar approach is used to correct errors between cooperative robots in an offline manner. The developed methodologies in both online and offline error correction methods are tested to demonstrate their accuracy and effectiveness. Vision-based inspection systems efficiently inspect drilled holes in aerospace composites. However, detecting damage in images is challenging due to dark and textured surfaces in Carbon Fiber Reinforced Polymers (CFRPs). This study presents a fully autonomous system to detect and segment damages and cracks around drilled CFRP holes. The system includes: 1) automated multi-light imaging to illuminate holes in varying light, 2) image processing for hole profile, damage

This Master’s thesis investigates the application of dynamically generated procedural terrain textures for texturing 3D representations of the Earth’s surface. The study explores techniques to overcome limitations of the currently most common method – projecting satellite imagery onto the mesh – such as insufficient resolution for close-up views and challenges in accommodating external lighting models. Textures for sand, rock and grass were generated procedurally on the GPU. Aliasing was prevented using a clamping technique, dynamically changing the level of detail when freely navigating across diverse landscapes. The general color of each terrain type was extracted from the satellite images, guided by land cover rasters, in a process where shadows were eliminated using HSV color space conversion and filtering. The procedurally generated textures provide significantly more details than the satellite images in close-up views, while missing some information in medium- to far-distance views, due to the satellite images containing information lacking in the 3D mesh. A qualitative analysis spanning six data sets from diverse global locations demonstrates that the proposed methods are applicable across a range of landscapes and climates.

Teknologiska framsteg påverkar tillverkningsindustrin genom att integrera tester för kretskortens status, lödning och elektriska egenskaper. En pick-and-place-maskin används för att verifiera komponentpositioner på kretskortet, men underhåll och re-paration av dessa maskiner medför stora kostnader. Industry 4.0 introducerar auto-matiseringslösningar som Internet of Things, Big Data, dataanalys och cybersäker-het, vilket kräver uppgradering av utrustning. Att erbjuda prisvärda uppgraderingar för äldre maskiner är avgörande för små och medelstora företags konkurrenskraft. Validering av kretskortkomponenterna på ett kostnadseffektiv och tillförlitligt sätt kan optimera kretskortsindustri.Det finns idag framgångsrika metoder för att lösa liknande problem. Konvolutionella neurala nätverk, finkornig bildigenkänning, funktionsmatchning samt funktions-extraktion är exempel på dessa. För att verifiera korrekt positionering av en krets-kortskomponent utnyttjas en automatisk maskin som systematiskt inspekterar kretskortet med hjälp av en pick-and-place fil som inkluderar exakta koordinater för varje position.Denna studie använder bildigenkänning för att finna en ekonomisk implementerbar lösning på identifiering av kretskortskomponenter. Detta för att undersöka om mo-biltelefonkameran är en möjlig implementationslösning för små till medelstora före-tag att identifiera kretskort. En mobilkamera användes för att ta bilder på kretskor-ten. Detta gjordes i varierande ljussättning, vinklar samt avstånd från kretskortet. En algoritm skapades för att undersöka ifall komponenterna är korrekt placerade på kretskortet. Segmentering, morfologiska processer och kunskapsbaserade metoder används i denna algoritm.Resultatet tyder på att det är en lämplig lösning att använda bildbehandling för att identifiera kretskortskomponenters position. Att identifiera korrekt satta eller fel-aktigt satta komponenter är möjligt förutsatt att bilderna är tagna med referensnära inställningar. Därav kan små, Technological advances are impacting the manufacturing industry by integrating tests for circuit board status, soldering, and electrical properties. A pick-and-place machine is used to verify components positions on the circuit board, but mainte-nance and repair of these machines incurs large costs. Industry 4.0 introduces auto-mation solutions such as the Internet of Things, Big Data, data analysis, and cyberse-curity, which require equipment upgrades. Offering affordable upgrades for older machines is critical to the competitiveness of small and medium-sized businesses. Validation of the circuit board components in a cost-effective and reliable way can optimize the circuit board industry.Today there are successful methods of solving similar problems. Convolutional neu-ral networks, fine-grained image recognition, feature matching, and feature extrac-tion are examples of these. To verify the correct positioning of a circuit board com-ponent, an automatic machine is used that systematically inspects the circuit board using a pick-and-place file that includes exact coordinates for each position.This study uses image recognition to find an economically implementable solution to the identification of printed circuit board components. This is to investigate whether the mobile phone camera is a possible implementation solution for small to me-dium-sized companies to identify circuit boards. The mobile phone camera was used to capture images of the circuit board. This was done in varying lighting, angles, and distance from the circuit board. An algorithm was created to check if the compo-nents are correctly placed on the circuit board. Segmentation, morphological pro-cesses, and knowledge-based methods are used in this algorithm.The result suggests that using image processing to identify the position of circuit board components is a suitable solution. Identifying correctly set or incorrectly set components is possible provided that the images are taken with reference-clos

