Search

In modern deep learning, character recognition in images is a very important field of study due to its has many real life applications. The goal of this paper is to create the state-of-the-art character recognition model using a stacking ensemble of convolution neural networks (CNNs).To develop the proposed ensemble model, we evaluated several CNN models. The models were judged on how well they performed on the Chars74k dataset. The dataset contains 74,103 images divided into 62 classes with labels [A-Z], [a-z], and [0-9]. The accuracy distribution based on the dataset's subgroups (uppercase, lowercase, and digit) is shown in results. The proposed ensemble model achieves state-of-the-art performance with a maximum accuracy of 92.31% on complete dataset, 99.22% on Uppercase alphabets, 98.66% on Lowercase alphabets, 99.77% on Digits, 91.97% on Uppercase+Lowercase alphabets. On the complete and partial datasets, a comparison report between the proposed model and other existing approaches is also displayed. A comparative study of the proposed work and the previous methods is also shown in this paper, in order to demonstrate the effectiveness of the proposed work. [ABSTRACT FROM AUTHOR]

The COVID 19 pandemic is highly contagious disease is wreaking havoc on people's health and well-being around the world. Radiological imaging with chest radiography is one among the key screening procedure. This disease contaminates the respiratory system and impacts the alveoli, which are small air sacs in the lungs. Several artificial intelligence (AI)-based method to detect COVID-19 have been introduced. The recognition of disease patients using features and variation in chest radiography images was demonstrated using this model. In proposed paper presents a model, a deep convolutional neural network (CNN) with ResNet50 configuration, that really is freely-available and accessible to the common people for detecting this infection from chest radiography scans. The introduced model is capable of recognizing coronavirus diseases from CT scan images that identifies the real time condition of covid-19 patients. Furthermore, the database is capable of tracking detected patients and maintaining their database for increasing accuracy of the training model. The proposed model gives approximately 97% accuracy in determining the above-mentioned results related to covid-19 disease by employing the combination of adopted-CNN and ResNet50 algorithms. [ABSTRACT FROM AUTHOR]

The very low resolution (VLR) issue arises in many face application systems because of the rising demand for surveillance camera-based applications. On the VLR face image, the face recognition algorithms in use are unable to work satisfactorily. Although face super-resolution (SR) techniques can be used to improve the resolution of the images, these VLR face images do not respond well to the existing learning-based face SR techniques. This work suggests a novel method for learning the link between the high-resolution image space and the VLR image space for face SR to solve this issue. The paper aims to address the problem of face image super-resolution and solutions for resolving the issues related to facial images. The major issues related to face image super-resolution during reconstruction end is a fusion of more than one kind of feature which may cause noise in the image, blind spot generation, low perceptual quality, and checkboard issues. The existing models are designed to improvise the perceptual quality of face super-resolution but still fail to generate better perceptual quality of image due to loss during the reconstruction stage. Therefore, this paper proposes a novel progressive face hallucination super-resolution (FHSR) model with a loss-aware upscaling network layer. The upscaling layer is integrated with correlation filters and used a combined loss function. The model is designed as cascading of upscaling modules that are progressive with dense skip connection layers. Loss functions from each upscaling module are evaluated and cascaded together and fed forwards to the next layer to minimize the reconstruction losses. The architecture upscaled the images with 3 cascading modules which magnify 2x,4x, and 8x. The paper also presented the ablation study that assured that the designed FHSR model is better as compared to baseline models. This architecture overcomes the issue of extraction of more features with minimum losses and constructs the high magnification at the reconstruction end. The comparative study shows enhanced performance over state-of-art models. [ABSTRACT FROM AUTHOR]

The world population is growing very fast and the lifestyle of human beings is changing with time and place. So, there is a need for disease management which includes disease diagnosis, its detection and classification, cure and lastly for future disease prevention. The outermost protective layer of a human body is the skin. Skin not only impacts a person's health but also psychologically impacts one's life. Computer-aided systems are very helpful in skin disease detection and classification and their application is growing rapidly in healthcare. This literature review paper aims to help the researchers to get a synthesized and appropriate information for the same. We have included papers from 2021 to 2023 for the review from the Scopus database. 45 studies are selected for the review of which 32 studies use deep learning techniques, 11 use machine learning techniques and 2 studies use a hybrid approach. The studies are compared on various parameters like models, datasets, and performance metrics. The work also identified some of the challenges like dealing with noise and also explained disease symptoms. [ABSTRACT FROM AUTHOR]

