Search

Medical imaging is crucial for many modern clinical applications such as diagnostics, disease progression monitoring, and treatment planning. For a large number of these applications, image segmentation serve as the first processing step, followed by quantitative image feature extraction to assess size, volume, shape characteristics, and intensity-related features. These quantitative features provide valuable information for interpretation of pathology and prognostics. However, the segmentation process is mostly done manually, which is time-consuming and lacks consistency. Automated and semi-automated methods have been proposed over the years to increase the efficiency, consistency, and accuracy of segmentation. Among the many methods, there are simple ones like thresholding, region-growing, watershed and more advanced ones like active contour methods, active shape model, and more recently, deep learning-based methods. Deep learning models have yielded a new generation of image segmentation methods with remarkable performance improvements, typically beating standard and traditional machine learning methods in various benchmarks. Another branch of medical image analysis is radiomics. A radiomics approach refers to methods that extracts features from medical images using data-characterisation algorithms, which can be utilized to discover and quantify characteristics of tumors. In this field, deep learning-based approaches are also appearing, showing promising improvements compared to using handcrafted features., This thesis aims to investigate and develop deep learning-based methods for the use in radiation oncology to facilitate the efficient analysis of volumetric medical image datasets. In particular, the segmentation performance of several state-of-the-art convolutional neural networks (CNN) and more recent transformer-based models for segmenting both normal and abnormal structures in PET images were studied and compared. For cerebellum segmentation, we discovered a modified U-Net that has significantly better segmentation performance than the original U-Net and V-Net. For head and neck cancer (HNC) lesion segmentation, we proposed a U-Net-CBAM that slightly outperforms the U-Net and several state-of-the-art transformer-based models. We also proposed a combined localization and segmentation approach for efficient segmentation of the pelvis in CT images with variant volume sizes and image qualities, achieving up to four times faster processing speed compared to a standard approach. Furthermore, this single object combined localization and segmentation approach is extended to process multiple objects in a scan at once. Specifically, the algorithm is applied and studied in the context of lumbar and thoracic vertebrae segmentation. In addition, we explored the performance of using CNN as well as a transformer-based models for HNC outcome prediction and found that the transformer-based model shows the best prognostic performance., In conclusion, our joint localization and segmentation approach for pelvis and lumbar and thoracic vertebrae demonstrated a novel way of utilizing CNNs to substantially improve the segmentation efficiency. We assessed the segmentation performance difference of several CNN and transformer based models on two different data sets. The difference can be substantial. Therefore, testing different models before starting a new project is imperative. In addition, comparison between CNN models and transformer based models on HNC outcome prediction performance showed the potential of transformer based models for radiomics applications. Furthermore, annotation quality, data size and class balance were found to play an important role in deep learning-based segmentation and classification performance, respectively.

Now and in the future, a face mask is a very important strategy to protect people when a new contagious life threatens disease spread through the air appears. Currently, there is a serious health emergency because of the coronavirus disease 2019 (COVID-19) epidemic. The negative consequences of this pandemic need to be protected in public areas. Numerous methods are advised by the World Health Organization (WHO) to reduce infection rates and prevent depleting the available medical resources in the absence of efficient antivirals. Wearing masks is a non-pharmaceutical strategy to lessen the susceptibility to COVID-19 infection. This research aims to create a face mask identification system that is efficient and uses deep learning, which has proven to be beneficial in many real-world applications. This system has also used a transfer learning method with the MobileNetV2 model to classify people who wear face masks properly, wear face masks improperly, and are without masks. The results demonstrate that the proposed system has an accuracy of 99.4% which is higher than current systems.

