Search

Researchers use many classification algorithms with image filers to work on image classification. The goal of this study is to see how well machine learning classifiers perform with different WEKA image filters on image dataset. In this paper we present the comparison of different ML classifiers with image filters using Waikato Environment for Knowledge Analysis (WEKA). WEKA is open source data mining software that comprises of a collection of machine learning algorithms. This paper, we introduces four important data mining techniques K-NN,J48, Naive Bayes and Random forest classifiers for image classification with image filters on the birds image dataset using WEKA tool. In this paper, we compare these classifiers with image filers over different parameters and selected optimal classification algorithm. As per our results Random forest classifier working well on our small image dataset and we got 90% accuracy with image filters. To improve the performance of the classifiers we used ensemble technique. Then after we got 96% accuracy with ensemble image filters on this resultant image dataset. In this paper we concluded that ensemble technique will improve the performance of the classifiers with image filters. [ABSTRACT FROM AUTHOR]

This paper mostly focuses on real-time detection and classification of biomedical waste objects. Traditional methods for detecting the objects like biomedical waste objects in machine learning are substituted by recent technology of object detection methods in deep learning by building Convolutional Neural Network (CNN) which is nothing but the components of deep learning. This paper proposes a deep learning approach for object detection of biomedical waste material by evaluating the TensorFlow object detection API, which can used for real time application, this biomedical object detection API further we can used for automate the biomedical waste segregation system, this proposed system has been trained using more than 500 images for each type of biomedical waste object. Tensorflow is useful and easy to object detection that can form powerful image recognition software, in this proposed work we are using Faster R-CNN algorithm for developing this object detection process. [ABSTRACT FROM AUTHOR]

The task of recognizing and categorizing vehicles in videos and images as objects poses a considerable challenge in terms of appearance-based representation. However, it holds great importance in the practical implementation of Intelligent Transportation Systems (ITS), particularly in real-time applications. The fast advancement of deep learning has resulted in an increasing need within the computer vision field for the development of efficient, robust, and outstanding services across diverse domains. This paper provides an extensive analysis of various methodologies for vehicle detection and classification, along with their utilization in real-time targets, estimating the density of traffic, and related domains through the implementation of Deep Learning techniques. The major findings of our survey highlight crucial insights obtained from an extensive analysis of existing literature, shedding light on the current state-of-the-art techniques. Through a comprehensive review of deep learning methodologies, performance metrics, benchmark datasets, and a comprehensive exploration of the challenges encountered, our survey offers valuable contributions to the field. By synthesizing and presenting the collective knowledge in this domain, our paper serves as a key resource for researchers and practitioners alike, providing a holistic understanding of the advancements and challenges in vehicle identification and categorization within deep learning architectures. [ABSTRACT FROM AUTHOR]

The weed plants of Urtica Dioica (UD) are considered as widely witnessed, injurious and vital weed witnessed in wide areas of grass. It needs rigid techniques of handling (removal or herbicide). These harmful plants have multiple effects on dairy and food production and quality degradation, its removal is a necessary procedure. Although manual techniques of removal are precise, they are extremely costly in the form of time and effort. On the other side, herbicides treatments cause negative impacts on environmental pollution and crops quality. Smart farming and weed processing depend on computer vision techniques to (automatically & efficiently) detect the weeds in order to treat them effectively. Traditional techniques of Machine learning (ML) have significant challenges in weed detection due to the lack of features in real-field imaging. The new fashion of neural networks (Deep learning (DL)) recorded higher accuracy regarding self-learning features, which enhances accurate weed detection. Some recent publications adopted a single DL method, achieving high performance in detecting well-separated weeds and controlled-illumination images, yet they recorded low performance for overlapping uncontrolled illumination or occluded leaves challenges. In this paper, an ensemble of DL networks is used, where each network is robust against specific challenges in UD detection. The proposed ensemble scheme adopts Convolutional Neural Networks (CNN) networks for feature extraction and image classification. The proposed framework includes three major stages such as data preparation, preprocessing, and classification phases. The 3-extractor ensemble scheme is adopted as the cornerstone of the classification stage. The adopted data in this paper is real-field captured images with (overlap, occlusion, and illumination) challenges. The three extractors are: robust Inception-v3 network against illumination issues, robust Visual Graphics Group-16 network (VGG-16) against overlapping issues and robust Residential Energy Services Network-50 (ResNet-50) against occlusion issues. Combining these networks to construct one ensemble scheme enhances the discrimination performance of real-field weed detection. The ensemble- experiments were tested on two benchmarking UD datasets, where both of them contain captured images with real-field challenges. The two datasets consist of (900 &677) samples, respectively, which are separately tested in the ensemble scheme. Set of standard evaluation measures are used, which showed considerable results like accuracy (96.89%), recall (98.08%), precision (94.45%), and F1-score (95.78%). This paper recorded superior performance in their accuracy and significance regarding the State of Art published literature. [ABSTRACT FROM AUTHOR]

State-of-the-art solutions in the areas of language modelling, generating text, speech recognition, generating image descriptions or video tagging, have been using Recurrent Neural Networks as the foundation for their approaches. Understanding the underlying concepts is therefore of tremendous importance if we want to investigate frontier methods and develop new approaches for building smart digital solutions. This paper proposes a smart city control system architecture based on deep convolutional neural networks. The control system has a multilayer architecture that combines loosely coupled intelligent components. As the main layer, a solution based on deep learning technology is applied, which allows solving several tasks simultaneously: segmentation, detection and classification of images received from surveillance cameras of the smart city system. The data obtained at the output of this layer is used for further analysis and decision-making in the smart city system. The proposed architecture has a high degree of modularity and allows the replacement of individual elements in a loosely coupled architecture. In this paper, deep learning and computer vision technologies are also considered, on the basis of which the image processing layer from video cameras is implemented. A masked recurrent neural network is used for this task. [ABSTRACT FROM AUTHOR]