Background. This Thesis addresses the critical issue of disease management in ricecrops, a key factor in ensuring both food security and the livelihoods of farmers. Objectives. The primary focus of this research is to tackle the often-overlooked challenge of precise disease localization within rice plants by harnessing the power of deep learning techniques. The primary goal is not only to classify diseases accurately but also to pinpoint their exact locations, a vital aspect of effective disease management. The research encompasses early disease detection, classification, andthe precise identification of disease locations, all of which are crucial components of a comprehensive disease management strategy. Methods. To establish the reliability of the proposed model, a rigorous validation process is conducted using standardized datasets of rice plant diseases. Two fundamental research questions guide this study: (1) Can deep learning effectively achieve early disease detection, accurate disease classification, and precise localizationof rice plant diseases, especially in scenarios involving multiple diseases? (2) Which deep learning architecture demonstrates the highest level of accuracy in both disease diagnosis and localization? The performance of the model is evaluated through the application of three deep learning architectures: Masked RCNN, YOLO V8, and SegFormer. Results. These models are assessed based on their training and validation accuracy and loss, with specific metrics as follows: For Masked RCNN, the model achieves a training accuracy of 91.25% and a validation accuracy of 87.80%, with corresponding training and validation losses of 0.3215 and 0.4426. YOLO V8 demonstrates a training accuracy of 85.50% and a validation accuracy of 80.20%, with training andvalidation losses of 0.4212 and 0.5623, respectively. SegFormer shows a training accuracy of 78.75% and a validation accuracy of 75.30%, with training and validation losses of 0.5678 and 0.6741, respect

We studied the benefits of a nine-component (9C) seismic survey over a landslide-prone area in southwest of Sweden to retrieve ultrahigh-resolution shear wave reflection images of the subsurface as well as crucial information on physical properties of the sediments. A complete, 1 m shot and receiver spacing, multicomponent 2-D seismic profile was acquired using three-component microelectromechanical-system-based landstreamer receivers, and a 5-kg sledgehammer strike in vertical and horizontal orientations as three-component seismic source. Given the rich number of shear wave reflections observed on all the 9C data, the processing work focused on their retrievals. It revealed three distinct reflections, two of which associated with coarse-grained materials and one with an extremely undulating bedrock surface. Given the extremely slow shear wave velocities on the order of 60–100 m s−1, we obtained ultrahigh-resolution shear wave sections avoiding temporal and spatial aliasing. Imaging results suggest vertical-source and horizontal-radial receiver (V–HR), and horizontal-transverse source–receiver orientations (HT–HT) provided the most optimum images of the subsurface. A non-hyperbolic algorithm was applied to the normal-moveout corrections justified by the traveltime differences of the bedrock reflection in different shear wave sections. The improved images by incorporating the anisotropy term suggest that the data set reveals moderate shear wave anisotropy along some portions of the profile. The Vp/Vs ratio obtained by using bedrock reflection in P- and S-wave sections suggests values ranging 10–16, which implies high water content. Areas with lower Vp/Vs coincides with greater anisotropic parameters and this can indicate disturbed clays or presence of sensitive clays.