A robust structural health monitoring approach measuring the structural responses of bridges such as displacements, strains etc. helps to ensure their safety and serviceability. Static and dynamic loads from vehicles and pedestrians influence the instantaneous responses of bridges, while thermal loads from daily and seasonal temperature variations influence bridge long-term responses. Vision based monitoring (VBM) is an emerging non-contact, non-destructive monitoring approach. It utilizes cameras to capture sequential images of the structure under load and suitable image processing algorithms for target tracking. VBM has shown promising accuracy in static and dynamic response measurements of bridges, however, the evidence of its accuracy in thermal response measurements is limited. This research illustrates the results of laboratory experiments implementing VBM for thermal response measurements. Thermal responses of a laboratory truss are monitored with VBM and contact sensors such as thermocou-ples and linear variable differential transformers (LVDT). Cyclic temperature loads are applied to the truss to simulate daily temperature variations. The truss is monitored with GoPro cameras and contact sensors. Measured response trends by VBM and LVDT are comparable, indicating the accuracy of VBM to measure thermal responses. Thermal responses measured by VBM are higher than those of LVDT, signifying requirement for measurement resolution enhancement. The measurement resolution of VBM is 0.099 mm/˚C and LVDT1 is 0.041 mm/˚C respectively. This discrepancy can be attributed to non-identical targets of VBM and LVDT, resolution of the camera, efficiency of the feature tracking algorithm and robustness of LVDT output. This case study illustrates the feasibility and challenges of VBM for thermal response measurement.

Spin-Transfer-Torque (STT) MRAM is configured with high access speed, readily hybrid integration and guaranteed endurance/retention, which becomes one of the most promising candidates for next generation memory. However, the major challenge of STT-MRAM is its high switching energy consumption, accompanied with unreliable writing operation. This work exploits two emerging design paradigms entitled approximation and speculation MRAM. A novel design scheme with timing speculation and approximate storage is proposed to improve energy efficiency in MRAM write operation by reducing write time. Simulation results show that latency and energy loss in entire write operation can be significantly reduced. The proposed technique obtains 84%/57.4% in the symmetric and unidirectional writing operations. Meanwhile, timing speculative method improves the failure/reliability in writing by re-write strategy. Related MRAM macro specification is applied to image processing, which obtains 10% (58.8%-67.2%) of the improvement of precision in image and higher image quality is obtained under the same accuracy rate.

Field imagery is an effective way to capture the state of the entire field; yet, current field inspection approaches, when accounting for image resolution and processing speed, using existent imaging systems, do not always enable real-time field inspection. This project involves the innovation of novel technologies by using an FPGA-based image processing (FIP) device that eliminates the technical limitations of the current agricultural imaging services available in the market and will lead to the development of a market-ready service solution. The FIP prototype developed in this study was tested in both a laboratory and outdoor environment by using a digital single-lens reflex (DSLR) camera and web camera, respectively, as the reference system. The FIP system had a high accuracy with a Lin’s concordance correlation coefficient of 0.99 and 0.91 for the DLSR and web camera reference system, respectively. The proposed technology has the potential to provide on-the-spot decisions, which in turn, will improve the compatibility and sustainability of different land-based systems.

Civil steel structures and infrastructures, such as truss railway bridges, are often subject to potential damage, mainly due to fatigue phenomena and corrosion. Therefore, damage detection algorithms should be designed and appropriately implemented to increase their structural health. Today, the vast amount of information provided by data processing techniques and measurements coming from a monitoring system constitutes a possible tool for damage identification in terms of both detection and description. For this reason, the research activity aims to develop a methodology for a preliminary description of the damage in steel railway bridges induced by fatigue phenomena. The proposed approach is developed through an integration of global and local procedures. At the global scale, vibration-based procedures will be applied to improve a forecast numerical model and, subsequently, to identify the zones most involved in fatigue problems. At the local scale, careful and refined local identification will be pursued via image processing techniques whose evidence will be analyzed and described through nonlinear numerical models. A case study of a historical railway bridge in Spain will illustrate the methodology’s performance, potentiality, and critical issue.

One of the primary issues with inside the agricultural region is crop sicknesses and automated detection is crucial for crop monitoring. Plant leaves generally display the maximum ailment symptoms, however professional laboratory leaf analysis is high priced and time-consuming. According to the Food and Agriculture Organization (FAO), agricultural pests lessen crop yields international via way of means of 20 to 40% according to year. Smart farming is good solution for farmers is to apply artificial intelligence techniques along with modern statistics and communication technologies to get rid of those dangerous insect infestations. Farmers can substantially lessen financial losses via way of means of treating plants directly and detecting sicknesses and pests in apple, mango and graphs leaves appropriately and timely. Image type accuracy has progressed notably because of latest advances in deep studying-primarily based totally convolutional neural networks (CNN). This article evolved strategies primarily based totally on deep studying to hit upon sicknesses and pests in apples, mango, and grape leaves. These strategies have been influenced via way of means of the achievement of CNNs in photograph type.