Identifying bird songs is essential in monitoring birds and their behaviour without physically interacting with them and disturbing their natural habitat. The field of audio classification is a recently emerging area of research that is mostly interpreted using supervised machine learning techniques. In recent studies, image classification methods and algorithms have shown great promise in this field. This paper analyses three different deep-learning models which are ANN, CNN and RNN-LSTM and compares these models comprehensively with the above-mentioned models. Metrics such as precision, recall and F1-score have been taken into consideration to decide which model among the three models is best for audio classification. The audio files are converted into Mel-Spectrogram to obtain MFCCs and are fed into the models. It is observed that RNN-LSTM provides the highest accuracy followed by CNN and ANN respectively. This comprehensive study of existing models will help the researchers to choose the perfect model for audio classification for further research. [ABSTRACT FROM AUTHOR]

Automatic systems for monitoring food intake have been widely used. Both diet management and controlling chronic diseases such as diabetes require nutrition management for daily food monitoring. Recently, large-scale and benchmark datasets are needed to improve food analysis systems towards global and diverse recognition. However, the definition of benchmark dataset demands almost all the food types, and to the best of this paper knowledge, no Moroccan food dataset has been proposed for food image recognition systems despite the fact that Moroccan cuisine has become more popular and consumed in other countries. In this paper, we introduce MFOOD-32: a food recognition dataset containing, as a start, 32 food categories consumed by Moroccan people, and a total of 6400 images collected from the Web and certain existing datasets. The dataset is evaluated using three pre-trained CNN models: MobileNet, DenseNet and EfficientNet. The evaluation performance have shown promising results achieving a classification accuracy up to 98%. [ABSTRACT FROM AUTHOR]

Sentiment, emotions, feelings, showing or giving judgments by face gestures, etc. can be communicated by text, speech or images. Analyzing image Polarity via images is now a burgeoning research field. An image can easily be interpreted by any human being and acts as a bridge to relate an image to the human's thoughts. The paper proposed a model for analyzing the sentiments or emotions of an image. We have used a Data-Mining Orange Tool for executing the experiment for classification and sentiment analysis of an image. The main challenge is to acquire the polarity of the unlabeled data using optimal machine learning techniques. To overcome this challenge for getting the polarity and sentiment of an unlabeled data the proposed model used various noteworthy machine learning models like Neural Network, Naïve Bayes and random forest for classifying and analyzing the image sentiment and also provide a comparison based on certain parameters like Area Under Curve, Precision, Recall, classification Accuracy, etc. The tool provides a complete package for performing image analysis using the widgets available with orange tool and a data visualization tool to get a visible representation of the output. The proposed model shows that Random Forest performance the best from the three compared algorithms with classification Accuracy of 0.747, recall & precision as 0.747 & 0.752 and the accuracy based on confusion matrix is 74.72%. [ABSTRACT FROM AUTHOR]

In recent years, AI deep learning technology has been recognized as a potential solution in many fields and applications. It has the ability to solve practical problems, especially the ability to identify and classify images. In addition to ordinary daily image recognition, it also includes applications in medical images. On the other hand, music therapy has been commonly used for children with developmental delays to stimulate children's visual, auditory and tactile sensory abilities and enhance children's developmental skills in different aspects. However, there was a lack of clear quantitative evidence of response time in previous music therapy treatments, mainly as there were no extra human resources to measure the response time accurately during the treatment process. In view of this, this paper aims to identify the time point when children turn their heads after being stimulated by the sound and calculate the reaction time after the sound stimulation for children with developmental delays in the process of music therapy, by using AI deep learning technology. In the experimental analysis, the reaction time is measured for children. The experimental results show that the average error of the reaction time measurement method proposed in this paper is less than 1 %, achieving the desired goal of accurately measuring the reaction time and analyzing the time error calculation. The method reduces the average error by about 6.5 times. [ABSTRACT FROM AUTHOR]

Image recognition plays a very important role in the application of computer vision, but there is a problem that object tracking is not ideal. The previous recognition method could not solve the problem of object tracking in the field of computer vision, and the recognition was inaccurate. Therefore, this paper proposes an artificial intelligence technology for object tracking image recognition and analysis. they used to process the image information, and the indicators are divided according to the requirements of artificial intelligence processing to reduce them Interference factors in image recognition. Then, the theory of human intelligence processes the image artificial intelligence, forms an artificial intelligence processing scheme, and recognizes the results of the image Conduct a comprehensive analysis. The revised results show that the calculation process in this paper is relatively good, and it is suitable for practical theoretical analysis, which has high significance and function, and can be used for later comprehensive judgment. [ABSTRACT FROM AUTHOR]

With an increase in the online picture database, the need for gender and age classification has become an active area of research. It helps in detection and investigation purposes. With the emergence of computer vision, deep learning (DL) is being applied in research due to its wide range of real-world applications in image classification. This paper investigates the performance of two DL models, the Convolution Neural Network (CNN) and the Residual Neural Network (ResNet18) model, for analyzing their efficacy in predicting gender and age classification tasks using facial images. The Adience dataset, a widely used benchmark dataset for these tasks, is used to train and test the respective models. This study uses a 3-fold empirical model evaluation on CNN and ResNet18 model, and results were found on three main assessment criteria-age accuracy, 1-off accuracy, and gender detecting accuracy. The ResNet18 model slightly outperformed the CNN model in both functions by showing a 94% exact match on age classification and 92% on gender prediction. This paper highlights the effectiveness of DL models in gender and age classification tasks using facial images and the importance of the model architecture and design in achieving high accuracy and robustness. [ABSTRACT FROM AUTHOR]