Background. This Thesis addresses the critical issue of disease management in ricecrops, a key factor in ensuring both food security and the livelihoods of farmers. Objectives. The primary focus of this research is to tackle the often-overlooked challenge of precise disease localization within rice plants by harnessing the power of deep learning techniques. The primary goal is not only to classify diseases accurately but also to pinpoint their exact locations, a vital aspect of effective disease management. The research encompasses early disease detection, classification, andthe precise identification of disease locations, all of which are crucial components of a comprehensive disease management strategy. Methods. To establish the reliability of the proposed model, a rigorous validation process is conducted using standardized datasets of rice plant diseases. Two fundamental research questions guide this study: (1) Can deep learning effectively achieve early disease detection, accurate disease classification, and precise localizationof rice plant diseases, especially in scenarios involving multiple diseases? (2) Which deep learning architecture demonstrates the highest level of accuracy in both disease diagnosis and localization? The performance of the model is evaluated through the application of three deep learning architectures: Masked RCNN, YOLO V8, and SegFormer. Results. These models are assessed based on their training and validation accuracy and loss, with specific metrics as follows: For Masked RCNN, the model achieves a training accuracy of 91.25% and a validation accuracy of 87.80%, with corresponding training and validation losses of 0.3215 and 0.4426. YOLO V8 demonstrates a training accuracy of 85.50% and a validation accuracy of 80.20%, with training andvalidation losses of 0.4212 and 0.5623, respectively. SegFormer shows a training accuracy of 78.75% and a validation accuracy of 75.30%, with training and validation losses of 0.5678 and 0.6741, respect

This paper proposes a novel multimodal deep weakly-supervised learning framework, SinkholeNet, to classify and localize sinkhole(s) in high-resolution RGB-slope aerial images. The SinkholeNet first employs a multimodal Convolutional Neural Network (CNN) architecture that simultaneously extracts features from the input RGB image and ground slope map and then fuses the extracted features. It then uses an improved ShuffleNet architecture on the fused features to classify patches as sinkholes or non-sinkholes. Finally, the last extracted feature maps, belonging to the sinkhole class, are used as input of gradient-weighted class activation mapping (Grad-CAM) to localize sinkhole(s) in a weakly-supervised setting. The proposed weakly-supervised framework intends to increase the available labeled data for training and decrease the cost of human annotation. We also introduce a novel publicly available weakly labeled sinkhole dataset comprising RGB-slope paired image patches to support reproducible research. The experimental results on the newly introduced dataset show that the SinkholeNet outperforms the other methods considered in this paper both for sinkhole classification and localization. © 2023 National Authority of Remote Sensing & Space Science

This paper proposes a novel multimodal deep weakly-supervised learning framework, SinkholeNet, to classify and localize sinkhole(s) in high-resolution RGB-slope aerial images. The SinkholeNet first employs a multimodal Convolutional Neural Network (CNN) architecture that simultaneously extracts features from the input RGB image and ground slope map and then fuses the extracted features. It then uses an improved ShuffleNet architecture on the fused features to classify patches as sinkholes or non-sinkholes. Finally, the last extracted feature maps, belonging to the sinkhole class, are used as input of gradient-weighted class activation mapping (Grad-CAM) to localize sinkhole(s) in a weakly-supervised setting. The proposed weakly-supervised framework intends to increase the available labeled data for training and decrease the cost of human annotation. We also introduce a novel publicly available weakly labeled sinkhole dataset comprising RGB-slope paired image patches to support reproducible research. The experimental results on the newly introduced dataset show that the SinkholeNet outperforms the other methods considered in this paper both for sinkhole classification and localization. © 2023 National Authority of Remote Sensing & Space Science

Background. This Thesis addresses the critical issue of disease management in ricecrops, a key factor in ensuring both food security and the livelihoods of farmers. Objectives. The primary focus of this research is to tackle the often-overlooked challenge of precise disease localization within rice plants by harnessing the power of deep learning techniques. The primary goal is not only to classify diseases accurately but also to pinpoint their exact locations, a vital aspect of effective disease management. The research encompasses early disease detection, classification, andthe precise identification of disease locations, all of which are crucial components of a comprehensive disease management strategy. Methods. To establish the reliability of the proposed model, a rigorous validation process is conducted using standardized datasets of rice plant diseases. Two fundamental research questions guide this study: (1) Can deep learning effectively achieve early disease detection, accurate disease classification, and precise localizationof rice plant diseases, especially in scenarios involving multiple diseases? (2) Which deep learning architecture demonstrates the highest level of accuracy in both disease diagnosis and localization? The performance of the model is evaluated through the application of three deep learning architectures: Masked RCNN, YOLO V8, and SegFormer. Results. These models are assessed based on their training and validation accuracy and loss, with specific metrics as follows: For Masked RCNN, the model achieves a training accuracy of 91.25% and a validation accuracy of 87.80%, with corresponding training and validation losses of 0.3215 and 0.4426. YOLO V8 demonstrates a training accuracy of 85.50% and a validation accuracy of 80.20%, with training andvalidation losses of 0.4212 and 0.5623, respectively. SegFormer shows a training accuracy of 78.75% and a validation accuracy of 75.30%, with training and validation losses of 0.5678 and 0.6741, respect