Writing is a commonplace activity that individuals partake in regularly. However, the implications behind it are often overlooked. When we write, various psychological factors come into play as the pen creates letters on the paper. Handwriting analysis has long been a subject of study, attracting researchers from diverse disciplines such as graphology, psychology, paleography, neuroscience, criminology, and computer science. Among the promising applications of handwriting analysis is gender classification, where a system can predict the gender of a writer based on their handwriting style. Since each individual's handwriting is unique, and variations exist between the handwriting of different genders, an automatic gender classification system can exploit these differences to make predictions. This paper presents a deep-learning-based gender classification system specifically designed for Urdu handwriting. The proposed approach utilizes a CNN network trained and tested on a self-created dataset contributed by 200 distinct male and 200 female Urdu writers. Through this method, the gender classification system achieved an impressive overall accuracy of 99.63%. The results obtained demonstrate that our technique for Urdu handwriting-based writer identification surpasses existing approaches. In the future, we intend to explore transfer learning techniques to further advance this field. [ABSTRACT FROM AUTHOR]

This research paper presents a review of the use of convolutional neural networks (CNNs) for human crowd analysis. The paper discusses the challenges and limitations of methods and highlights the potential of CNNs in addressing these limitations. This study reveals and provides an in-depth analysis of the different techniques, architectures, and algorithms used in CNNs for human crowd analysis and their respective advantages and limitations. Additionally, the paper discusses the potential applications of CNNs in crowd analysis, including pedestrian detection, crowd counting, and crowd behavior recognition. The review also provides insights into the performance evaluation metrics commonly used in this area and the datasets used for training and testing CNNs. Overall, this review provides a comprehensive overview of the latest developments in the use of CNNs for human crowd analysis, as well as insights into future research directions in this field. [ABSTRACT FROM AUTHOR]

Due to the considerable number of deaths and vehicle accidents caused by a driver's inattention, as reported by WHO, automobile manufacturers are aiming to combine advanced driver assistance systems (ADAS) with artificial intelligence algorithms, particularly deep learning and computer vision techniques. One feature that assists drivers is traffic sign recognition, which is a technique that allows vehicles to detect and recognize road signs placed on the road. This can be achieved by the aid of computer vision and Convolutional Neural Networks (CNN). The main aim of this research is to propose and improve a CNN based-model that can be efficiently and accurately applied for embedded applications, this might be accomplished with the help of several preprocessing algorithms. An improved network model called LeNet-5 has been developed for the classification of traffic signs. Furthermore, the proposed model network is trained using both German Traffic Sign Recognition Benchmark (GTSRB) and extended GTSRB (EGTSRB) datasets. According to the test results, the improved LeNet-5 architecture obtained an accuracy of 99.12% on GTSRB and 99.78% on EGTSRB datasets respectively, which has a positive performance compared to other state-of-the-art papers in terms of accuracy. [ABSTRACT FROM AUTHOR]

Object detection is the most important problem in computer vision tasks. After AlexNet proposed, based on Convolutional Neural Network (CNN) methods have become mainstream in the computer vision field, many researches on neural networks and different transformations of algorithm structures have appeared. In order to achieve fast and accurate detection effects, it is necessary to jump out of the existing CNN framework and has great challenges. Transformer's relatively mature theoretical support and technological development in the field of Natural Language Processing have brought it into the researcher's sight, and it has been proved that Transformer's method can be used for computer vision tasks, and proved that it exceeds the existing CNN method in some tasks. In order to enable more researchers to better understand the development process of object detection methods, existing methods, different frameworks, challenging problems and development trends, paper introduced historical classic methods of object detection used CNN, discusses the highlights, advantages and disadvantages of these algorithms. By consulting a large amount of paper, the paper compared different CNN detection methods and Transformer detection methods. Vertically under fair conditions, 13 different detection methods that have a broad impact on the field and are the most mainstream and promising are selected for comparison. The comparative data gives us confidence in the development of Transformer and the convergence between different methods. It also presents the recent innovative approaches to using Transformer in computer vision tasks. In the end, the challenges, opportunities and future prospects of this field are summarized. [ABSTRACT FROM AUTHOR]