The paper considers math models and algorithms for modeling the propagation of an acoustic signal of side scan sonar. A mathematical model is presented for seabed generating in order to simulate the antennas signals and hydroacoustic echograms. Mathematical models for calculating receiving antenna signals are described. The advantages of the approach include the relatively low computational complexity of applied algorithms (in comparison with the solution of the wave equation with boundary conditions), as well as the possibility of parallelizing computational processes on several threads and computers. The presented results are applicable for sonar signal algorithms debugging, object detection methods developing and other tasks related to hydroacoustic images simulation. Further work is seen in the implementation of described approach as part of a modeling complex for simulating hydroacoustic signals in real time. [ABSTRACT FROM AUTHOR]

Texture identification and classification is an ever existing problem in the image processing scenario over the past and the present decades. A statistical design of experiments approach for texture identification and representation has been successfully attempted for a larger grid size of an image. There were four groups of feature vectors (FV) suggested for texturedescriptors. The first FV consisting of off - the main diagonal elements in the interaction groups which are considered to be the feature descriptors in a larger grid under consideration while using the design of experiments based approach. In this paper, the proposed texture descriptor has been demonstrated to work efficiently for the classification of texture images in a supervised manner. The global descriptor (FV) is used for representing the texture content in the image to be classified. The main idea of classification demonstrated in this proposal, between the texture images and the reference image is that the texture from the same group will have zero difference or near zero in the distance measure between the corresponding global descriptors(FV). The integrated absolute difference (IAD) has been used for identifying the closeness between regions and the reference texture through respective FVs. The proposal has been successfully applied for a number of texture images collected from various standard textural albums and the results are reported with the available results. The scheme is being extended for identifying the texture classification with other FVs and in an unsupervised manner also. [ABSTRACT FROM AUTHOR]

In recent years, with the advancement of deep learning, the combination of computer vision and natural language processing has gotten a lot of attention. Image captioning is an example of this field, in which a computer learns to interpret the visual information of an image by using one or more phrases. The capacity to analyze the state, properties and relationships among these objects is required not only for the meaningful description generation process of high-level image semantics but also for the ability to recognize the object and scene. Despite the fact that image captioning is a complex and demanding process, several researchers have made great progress. In this paper, we propose a real-time system using an encoder-decoder architecture to generate captions for the scene. [ABSTRACT FROM AUTHOR]

In the task of remote sensing, the Hyperspectral image (HSI) classification to analyze land cover is an established research topic. However, the nature of remote sensing data still poses several challenges including, the curse of dimensionality, the negligible number of samples during training or the presence of unbalanced data which makes learning difficult. Having a training set of pixels with the label of the assigned class, the operation that is performed in the classification of hyperspectral images is to assign a class label to each pixel in the test set based on the knowledge acquired with the training set. This paper discusses a new approach in the supervised classification of HS images considering the statistical tool of Copulas. Comparison with well-established techniques shows the good behaviour of this technique. [ABSTRACT FROM AUTHOR]

This paper presents the first results of the Bayesian model in the problem of image classiffication using feature extraction. The proposed approach is similar to the well-known Markov processes and is based on conditional probabilities determined from the given data using the regression process with Kullback-Leibler divergence. In the contribution, we shortly review the complete process consisting of the Scale Invariant Feature Transform technique for the detection of key features, Bag of Visual Words, and classiffication. We compare the new methodology with the standard approach based on Support Vector Machine on a non-trivial benchmark containing real photographs of cats and dogs. [ABSTRACT FROM AUTHOR]

Monkeypox disease was declared a Public Health Emergency of International Concern (PHEIC) by the World Health Organization (WHO) on July 23rd, 2022. The purpose of this designation is to encourage the global community to take coordinated measures to contain the spread of the disease and protect communities. The recovery period for Monkeypox disease typically ranges from 2 to 4 weeks, and in some cases, the disease can resolve spontaneously with or without treatment. However, severe cases have been reported, which can lead to fatalities. The fatality ratio for Monkeypox disease is estimated between 3% to 6% at present. It is essential to take precautionary measures to prevent the spread of this disease and seek medical attention if any symptoms are observed. Monkeypox is not highly contagious, and it generally spreads through close contact with an infected person. However, in May 2022, a significant number of cases of Monkeypox disease were identified, indicating that the disease was spreading rapidly. In light of this situation, it is crucial to track and isolate suspected cases promptly to prevent further transmission. Rapid and effective contact tracing, along with timely isolation of cases, is essential to contain the spread of the disease and prevent it from becoming an epidemic. Identifying, controlling, and taking measures to control the spread of Monkeypox disease is of utmost importance. To aid in this effort, this paper aims to create a large-scale dataset of Monkeypox disease images using Generative Adversarial Networks (GANs). GANs have proven to be highly effective in generating high-quality images and can aid in creating a more comprehensive and diverse dataset. The generated images will be used to train different deep learning models for image classification to diagnose Monkeypox disease. The deep learning models will be compared, and their performance will be evaluated using images of infected human skin with Monkeypox disease and healthy skin images. The results obtained from this paper will improve the accuracy of previous works, aiding in the early diagnosis and treatment of Monkeypox disease. [ABSTRACT FROM AUTHOR]