Background. This Thesis addresses the critical issue of disease management in ricecrops, a key factor in ensuring both food security and the livelihoods of farmers. Objectives. The primary focus of this research is to tackle the often-overlooked challenge of precise disease localization within rice plants by harnessing the power of deep learning techniques. The primary goal is not only to classify diseases accurately but also to pinpoint their exact locations, a vital aspect of effective disease management. The research encompasses early disease detection, classification, andthe precise identification of disease locations, all of which are crucial components of a comprehensive disease management strategy. Methods. To establish the reliability of the proposed model, a rigorous validation process is conducted using standardized datasets of rice plant diseases. Two fundamental research questions guide this study: (1) Can deep learning effectively achieve early disease detection, accurate disease classification, and precise localizationof rice plant diseases, especially in scenarios involving multiple diseases? (2) Which deep learning architecture demonstrates the highest level of accuracy in both disease diagnosis and localization? The performance of the model is evaluated through the application of three deep learning architectures: Masked RCNN, YOLO V8, and SegFormer. Results. These models are assessed based on their training and validation accuracy and loss, with specific metrics as follows: For Masked RCNN, the model achieves a training accuracy of 91.25% and a validation accuracy of 87.80%, with corresponding training and validation losses of 0.3215 and 0.4426. YOLO V8 demonstrates a training accuracy of 85.50% and a validation accuracy of 80.20%, with training andvalidation losses of 0.4212 and 0.5623, respectively. SegFormer shows a training accuracy of 78.75% and a validation accuracy of 75.30%, with training and validation losses of 0.5678 and 0.6741, respect

This paper proposes a novel multimodal deep weakly-supervised learning framework, SinkholeNet, to classify and localize sinkhole(s) in high-resolution RGB-slope aerial images. The SinkholeNet first employs a multimodal Convolutional Neural Network (CNN) architecture that simultaneously extracts features from the input RGB image and ground slope map and then fuses the extracted features. It then uses an improved ShuffleNet architecture on the fused features to classify patches as sinkholes or non-sinkholes. Finally, the last extracted feature maps, belonging to the sinkhole class, are used as input of gradient-weighted class activation mapping (Grad-CAM) to localize sinkhole(s) in a weakly-supervised setting. The proposed weakly-supervised framework intends to increase the available labeled data for training and decrease the cost of human annotation. We also introduce a novel publicly available weakly labeled sinkhole dataset comprising RGB-slope paired image patches to support reproducible research. The experimental results on the newly introduced dataset show that the SinkholeNet outperforms the other methods considered in this paper both for sinkhole classification and localization. © 2023 National Authority of Remote Sensing & Space Science

Background. This Thesis addresses the critical issue of disease management in ricecrops, a key factor in ensuring both food security and the livelihoods of farmers. Objectives. The primary focus of this research is to tackle the often-overlooked challenge of precise disease localization within rice plants by harnessing the power of deep learning techniques. The primary goal is not only to classify diseases accurately but also to pinpoint their exact locations, a vital aspect of effective disease management. The research encompasses early disease detection, classification, andthe precise identification of disease locations, all of which are crucial components of a comprehensive disease management strategy. Methods. To establish the reliability of the proposed model, a rigorous validation process is conducted using standardized datasets of rice plant diseases. Two fundamental research questions guide this study: (1) Can deep learning effectively achieve early disease detection, accurate disease classification, and precise localizationof rice plant diseases, especially in scenarios involving multiple diseases? (2) Which deep learning architecture demonstrates the highest level of accuracy in both disease diagnosis and localization? The performance of the model is evaluated through the application of three deep learning architectures: Masked RCNN, YOLO V8, and SegFormer. Results. These models are assessed based on their training and validation accuracy and loss, with specific metrics as follows: For Masked RCNN, the model achieves a training accuracy of 91.25% and a validation accuracy of 87.80%, with corresponding training and validation losses of 0.3215 and 0.4426. YOLO V8 demonstrates a training accuracy of 85.50% and a validation accuracy of 80.20%, with training andvalidation losses of 0.4212 and 0.5623, respectively. SegFormer shows a training accuracy of 78.75% and a validation accuracy of 75.30%, with training and validation losses of 0.5678 and 0.6741, respect