Exposure to UV rays due to global warming can lead to sunburn and skin damage, ultimately resulting in skin cancer. Early prediction of this type of cancer is crucial. A detailed review in this paper explores various algorithms, including machine learning (ML) techniques as well as deep learning (DL) techniques. While deep learning strategies, particularly CNNs, are commonly employed for skin cancer identification and classification, there is also some usage of machine learning and hybrid approaches. These techniques have proven to be effective classifiers of skin lesions, offering promising results for early detection. The paper analyzes various researchers' reviews on skin cancer diagnosis to identify a suitable methodology for improving diagnostic accuracy. A publicly available dataset of dermoscopic images retrieved from the ISIC archive has been trained and evaluated. Performance analysis is done, considering metrics such as test and validation accuracy. The results indicate that the RF(random forest) algorithm outperforms other machine learning algorithms in both scenarios, with accuracies of 58.57% without augmentation and 87.32% with augmentation. MobileNetv2, ensemble of Dense Net and Inceptionv3 exhibit superior performance. During training without augmentation, MobileNetv2 achieves an accuracy of 88.81%, while the ensemble model achieves an accuracy of 88.80%. With augmentation techniques applied, the accuracies improved to 97.58% and 97.50%, respectively. Furthermore, experiment with a customized convolutional neural network (CNN) model was also conducted, varying the number of layers and applying various hyperparameter tuning methodologies. Suitable architectures, including a CNN with 7 layers and batch normalization, a CNN with 5 layers, and a CNN with 3 layers were identified. These models achieved accuracies of 77.92%, 97.72%, and 98.02% on the raw data and augmentation datasets, respectively. The experimental results suggest that these techniques hold promise for integration into clinical settings, and further research and validation are necessary. The results highlight the effectiveness of transfer learning models, in achieving high accuracy rates. The findings support the future adoption of these techniques in clinical practice, pending further research and validation. [ABSTRACT FROM AUTHOR]

Change detection in SAR is currently known to be a happening field of research in the domain of computer vision and remote sensing. There are numerous approaches and techniques available to detect the change in varied types of images captured in various areas. In this paper, we present a technique for detecting changes in SAR images, which generally happen to be poor contrast and poor brightness grayscale images. Consequently, they are complicated to change detection. We developed a change detection strategy in this study that employs a Convolution Neural Network (CNN) as a classification model. Further, we have used the Fuzzy Local Information C-Means approach to find out interesting pixels which have a high possibility of being changed or unchanged. By integrating the CNN classification result with the pre-classification result, the final change map is created. In this paper, we have tried to generate some virtual samples to supplement the lack of training samples. The efficiency and robustness of the planned approach have been ascertained with investigational results on four real SAR image data sets compared to several existing methodologies. [ABSTRACT FROM AUTHOR]

According to the problem that emotional features cannot be well represented by a single feature and it is difficult to extract in the task of Speech Emotion Recognition (SER), we propose a Feature-Level (FL) fusion method for 9 types of acoustic features and a combined CNN-LSTM network with attention (CNN-A-LSTM). The feature vector set after feature-level fusion is used as the input of CNN-A-LSTM network, which contains prosodic features and spectral features. High-level Statistical Functions (HSFs) are also added to receive global features. The feature extraction network CNN-A-LSTM can more effectively read time-series input and extract speech emotion information. Finally, Softmax is applied as a classifier to obtain the final emotion classification results. The experiment is verified on SAVEE and CASIA datasets. The experimental results show that the method in this paper has the best effect compared with the state-of-the-art, and the accuracy rates of 94.5% and 96.7% are respectively gained on SAVEE and CASIA. The above results prove the effectiveness of the algorithm in this paper with some generalization. [ABSTRACT FROM AUTHOR]

Postpartum haemorrhage is a type of blood loss that occurs after the birth of a baby. When you lose more than 500 ml of blood, your blood pressure drops, and you may suffer and die as a result. Deep learning techniques can predict postpartum hemorrhage earlier. As a result, we would be able to save the human. This paper discusses various types of deep learning techniques. This paper focuses on the concept of Convolutional neural networks and divides it into two sections: ZFnet and VGG-16net. By comparing the results of two nets, we can determine which of the techniques is best for predicting postpartum hemorrhage at an earlier stage. This study will be more beneficial to pregnant women in the future. The paper focuses on two nets that are said to be more useful and to be a standardized technique that also helps to give relevant medicine to patients at the appropriate time. In this paper, the algorithm is used for the VGG-16net, and the Confusion matrix is used for both nets to improve performance. Many metrics are used in this research to improve accuracy and results. Finally, the convolutional neural network concept of VGG-16net produced better results than ZF-net. [ABSTRACT FROM AUTHOR]