Convolutional Neural Networks (CNN's) are known to perform well on computer vision tasks such as image classification, image segmentation, and object detection. However, one major drawback of CNN's is the huge amount of computing and memory resources needed to train them. In this paper, we propose an architectural unit which we call Upsampling-Based Wavelet Residual Block (UBWRB), that utilizes the 2D discrete wavelet transform coupled with upsampling operators and a residual connection to extract features from image data while having relatively fewer trainable parameters as compared to traditional convolutional layers. The discrete wavelet transform is a family of transforms that find extensive applications in signal processing and time-frequency analysis. For this paper, we use the filter-bank implementation of the discrete wavelet transform, allowing it to act in a similar fashion to a convolutional layer with fixed kernel weights. We demonstrate the performance and parameter-efficiency of CNN's with UBWRB's in the task of image classification by training them on the MNIST, Fashion-MNIST, and CIFAR-10 datasets. Our best-performing models achieve a test accuracy of 99.34% on the MNIST dataset while having less than 120,000 trainable parameters, and 92.90% and 84.27% on the Fashion-MNIST and CIFAR-10 datasets respectively, with both having less than 180,000 trainable parameters. [ABSTRACT FROM AUTHOR]

In today's world, the number of vehicles plying on the road is increasing day by day. Due to huge traffic, emergency vehicles like ambulances are not able to reach their destinations in time, resulting in the loss of human lives. Though people are aware of giving way to ambulances, it may not be always possible. Also, at red traffic signal, ambulances must stop. Else, severe accident may happen due to other vehicles coming from other directions. Hence, there is a need for a solution, that allows the movement of an ambulances on priority, while all other vehicles are halted. This paper has proposed a real-time system based on image and audio modality. It is developed using Nvidia's Jetson nano bord, web camera, and microphone. Combined audio and image classification technique helps in regulating the traffic by detecting the siren sound and then by identifying actual ambulance. [ABSTRACT FROM AUTHOR]

The objective of this paper is to build an Object Recognition model using 'TINY YOLO' and FPGA. Object Recognition is a CV (i. e., Computer Vision) technique for identifying and detecting objects in images or videos. Object recognition is a key output of DL (i. e., Deep Learning) and ML (i. e., Machine Learning). When we humans look at an image or video, we can readily spot people, object, scene and visual details. The aim of this is to teach a computer to do what comes innately to a human being. Object recognition is done with Image Classification, Object Localization and Object Detection that combines these two tasks and localizes and classifies an object in an image. Object Recognition has become a pivotal technology for driverless cars, security purpose, traffic surveillance etc., It is also used in a various application such as disease detection in bio imaging, industrial examination, and robotic vision. Hence, in a nutshell, 'Object Recognition is the staple of automation industry'. [ABSTRACT FROM AUTHOR]

Radio or News Paper?... And the winner is Radio. The same way many people prefer radio to newspaper which means they prefer listening to reading. If not listening one must know to read and with radio we can simply listen. If the text is in other language, there must be a translator. If we analyze a situation where two parties are desperate to communicate but they does not have a common communication medium, In that case they need a translator in between them for communication. Today's world is moving towards socialization and there can be situations where people from one place connect to others. Text messages, Emails, Phone calls etc. has been a great deal for communication. In these cases, most of the applications have done speech to text conversion and vice versa for effective communication. The "Go-Voice" application was developed in order to read out the text from the text files and any text from images by Text Recognition. [ABSTRACT FROM AUTHOR]

At present, the verification method of steel tape has characteristics of high labor intensity, low working efficiency and low measuring accuracy. To solve these problems, an automatic steel tape verification system is built in this paper. This system takes the stainless steel as the calibration platform, uses the CCD camera and image processing technology to achieve the indication errors between lines. It realizes the automatic verification through precise motion control technology and special measuring software. This device completes the verification by trolley motion control and CCD image acquisition, and carries on the image recognition to analyze errors and evaluate results. Experiments show this device has comprehensive functions, high level of automation, practical application value and broad market prospects. [ABSTRACT FROM AUTHOR]

The main objective of this paper is to improve the accuracy for automatic classification of meal waste from innovative trash data with the help of image processing. There are 2572 images for the classification of meal waste were used for this paper. The images are labeled as "Cardboard", "Plastic", "Paper", "Metal", "Glass", "Trash" and there are 20 number of images have been used for RF classifier taken as first set of machine learning algorithm and is compared with SVM algorithm taken as second set of machine learning algorithm With a g-power value of 80%, the revolutionary garbage data images, a threshold of 0.05%, a confidence interval of 95%, and a standard deviation, these photographs were gathered from various web sources. When compared to the SVM method, which had an accuracy of 61.45%, the proposed system's accuracy was enhanced to 84.81%, with a significant value of 0.001 (p 0.05) with a 95% confidence interval. This study found the meal waste from trash using the RF is significantly better than SVM algorithm. [ABSTRACT FROM AUTHOR]

Image denoising is one of the most fundamental difficulties in image processing and computer vision, with the objective of recovering the original image by suppressing noise from a noise-contaminated version of the image. Image noise can be created by a variety of internal (sensor) and external (environment) factors that are difficult to prevent in real-world settings. As a result, image denoising is useful in a variety of applications, including image restoration, visual tracking, image registration, image segmentation, and image classification, where recovering the original image content is essential for good performance. While various methods for image denoising have been proposed, the problem of image noise suppression remains unsolved. The purpose of this paper is to present a comprehensive study on vector median filters for removing the adverse impact of impulse noise from color images. Color images are vectored value signals that are nonstationary in nature. As a result, nonlinear filters such as vector median filters outperform linear filters. In the literature, a variety of nonlinear filters have been proposed. [ABSTRACT FROM AUTHOR]