In this study, a combined system of electronic nose (e-nose) and computer vision was developed for the detection of adulteration in extra virgin olive oil (EVOO). The canola oil was blended with the pure EVOO to provide adulterations at four levels of 5, 10, 15, and 20%. Data collection was carried out using an e-nose system containing 13 metal oxide gas sensors, and a computer vision system. Applying principal component analysis (PCA) on the e-nose-extracted features showed that 93% and 92% of total data variance was covered by the three first PCs generated from Maximum Sensor Response (MSR), Area Under Curve (AUC) features, respectively. Cluster analysis verified that the pure and impure EVOO samples can be categorized by e-nose properties. PCA-Quadratic Discriminant Analysis (PCA-QDA) classified the EVOOs with an accuracy of 100%. Multiple Linear Regression (MLR) was able to estimate the adulteration percentage with the R2 of 0.8565 and RMSE of 2.7125 on the validation dataset. Moreover, factor analysis using Partial Least Square (PLS) introduced the MQ3 and TGS2620 sensors as the most important e-nose sensors for EVOO adulteration monitoring. Application of Response Surface Methodology (RSM) on RGB, HSV, L*,a*, and b* as color parameters of the EVOO images revealed that the color parameters are at their optimal state in the case up to 0.1% of canola impurity, where the obtained desirability index was 97%. Results of this study demonstrated the high capability of e-nose and computer vision systems for accurate, fast and non-destructive detection of adulteration in EVOO and detection of food adulteration may be more reliable using these artificial senses.&nbsp

Plastic straws are well-known tools to assist human beings in drinking fluid, but most of them have micro-defects including black spot defects, head problems, pressure tube defects, and sealing wrinkles. The manual detection of these defects has drawbacks such as low efficiency, a high false detection rate, and excessive labor. This paper proposed machine vision-based detection with self-adaption and high-accuracy characteristics. A serial synthesis of algorithms including homomorphic filtering, Nobuyuki Otsu, and morphological opening operations is proposed to obtain plastic straws with binary images with good performance, and it was further found that the convolutional neural network can be designed to realize the real-time recognition of black spot defects, where the corner detection algorithm demonstrates the linear fitting of the edge point of the straw with the effective detection of sealing wrinkle defects. We also demonstrated that the multi-threshold classification algorithm is used to detect defects effectively for head problems and pressure tube defects. The detection system based on machine vision successfully overcomes shortcomings of manual inspection, which has high inspection efficiency and adaptively detects multiple defects with 96.85% accuracy. This research can effectively help straw companies achieve high-quality automated production and promotes the application of machine vision in plastic straw defects with the aid of machine learning.

The paper is devoted to machine learning methods that focus on texture-type image enhancements, namely the improvement of objects in images. The aim of the study is to develop algorithms for improving images and to determine the accuracy of the considered models for improving a given type of images. Although currently used digital imaging systems usually provide high-quality images, external factors or even system limitations can cause images in many areas of science to be of low quality and resolution. Therefore, threshold values for image processing in a certain field of science are considered. The first step in image processing is image enhancement. The issues of signal image processing remain in the focus of attention of various specialists. Currently, along with the development of information technology, the automatic improvement of images used in any field of science is one of the urgent problems. Images were analyzed as objects: state license plates of cars, faces, sections of the field on satellite images. In this work, we propose to use the models of Super-Resolution Generative Adversarial Network (SRGAN), Extended Super-Resolution Generative Adversarial Networks (ERSGAN). For this, an experiment was conducted, the purpose of which was to retrain the trained ESRGAN model with three different architectures of the convolutional neural network, i.e. VGG19, MobileNet2V, ResNet152V2 to add perceptual loss (by pixels), also add more sharpness to the prediction of the test image, and compare the performance of each retrained model. As a result of the study, the use of convolutional neural networks to improve the image showed high accuracy, that is, on average ESRGAN+MobileNETV2 – 91 %, ESRGAN+VGG19 – 86 %, ESRGAN+ResNet152V2 – 96 %., Робота присвячена методам машинного навчання, орієнтованим на покращення зображень текстурного типу, а саме на поліпшення об'єктів на зображеннях. Метою дослідження є розробка алгоритмів покращення зображень та визначення точності розглянутих моделей для покращення даного типу зображень. Незважаючи на те, що використовувані в даний час системи цифрової обробки зображень зазвичай забезпечують високу якість зображень, зовнішні фактори або навіть системні обмеження можуть привести до того, що зображення в багатьох областях науки матимуть низьку якість та роздільну здатність. Тому розглядаються порогові значення для обробки зображень у певній галузі науки. Першим етапом обробки зображень є покращення зображення. Питання обробки зображень сигналів залишаються у центрі уваги різних фахівців. В даний час, поряд з розвитком інформаційних технологій, одним із актуальних завдань є автоматичне покращення зображень, що використовуються в будь-якій галузі науки. Зображення аналізувалися як об'єкти: державні номерні знаки автомобілів, особи, ділянки поля на супутникових знімках. У даній роботі ми пропонуємо використовувати моделі генеративно-змагальної мережі надроздільної здатності (SRGAN), розширених генеративно-змагальних мереж надроздільної здатності (ERSGAN). Для цього був проведений експеримент, метою якого було перенавчання навченої моделі ESRGAN з трьома різними архітектурами згорткової нейронної мережі, тобто VGG19, MobileNet2V, ResNet152V2 для додавання втрат сприйняття (за пікселями), а також додавання різкості до прогнозування тестового зображення та порівняння продуктивності кожної перенавченої моделі. В результаті дослідження використання згорткових нейронних мереж для покращення зображення показало високу точність, тобто в середньому ESRGAN+MobileNetV2 – 91 %, ESRGAN+VGG19 – 86 %, ESRGAN+ResNet152V2 – 96 %