In order to detect the defects of smart meter liquid crystal display (LCD) screen quickly and accurately, this paper proposes a method based on line segment detector (LSD) and deep learning. Characters of smart meter LCD screen can be divided into two types: ordinary characters and digital tube characters. Convolutional neural network (CNN) is used to locate and identify the ordinary characters. Since there may be certain missing segment in digital tube characters, conventional text detection methods are not applicable, and template matching method is used for detecting them. Compared with the whole screen, the digital tube area is very small, and the relative positions of the digital tube characters of smart meter of most manufacturers are nearly the same, so the detection rate of the template matching method is extremely high and the method can be commonly used. The whole detection process is as follows: First, horizontal straight lines got by LSD are used for tilt correction. Secondly, CNN is used to locate and identify an ordinary character of the meter screen to determine whether it is missing or not. Then, whether there are missing segments in the digital tube area is determined by template matching method. At last, the experimental results show that the accuracy of the method for detecting the character defects of LCD screen is about 96%, and the accuracy of the electric quantity identification function of the method is about 97%. [ABSTRACT FROM AUTHOR]

Digital image forgeries become one of the most challenges faced by technology users today. A copy-move forgery is one of the most common types of these forgeries. This type is not only common, but it is also very simple to execute to create deceptive images. Many algorithms use various technologies to create a high-performance Copy Move Forgery Detection (CMFD) system. There are currently two major CMFD trends: traditional CMFD algorithms and deep-learning-based CMFD algorithms. The main concept behind CMFD algorithms is to extract image features using various techniques and then search for matching features. This main idea can be carried out by using different feature extraction techniques with various matching techniques. For traditional algorithms, there are many types of features extractors used such as invariant image moments, texture and intensity descriptors, Scale-Invariant Feature Transform (SIFT), Speed Up Roust Features (SURF), Discrete Cosine Transform (DCT), Singular Value Decomposition (SVD), Principal Component Analysis (PCA), and other extractors. The strength of these extractors varies according to the image properties and the forgery's performance perfection. Deep learning-based CMFD algorithms operate as a black box, by extracting image features using layers with complex training and validation processes to achieve the best scenario, such as Convolutional Neural Network (CNN), Convolutional Long Short-Term Memory (CovLSTM), Multi-domain CNN, and Buster-Net. This paper discusses different types of digital image forgery with focusing on copy-move forgery. A detailed survey of various types of CMFD algorithms, on the other hand, has been displayed. The paper provides a comprehensive comparison of various algorithms and evaluates them using various types of evaluation parameters. Finally, the paper focuses on the different datasets used by various algorithms to evaluate their performance accuracy and testing time. These datasets are detailed in terms of image numbers, the number of original images, the number of forged images, the resolution variety in each dataset, the size of the tampered region in each dataset, and the strength of each dataset compared with each other. [ABSTRACT FROM AUTHOR]

Image watermarking techniques are used to provide copyright protection and verify ownership of media/entities. This technique refers to the concept of embedding of secret data/information of an owner in a given media/entity for determining any ownership conflicts that can arise. Many watermarking approaches have been offered by various authors in the last few years. However, there are not enough studies and comparisons of watermarking techniques in soft computing environments. Nowadays, soft computing techniques are used to improve the performance of watermarking algorithms. This paper surveys soft computing-based image watermarking for several applications. We first elaborate on novel applications, watermark characteristics and different kinds of watermarking systems. Then, soft computing based watermarking approaches providing robustness, imperceptibility and good embedding capacity are compared systematically. Furthermore, major issues and potential solutions for soft computing-based watermarking are also discussed to encourage further research in this area. Thus, this survey paper will be helpful for researchers to implement an optimized watermarking scheme for several applications. [ABSTRACT FROM AUTHOR]

With the continuous in-depth study of convolutional neural network in computer vision, how to improve the performance of network structure has been the focus of current research. Recent works have shown that multi-scale feature concatenation, shortcut connection and grouping convolution can effectively train deeper networks and improve the accuracy and effectiveness of the network. In this paper, we present a novel feature transformation strategy of fragmented multi-scale feature fusion. Moreover, an efficient modularized image classification network, IX-ResNet, is proposed based on this new strategy. IX-ResNet consists of many large isomorphic modules stacked in the form of residual network while Each large module can be composed of many small heterogeneous modules. The performance of IX-ResNet is verified on cifar-10, cifar-100 and ImageNet-1 K datasets, which indicates that IX-ResNet model using fragmented multi-scale feature fusion strategy can further improve accuracy compare to the original grouping convolution network ResNeXt with the same or even lower parameters. [ABSTRACT FROM AUTHOR]