Remote sensing multi-band hyperspectral image classification is a challenging task and is of current interest too. Proper extraction of spectral features of the data results in higher classification accuracy. This paper proposes a hybrid approach to obtain maximum classification accuracy of hyperspectral data. First and foremost, the Fast Fuzzy Cluster Means (FastFCM) algorithm is presented for overcoming the limitation of high computational time of conventional FCM for high-dimensional data. The segmented outcome as spectral information is then combined with the spatial information obtained by the Extended Multiattribute Profiles (EMAPs) to form a composite kernel (filter). In this experiment, the Support Vector Machine (SVM) is used as a classifier to train the data. The classification outcome in terms of three parameters Overall Accuracy (OA), Average Accuracy (AA), and Kappa Index (KI) of the proposed approach (named as (FastFCM-SVM)) has been compared with some current state-of-art methods to investigate its better stability and classification performance. [ABSTRACT FROM AUTHOR]

The technological advancements made in the Information Communication and Technologies (ICT) system has impressed the smart classroom management applications. The communication networks play a vital role in connecting the different entities of a smart classroom system. Detecting (or) predicting cheating behavior during online examination at an earlier stage is a never-ending challenging task. This paper is an extensive survey on cheating behavior detection in smart classroom management using object recognition models. The conducted survey covers the significance of cheating behavior detection by means of object recognition approach and the image processing techniques. The survey analysis states that the research challenges such as representation of objects, constancy in scenes, classification of features, searching complexity and the presence of objects and their occlusion needs to be addressed while developing an intelligent cheating behavior detection process. [ABSTRACT FROM AUTHOR]

This paper proposed an improved image registration method combining Hu-based invariant moment contour information and feature points detection, aiming to solve the problems in traditional image stitching algorithm, such as time-consuming feature points extraction process, redundant invalid information overload and inefficiency. First, use the neighborhood of pixels to extract the contour information, employing the Hu invariant moment as similarity measure to extract SIFT feature points in those similar regions. Then replace the Euclidean distance with Hellinger kernel function to improve the initial matching efficiency and get less mismatching points, further, estimate affine transformation matrix between the images. Finally, local color mapping method is adopted to solve uneven exposure, using the improved multiresolution fusion algorithm to fuse the mosaic images and realize seamless stitching. Experimental results confirm high accuracy and efficiency of method proposed in this paper. [ABSTRACT FROM AUTHOR]

This study looks into a number of recently developed pre-trained models and assesses how well they perform using the ImageNet Dataset. In the disciplines of image processing, computer vision, and machine learning, image Classification is a well-known issue. In this study, we studied six different models—three transformer-based models and three convolutional neural network-based models. The findings of this research offer insightful information on the perks and drawbacks of using pre-trained models for image classification tasks. The results indicate that while transformer-based models exhibit promising outcomes, convolutional neural network-based models continue to perform better overall and with regard to accuracy. Additionally, the pre-trained models' performance may be greatly enhanced by fine-tuning them on a smaller dataset. This paper provides an assessment of recently developed pre-trained models and their performance on the ImageNet dataset, which adds to the current research in the fields of image processing, computer vision, and machine learning. The knowledge gathered from this work can guide the creation of models for image classification tasks that are more effective and accurate. [ABSTRACT FROM AUTHOR]

Agriculture's production is heavily impacted by the state of the economy. Plant disease is increasingly common in agricultural settings and is now easier to detect as a result of the previously described issue. It is receiving more attention due to the monitoring of crops in a wide range of settings. Farmer transitions between different disease control strategies and encounters significant challenges. We are able to identify or detect leaf illnesses, which is how surveillance and monitoring experts usually find them. If the appropriate control measures aren't followed, the plants suffer a major consequence, which symbolizes how the quality of the plants' production will be impacted. The adoption of a mechanical approach or method for identifying diseases is effective and beneficial since it lessens the burdensome work of surveillance in the vast cultivation. The early stages of plant illnesses allow us to identify the signs even before they manifest on the plant's leaves. The early stage denotes the starting period which is one or two days after affecting the disease. The paper's primary objective is to use ResNet50 for early detection of plant disease. ResNet50 has been successfully applied to plant disease image classification tasks and has shown promising results. ResNet architecture, with its deep layers and residual connections, has the ability to capture intricate patterns and features, which is useful for identifying the differences in plant images. The results proved that the ResNet 50 has better accuracy, precision and recall than Support vector machines (SVM), artificial neural networks (ANN), Convolutional neural networks (CNN) for plant disease prediction. [ABSTRACT FROM AUTHOR]

Glomeruli are tiny clusters of blood vessels found in the kidneys which filter bodily waste. However, various diseases such as diabetes and obesity are risk factors for glomerulosclerosis, the scarring of glomeruli. This impedes the filtration process and leads to more advanced renal diseases and even kidney failure. Diagnosis currently involves a process of biopsies and classification of numerous glomeruli whole slide images. The classification of these images are time-consuming and prone to human error. Thus, there is an opportunity to utilize image processing and artificial intelligence to aid in the automation of the process, thereby aiding nephrologists to shorten diagnostic time. To this end, current literature explores the use of image classification using neural networks in the identification of glomerulosclerosis. However, existing studies make use of single convolutional neural network architectures. To improve the results, ensemble learning can be used. Ensemble learning is a method that aggregates multiple CNN architectures to reduce bias and to improve overall accuracy. Previous works demonstrated that using ensemble learning can increase accuracy in the classification of skin lesions, showing promise to its application in other fields such as glomerulosclerosis. As such, this paper uses an ensemble of CNNs to classify glomerulosclerosis in a dataset of 2,340 whole slide images of glomeruli as either "normal" or "sclerosed." Results are further optimized through the testing of different weight distributions using the weighted averaging ensemble aggregation method. Ultimately, we show that the use of ensemble learning in the binary classification of glomeruli can increase the accuracy of architectures compared to single neural networks. [ABSTRACT FROM AUTHOR]