Violating the observation conditions for the investigated objects leads to the formation of diverse typical forms of objects throughout the frame in the series. As a consequence, determining the exact position of the object on the frame becomes difficult. To this end, a method was devised to determine the position of an object using the typical form of its image on a series of frames. This method is based on the formation of a typical form of a digital image of an object based on data from all frames of the series. This makes it possible to take into account the peculiarities of the very formation of the digital image of an object on each frame of the original series. Based on this, a more accurate assessment of the initial approximation of the parameters of all Gaussians of the object's image is performed. Adapting the method specifically for the typical form allows for a more accurate assessment of the positional parameters (coordinates) of the object in comparison with the analytically set profile. The estimation of the position of an object was obtained using the method of least squares. After that, minimization was performed using the Levenberg-Marquardt algorithm. Also, the use of the method makes it possible to improve identification with reference objects and reduce the number of false detections. The study showed a reduction in the standard deviation of frame identification errors by 7–10 times when using a typical digital image shape. The method devised for determining the position of an object using the typical form of its image was tested in practice within the framework of the CoLiTec project. It was implemented in the intraframe processing unit of the Lemur software to automatically detect new objects and track known ones. Owing to the use of Lemur software and the proposed computational method implemented in it, more than 700,000 measurements of various objects under study were successfully processed and identified

Granular soils are used in different areas of civil engineering due to their easy accessibility and low cost as a material. The sieve analysis method, which has been conducted in rock physics measurements for many years, has been practicing determining the particle size distributions for those materials. In this study, a method based on image analysis technique has been developed as an alternative to traditional sieve analysis method for determining the particle size distribution in granular soils. This technique based on image processing to determine particle distributions via the experimental setup that consist of a camera, tripod, light box and mechanical shaking apparatus. In order to assess the reliability of this technique, each sample was subjected to traditional sieve analysis and the results of both analysis methods were compared. In conclusion, it was observed that the results obtained with the image processing technique had a minimum 95.22% closeness with the sieve analysis experiment data.

Generative adversarial networks (GANs) are a hot research topic recently. GANs have been widely studied since 2014, and a large number of algorithms have been proposed. However, there is few comprehensive study explaining the connections among different GANs variants, and how they have evolved. In this paper, we attempt to provide a review on various GANs methods from the perspectives of algorithms, theory, and applications. Firstly, the motivations, mathematical representations, and structure of most GANs algorithms are introduced in details. Furthermore, GANs have been combined with other machine learning algorithms for specific applications, such as semi-supervised learning, transfer learning, and reinforcement learning. This paper compares the commonalities and differences of these GANs methods. Secondly, theoretical issues related to GANs are investigated. Thirdly, typical applications of GANs in image processing and computer vision, natural language processing, music, speech and audio, medical field, and data science are illustrated. Finally, the future open research problems for GANs are pointed out.