This paper proposes a novel multimodal deep weakly-supervised learning framework, SinkholeNet, to classify and localize sinkhole(s) in high-resolution RGB-slope aerial images. The SinkholeNet first employs a multimodal Convolutional Neural Network (CNN) architecture that simultaneously extracts features from the input RGB image and ground slope map and then fuses the extracted features. It then uses an improved ShuffleNet architecture on the fused features to classify patches as sinkholes or non-sinkholes. Finally, the last extracted feature maps, belonging to the sinkhole class, are used as input of gradient-weighted class activation mapping (Grad-CAM) to localize sinkhole(s) in a weakly-supervised setting. The proposed weakly-supervised framework intends to increase the available labeled data for training and decrease the cost of human annotation. We also introduce a novel publicly available weakly labeled sinkhole dataset comprising RGB-slope paired image patches to support reproducible research. The experimental results on the newly introduced dataset show that the SinkholeNet outperforms the other methods considered in this paper both for sinkhole classification and localization. © 2023 National Authority of Remote Sensing & Space Science

[Resumen] En entornos con bajo número de Reynolds (Re), los robots nadadores necesitan realizar movimientos no recíprocos para propulsarse. El método fundamental para lograr este tipo de movimiento es mediante la generación de ondas progresivas (también conocidas como ondas viajeras) que recorren el flagelo del robot desde la cabeza al extremo libre. Este trabajo se centra en una forma sencilla de generar ondas progresivas en condiciones de bajo Re que consiste en la oscilación periódica de un flagelo flexible pasivo. Para las pruebas de propulsión, se presenta un prototipo de robot nadador con actuación única basada en el mecanismo yugo escocés y palanca, que permite convertir el movimiento de rotación de un motor en una oscilación angular que viaja a lo largo del flagelo. Asimismo, se desarrolla en MATLAB un algoritmo basado en imágenes para identificar el movimiento realizado por el robot, es decir, para determinar las características del nadador en la propulsión. Los resultados experimentales demuestran que el robot es capaz de realizar un movimiento no recíproco en condiciones de bajo Re., [Abstract] Within low Reynolds number (Re) environments, swimming robots need to perform non-reciprocal motions to propel themselves. The fundamental method to achieve this type of motion is by generating progressive waves (also known as traveling waves) that travel along the robot flagellum from the head to the free end. This work focuses on a simple way to generate progressive waves under low Re conditions that consists of the periodic oscillation of a passive flexible flagellum. For propulsion testing, a prototype of single-acting swimming robot based on the scotch-yoke-lever mechanism is presented, which allows to convert the rotational motion of a motor into an angular oscillation traveling along the flagellum. Also, an image-based algorithm is developed in MATLAB to identify the motion performed by the robot, i.e., to determine the characteristics of the swimmer in propulsion. Experimental results show that the robot is able to perform a non-reciprocal motion under low Re conditions.

Multimodal measurements are increasingly becoming common in the domain of spectral sensing and imaging for fresh produce. Often multiple sensors are expected to carry complementary information which allows precise estimation of responses. In this study, a novel case of multimodal hyperspectral imaging is described where two different spectral cameras working in the complementary spectral ranges were integrated into a fully standalone system for spectral imaging for fresh produce analysis. Furthermore, a comparative analysis of different multiblock predictive modelling approaches for fusing data from these two complementary spectral cameras is demonstrated. Both multiblock latent space and multiblock variable selection approaches to identify key variables of interest was examined and compared with the analysis carried out on individual data blocks. Prediction of the soluble solids content in grapes was used to demonstrate the application. The presented approach can increase the applications of multimodal hyperspectral imaging for non-destructive analysis.

Producción Científica, With the advancement in information technology, digital data stealing and duplication have become easier. Over a trillion bytes of data are generated and shared on social media through the internet in a single day, and the authenticity of digital data is currently a major problem. Cryptography and image watermarking are domains that provide multiple security services, such as authenticity, integrity, and privacy. In this paper, a digital image watermarking technique is proposed that employs the least significant bit (LSB) and canny edge detection method. The proposed method provides better security services and it is computationally less expensive, which is the demand of today’s world. The major contribution of this method is to find suitable places for watermarking embedding and provides additional watermark security by scrambling the watermark image. A digital image is divided into non-overlapping blocks, and the gradient is calculated for each block. Then convolution masks are applied to find the gradient direction and magnitude, and non-maximum suppression is applied. Finally, LSB is used to embed the watermark in the hysteresis step. Furthermore, additional security is provided by scrambling the watermark signal using our chaotic substitution box. The proposed technique is more secure because of LSB’s high payload and watermark embedding feature after a canny edge detection filter. The canny edge gradient direction and magnitude find how many bits will be embedded. To test the performance of the proposed technique, several image processing, and geometrical attacks are performed. The proposed method shows high robustness to image processing and geometrical attacks.