This paper discusses the application of the Content-Based Image Retrieval (CBIR) system in image classification for various types of batik in the region. The aim is to detect and recognize certain types of batik with high accuracy, as well as develop a digital image-based security system for various types of batik with various motifs, patterns, and colors. In this research, a typical South Sulawesi batik type, namely Lontara motif batik, was used as a sample. The methods used to perform feature extraction and classification are the Gray Level Co-Occurrence Matrix (GLCM) and Multilayer Perceptron (MLP). The use of GLCM is equipped with the Angular Second Moment (ASM) feature with contrast, correlation, energy, and homogeneity parameters, so that the image data extraction results will be more accurate. Apart from that, the combination with the Multi-Layer Perceptron (MLP) algorithm will make it easier for the system to carry out classification because MLP uses three training functions, namely Levenberg-Marquardt (LM), Variable Learning Rate Backpropagation (GDX), and Scaled Conjugate Gradient (SCG), which produce high accuracy values at smaller epochs. The best accuracy value obtained in the test results was 88.89%. The combination of the GLCM-MLP method can be used as a solution to accurately detect various batik motifs. [ABSTRACT FROM AUTHOR]

In our research, we are interested in the automatized detection of corrosion on images of steel structures. The paper presents the influence of the roughness of image and the value of widely used measures, namely properties of grey-level co-occurrence matrix (GLCM) and grey level run length matrix (GLRLM). The results show the capability of these measures for the description of roughness in the image, which is typically considered significant evidence of corrosion. We suggest that image classification based on reviewed measures can be further utilized for standard classification methods. [ABSTRACT FROM AUTHOR]

The degree of Guyabano fruit is decided commercially by human assessment of color and texture to confirmthat the fruits are of good quality. Manual grading of Guyabano fruit using human vision is both time-consuming andharmful. Guyabano is graded for ripeness using vision and touch to assess color and texture. Farmers and consumers who are ready to harvest and eat Guyabano fruit but are having trouble determining maturity will be able to do utilizingCNN techniques. In today's world, the most widely utilized class of image identification and classification models is the Convolutional Neural Network (CNN). The application of strategies for grading the quality of Guyabano fruit is discussed in this paper. From hand grading, an automatic gradation is done to ensure that the fruits are of the greatest possible quality. Based on the results, the study looked at various CNN designs to determine the best prediction classifier. The goal of this study is to develop a non-destructive way for the significant maturity of Guyabano fruits. We employed a MobileNetV2 network that had been pre-trained by the ImageNet dataset as a base network and then substitute the top layer of the base network with a normal convolution layer. The test results were reviewed using generated accuracy and loss evaluation to establish the model's overall effectiveness. The average grading accuracy for Guyabano fruit is 99%. [ABSTRACT FROM AUTHOR]

Recently, there have been several automations percolating around the Banking Sector. Banks use several automation tools to fasten their banking jobs which makes them cumbersome if not used. Vehicle Insurance Policy is a huge banking section which involves direct involvement of bank employees for vehicle damage verification and customer authentication. Insurance policies do not cover all sorts of vehicle damage, hence the type of damage incurred to the vehicle has to be verified. This process can be automated by Image Recognition using convolutional neural networks (CNNs) which has the capability to detect the type of vehicle damage (bumper dent, tail lamp dent, door scratchetc.,). The required input is an Image of the damaged part of the vehicle. In this paper, several hybrid CNN models are established to show the changes evolved in the accuracy metrics. Experimental results show that transfer learning technique performs a great deep feature learning on the present limited dataset. [ABSTRACT FROM AUTHOR]

A key component of tumour diagnosis, treatment planning, and creating an outcome for assessing brain tumours is highly accurate and dependable segmentation. Magnetic resonance imaging (MRI) is a tool used in medicine to help find very small abnormal growths in any human body part. Deep neural networks (NNs) and machine learning methods are efficient at segmenting 2D images, but it is challenging for NNs to distinguish significant organs from 3D medical MR images. The segmentation for tumour detection includes a number of processing techniques thatfall into the categories of pre-processing, segmentation, optimization, and feature extraction. The study focuses mainlyon 3D-based Convolution neural networks (CNN), ANN (Artificial Neural Networks), SVM, and Multi-class Support vector machines for deeper segmentation. In order to lessen the computational burden of processing 3D medical scans,this survey paper will examine recent developments in image segmentation and image classification based on effective and efficient processing of tumor-infected human brain MRI adjacent image patches that can pass through the network. The focus will be on gliomas. More discriminative 3D NNs and computational machine learning are used to process the input images at different scales simultaneously. This article also suggests the state of the art in tumor-based imageprocessing using deep learning models for tumour segmentation and detection. [ABSTRACT FROM AUTHOR]

Implementation of Machine Learning/Deep Learning techniques in Image Processing have paved way to advanced object detection, classification and tracking in a real time environment. Recent advancements are enabling these techniques to be used in the field of real time Infra Red (IR) object detection. IR image processing is being used extensively in the field of military, industrial and medical applications. This paper emphasizes upon the various Deep learning techniques which can be used for automatic target detection/recognition in infra red maritime imagery of sea borne military assets. Our review begins with the introduction to Object Detection, IR imaging, techniques available and various architectures for feature extraction and object detection. Then, we also provide a comparison of performance/detection capabilities on widely used PASCAL VOC 2007 and 2012 datasets. [ABSTRACT FROM AUTHOR]

The COVID-19 Pandemic has brought changes in various aspects, particularly in education sector. Consequently, this situation has encouraged society to learn as well as use Artificial Intelligence (AI) technology much further and lead to an era of technologies that are dynamic, adaptive, productive, and at the same time making the world more digitally connected. AI facilitates the existing technology to be more applicable in daily lives by using image recognition which is already being developed. As the field of AI develops, so will the potential to protect the environment. From the land and air to both drinking and ocean water, AI is shaping up to be the key that government, organizations, and individuals can tap to work toward a cleaner planet. Furthermore, Telkom University as one of universities that supports Green Campus concept, will be the case study to be reviewed. This paper focuses on the role of AI in achieving the sustainable environment of Green Campus through Image Recognition to detect another part that is not perceived to be healthy/green in terms of the environment. To sum up, it is proven that AI offers beneficial uses in diverse purposes, notably in underpinning environmental protection effort. [ABSTRACT FROM AUTHOR]

China's urban railway network already covers 80 percent of the country. The support of a continuous line is inextricably linked to the rapid expansion of railways. The seamless line has a complex structure, a long working period, and is susceptible to environmental and other factors. This paper mainly studies the design and implementation of seamless line displacement data detection by using image recognition technology and machine learning technology. This topic trains and learns the digital content in the seamless line displacement scale through the machine learning SVM classification algorithm, and then identifies the scale mark and other data by image analysis to calculate the final displacement. In this way, the accuracy of the identified displacement data is very high. This study detects the displacement of the seamless line and gives early warning when the displacement exceeds the safety range. After receiving the early warning information, the management personnel can handle it in time to avoid danger. Some railway portions in Harbin have been subjected to the study. The research can assist railway management departments in detecting seamless line facilities and improving maintenance efficiency. At the same time, it can provide timely warnings of potentially risky information, reducing the likelihood of accidents. [ABSTRACT FROM AUTHOR]

This paper studies one of the important structural methods in the classification of texture images, which is scale-invariant feature transform (SIFT), where the Brodatz database was used as sample images to test the proposed algorithm. The k-mean classifier and the frequency density function are used to classify the images based on classifying the obtained feature pointsﻭ where an analysis of the proposed classification algorithm is presented. The results confirmed the possibility of the proposed algorithm in classifying histological images. [ABSTRACT FROM AUTHOR]

Multiresolution analysis has wide-ranging applications in signal processing such as fault detection and isolation in sensors. Wavelet transforms can be used for feature extraction and compressing the signals without losing their characteristics. Coefficients of wavelet transform of a signal can be represented in the form of scalogram images. Neural networks helped researchers in replacing human interventions which can also be trained for specific tasks. Researchers have also demonstrated the high level of accuracy and usefulness of Deep Neural Networks(DNNs) in image classification. DNNs pre-trained with millions of images can be used for feature extraction or can be modified and fine-tuned for specific tasks. Scalograms contain the fault signatures of the transformed signal, which can then be classified with the help of a Deep Neural Network(DNN). Though DNNs are efficient in image classification it requires a lot of computing resources for training. To overcome this disadvantage we used transfer learning i.e. modifying and fine-tuning a pre-trained DNN. In this paper, we proposed a model for fault detection and isolation in sensors using transfer learning of DNNs and scalograms. A publicly available temperature sensor dataset was used for fault injection and classification purposes. The proposed model successfully classified sensor stuck faults with 95% accuracy by using scalograms as input to the fine-tuned DNN. [ABSTRACT FROM AUTHOR]

Multi-class image classification is one of the tasks of computer vision in which convolutional neural networks are widely used. In this paper, we create a novel ensemble learner by stacking InceptionV3, Xception, MobileNetV2, and ResNet50 convolutional neural network in an attempt to outperform the best performance of each individual models. In an ensemble learning algorithm, by combining multiple different models with different architectures, an error of prediction from a single model can be corrected by other models, thus reducing the likelihood of selecting a wrong prediction. The dataset that we used to do benchmarking is the Intel Image Classification dataset which consists of 6 classes that displays images of fully colored natural scenes. The training set consists of 14034 images, while the test set consists of 3000 images, all having 150x150 as their default size. The training process on all models involves reducing their learning rate when the performance plateaus, and early stopping if there aren't any significant improvements from an epoch to the next. The result is that each individual models have an average of 83% accuracy, yet our ensemble model managed to reach 88% on the test set. [ABSTRACT FROM AUTHOR]

Covid-19 is a deadly disease that has spread all over the world. It has a larger social influence. There are various ways for diagnosing the virus, including RTPCR, Xray, and others, but chest computed tomography(CT) is one of the most effective. CT images are employed to identify if there any abnormalities and level of abnormalities identification of abnormalities in CT images is highly challenging. This paper focuses on the classification of CT images into normal or abnormal and also to localize the lesions in the CT images. Here lung lesion localization for accurate screening of covid 19 in CT images is proposed by incorporating transfer learning and K-mean clustering. The proposed approach is based on Resnet 50 architecture followed by dense network for classification [ABSTRACT FROM AUTHOR]

In recent years, improvements in natural language processing and computer vision have come together to provide automatic image caption generation. Image captioning is the process of creating a description for an image. Captioning an image needs the recognition of significant items, their properties, and their connections within the image. Additionally, it must create phrases that are syntactically and semantically accurate. Deep learning approaches can address the complexities and difficulties associated with image captions. This paper describes a joint model which is capable of automatically captioning images using EfficientNet-B0 and a transformer with multi-head attention. The model is an aggregation of an EfficientNet & Transformer single encoder and decoder. The encoder utilizes EfficientNet-B0, a convolutional neural network-based algorithm that generates a detailed input image, represented by embedding them into a fixed-length vector. The decoder employs a transformer, and a multi-head attention mechanism to selectively concentrate attention on certain regions of images to predict the sentence. The proposed model was trained using a large dataset Flickr8k to optimize BLEU N-Gram (N=1,2,3,4), METEOR score, and CIDEr, to assess the probability of the target description phrase given in the training images. Our studies show that the proposed model can produce captions for images automatically. [ABSTRACT FROM AUTHOR]