Producción Científica, Automated dental imaging interpretation is one of the most prolific areas of research using artificial intelligence. X-ray imaging systems have enabled dental clinicians to identify dental diseases. However, the manual process of dental disease assessment is tedious and error-prone when diagnosed by inexperienced dentists. Thus, researchers have employed different advanced computer vision techniques, as well as machine and deep learning models for dental disease diagnoses using X-ray imagery. In this regard, a lightweight Mask-RCNN model is proposed for periapical disease detection. The proposed model is constructed in two parts: a lightweight modified MobileNet-v2 backbone and region-based network (RPN) are proposed for periapical disease localization on a small dataset. To measure the effectiveness of the proposed model, the lightweight Mask-RCNN is evaluated on a custom annotated dataset comprising images of five different types of periapical lesions. The results reveal that the model can detect and localize periapical lesions with an overall accuracy of 94%, a mean average precision of 85%, and a mean insection over a union of 71.0%. The proposed model improves the detection, classification, and localization accuracy significantly using a smaller number of images compared to existing methods and outperforms state-of-the-art approaches.

We studied the benefits of a nine-component (9C) seismic survey over a landslide-prone area in southwest of Sweden to retrieve ultrahigh-resolution shear wave reflection images of the subsurface as well as crucial information on physical properties of the sediments. A complete, 1 m shot and receiver spacing, multicomponent 2-D seismic profile was acquired using three-component microelectromechanical-system-based landstreamer receivers, and a 5-kg sledgehammer strike in vertical and horizontal orientations as three-component seismic source. Given the rich number of shear wave reflections observed on all the 9C data, the processing work focused on their retrievals. It revealed three distinct reflections, two of which associated with coarse-grained materials and one with an extremely undulating bedrock surface. Given the extremely slow shear wave velocities on the order of 60–100 m s−1, we obtained ultrahigh-resolution shear wave sections avoiding temporal and spatial aliasing. Imaging results suggest vertical-source and horizontal-radial receiver (V–HR), and horizontal-transverse source–receiver orientations (HT–HT) provided the most optimum images of the subsurface. A non-hyperbolic algorithm was applied to the normal-moveout corrections justified by the traveltime differences of the bedrock reflection in different shear wave sections. The improved images by incorporating the anisotropy term suggest that the data set reveals moderate shear wave anisotropy along some portions of the profile. The Vp/Vs ratio obtained by using bedrock reflection in P- and S-wave sections suggests values ranging 10–16, which implies high water content. Areas with lower Vp/Vs coincides with greater anisotropic parameters and this can indicate disturbed clays or presence of sensitive clays.

The interplay between solidification and convection, which are usually strongly coupled, occurs via many different mechanisms resulting in very complex dynamics. Melt convection changes the solutal field near the solidification front leading to different microstructures or formation of freckle defects. Quantitative of dendritic structure evolution and melt flow during in situ solidification experiments is rather challenging and requires new/improved approaches to image processing. We present an image processing algorithm designed for quantitative analysis of meso-scale solidification of metal alloys in thin cells via X-ray imaging. Our methodology enables one to identify the bulk liquid volume, liquid channels and cavities, and separate them from the solidified structures. It also enables morphological analysis within the solid domain, including automatic decomposition into dominant grains by orientation and connectivity. Furthermore, convective plumes within the bulk liquid can also be studied. The applied image filters enable the developed code (will be made open-source) to reliably operate even for single images with low signal- and contrast-to-noise ratio at low image resolution. This is demonstrated by applying the code to several in situ dynamic X-ray imaging experiments involving a solidifying gallium-indium alloy in a thin cell. We show that primary spacings, grain (and global) dendrite orientation statistics, convective plume parametrization, etc. can be obtained. The limitations of the presented approach are also explained.

Original video camera data, and time-averaged, beam-hardening-corrected, drift-corrected dynamic and static UFXCT image data used in the associated publication; code used for the final processing; and the final processed data. More details are found in the publication and in the info in the respective folders.