Cervical cancer is one of the most known health problems faced by women around the world. Early detection of cervical cancer may reduce the mortality rate. Pap smear images are new techniques used for screening cervical cancer. This paper excludes the nucleus of pap smear images, and the resultant images are classified into seven classes based on the surrounding region nucleus. Automated features are extracted using three pre-trained convolutional neural networks (CNN). The resultant features are twenty-one. The principal component analysis reduces the dimensionality and selects the most significant features into ten features. These features are fed to two types of machine learning algorithms: support vector machine (SVM) classifier and random forest classifier. The support vector machine classifier achieved the highest accuracy for seven classes, reaching 93.1%. This method will help the physicians in the diagnosis of cervical cancer depending on the tissues, not the nucleus. Furthermore, the result can be enhanced using a huge amount of data. [ABSTRACT FROM AUTHOR]

The paper aimed to detect the brain tumor with MRI images through Novel CNN and predict the accuracy value. Materials and Methods: Proposed work involves two groups for brain tumor detection. Group 1 was the Novel Convolutional Neural Network (CNN) algorithm and Group 2 was Decision Tree (DT) algorithm. The sample size was calculated. It was identified that 10 samples/group and 20 samples were taken totally. The improved CNN algorithm was compared with the DT algorithm for predicting the accuracy. Results and Discussion: Data collection was carried out and the analysis was done in the Google Collab for the execution of results and for measuring the accuracy of given algorithms. The Novel CNN algorithm produces high accuracy with 92% whereas the DT algorithm gives 91% of accuracy. In this work, the improved CNN algorithm was compared with the existing DT Algorithm. Conclusion: The data was collected from various resources for the usage of brain tumor detection. The Novel CNN algorithm obtains high accuracy when analyzed with other existing algorithms. [ABSTRACT FROM AUTHOR]

This paper presents a method for achieving the aesthetic quality of images captured by (continuous shooting mode), To tell the difference between high-quality and low-quality pictures, we used a type of convolutional neural network (AlexNet) for feature extraction and image classification. The experimental results showed that our approach can achieve good results in assessing the aesthetic quality of images. The database adopted in our method is the AVA DP challenge portal. [ABSTRACT FROM AUTHOR]

The theory of pattern recognition is used in computers, in city's video surveillance cameras, when recognizing car license plates of the violators, for unlocking of smartphones, in drones, in security systems, etc. The paper is considering the solution to the problem of using image recognition methods for detecting road's signs. The purpose of the research is to develop a mobile application for the recognizing of road signs. The methodology includes a description of the stages of the video sequence for recognizing of road signs, choosing methods for their localization, classification and data storage in recognition systems. A significant result of the research is the development of a model based on convolutional neural networks for the classification of road signs. The developed model is used in a mobile application applying the methods of the libraries of computer vision. The application can be used by car drivers with installed file on the device and an autofocus camera. Testing of the application confirms its ability to perform the assigned tasks. In conclusion, findings are drawn about the effectiveness of the application. And options are proposed for improving the model for recognizing of road signs. [ABSTRACT FROM AUTHOR]

In this paper we propose a novel face recognition approach based on slow feature analysis (SFA) in contourlet transform domain. This method firstly use contourlet transform to decompose the face image into low frequency and high frequency part, and then takes technological advantages of slow feature analysis for facial feature extraction. We named the new method combining the slow feature analysis and contourlet transform as CT-SFA. The experimental results on international standard face database demonstrate that the new face recognition method is effective and competitive. [ABSTRACT FROM AUTHOR]

The technique of obtaining a wider field-of-view of an image to get high resolution integrated image is normally required for development of panorama of a photographic images or scene from a sequence of part of multiple views. There are various image stitching methods developed recently. For image stitching five basic steps are adopted stitching which are Feature detection and extraction, Image registration, computing homography, image warping and Blending. This paper provides review of some of the existing available image feature detection and extraction techniques and image stitching algorithms by categorizing them into several methods. For each category, the basic concepts are first described and later on the necessary modifications made to the fundamental concepts by different researchers are elaborated. This paper also highlights about the some of the fundamental techniques for the process of photographic image feature detection and extraction methods under various illumination conditions. The Importance of Image stitching is applicable in the various fields such as medical imaging, astrophotography and computer vision. For comparing performance evaluation of the techniques used for image features detection three methods are considered i.e. ORB, SURF, HESSIAN and time required for input images feature detection is measured. Results obtained finally concludes that for daylight condition, ORB algorithm found better due to the fact that less tome is required for more features extracted where as for images under night light condition it shows that SURF detector performs better than ORB/HESSIAN detectors. [ABSTRACT FROM AUTHOR]

Picture captioning has recently become a new difficult challenge that gathered everyone interest, which is capable to define automatically an image's content with properlyformatted text English sentences. it makes a great impact by assisting for visually impaired people better recognition of their circumstances. By taking the images of neighboring environment that make use of these photos to generate captions that are examined visual amplification impaired, so that they can get a better sense of what's going on around them. In this paper to extract features, a mixture of convolutional neural networks of the images & then LSTM was used (Long short-term memorys) to regenerate text from these features. The obtained text is converted into speech so that it can be read out. Our model generates highly descriptive captions that canpotentially greatly improve the lives of visually impaired people. [ABSTRACT FROM AUTHOR]