With the continuous in-depth study of convolutional neural network in computer vision, how to improve the performance of network structure has been the focus of current research. Recent works have shown that multi-scale feature concatenation, shortcut connection and grouping convolution can effectively train deeper networks and improve the accuracy and effectiveness of the network. In this paper, we present a novel feature transformation strategy of fragmented multi-scale feature fusion. Moreover, an efficient modularized image classification network, IX-ResNet, is proposed based on this new strategy. IX-ResNet consists of many large isomorphic modules stacked in the form of residual network while Each large module can be composed of many small heterogeneous modules. The performance of IX-ResNet is verified on cifar-10, cifar-100 and ImageNet-1 K datasets, which indicates that IX-ResNet model using fragmented multi-scale feature fusion strategy can further improve accuracy compare to the original grouping convolution network ResNeXt with the same or even lower parameters. [ABSTRACT FROM AUTHOR]

This paper presents survey on various techniques used to classify plants and its disease. Classification is concerned with classifying each sample into different classes. Classification is a method of separating a healthy and diseased leaf on its morphological features such as texture, color, shape, pattern and so on. Due to resemblance in the visual properties among plants, sorting and classification are complicated to carry out especially in large area. There are various methods based on image processing techniques and computer vision. Choosing the suitable classification technique is quite difficult as the result varies on different input data. Classification of leaf diseases in plants has wide applications in different fields such as agriculture and biological research. This paper provides a general idea of few existing methods, its pros and cons, state of art of different techniques used by several authors in leaf disease identification and classification such as preprocessing techniques, feature extraction and selection techniques, datasets used, classifiers and performance metrics. Apart from these some challenges and research gaps are identified and their probable solutions are pointed out. [ABSTRACT FROM AUTHOR]

This paper presents survey on various techniques used to classify plants and its disease. Classification is concerned with classifying each sample into different classes. Classification is a method of separating a healthy and diseased leaf on its morphological features such as texture, color, shape, pattern and so on. Due to resemblance in the visual properties among plants, sorting and classification are complicated to carry out especially in large area. There are various methods based on image processing techniques and computer vision. Choosing the suitable classification technique is quite difficult as the result varies on different input data. Classification of leaf diseases in plants has wide applications in different fields such as agriculture and biological research. This paper provides a general idea of few existing methods, its pros and cons, state of art of different techniques used by several authors in leaf disease identification and classification such as preprocessing techniques, feature extraction and selection techniques, datasets used, classifiers and performance metrics. Apart from these some challenges and research gaps are identified and their probable solutions are pointed out. [ABSTRACT FROM AUTHOR]

Identification of text and non-text regions in a document image is necessary before feeding it to an Optical character recognition (OCR) engine for the generation of editable version. This is because OCR engines only process text regions. Presence of non-text may constrain their performance. This makes text non-text region classification a necessary step. So far, texture based feature descriptors are widely considered for document region classification. These descriptors mostly consider the local patterns to estimate the region texture. Even the convention neural networks (CNN) also emphasise on the local connectivity. However, for better characterization of the region texture it is necessary to capture the relative context too. To address this issue, in this paper, a multi-scale dilated convolution neural network is designed to classify the document regions as text or non-text. This network can effectively capture the local patterns as well as the relative context at different scale. The proposed method is evaluated on a publicly available dataset AUTNT (Khan and Mollah, Multimed Tools Appl 78(22):32159–32186) where it outperforms the benchmark result by 1.63% by obtaining 97.91% accuracy. The proposed method also outperforms some state-of-the-art methods while evaluated on the said dataset. Additionally, the performance of the proposed network is evaluated using two standard datasets MNIST and Fashion-MNIST to observe its applicability in multi-class problem. The network obtains 99.31% accuracy in MINST dataset and 90.68% accuracy in Fashion-MNIST dataset. [ABSTRACT FROM AUTHOR]

The COVID-19 pandemic has spread worldwide for over 2 years now. The pandemic raises a significant threat to global health due to its transmissibility and high pathogenicity. The current standard detection method for COVID-19, namely, reverse transcription–polymerase chain reaction (RT–PCR), is slow and inaccurate to help fight the pandemic. RT–PCR takes hours to days to report a single test result and has a high false-negative rate. As a result, an infected person with a negative test result may unknowingly continue to spread the virus. Thus, better detection methods are required to improve the control of COVID-19. With technology advancements in artificial intelligence and machine learning, deep-learning diagnostic studies to detect COVID-19 infection using medical chest imaging have emerged. In this paper, we review these studies by analyzing their approaches and highlighting their major challenges. These challenges include dataset cleanness, public dataset availability, capability to differentiate COVID-19 from unrelated viral pneumonia, and the difficulty in dealing with images from multiple points of view. Finally, we discuss various ideas and solutions to address the highlighted challenges in the reviewed papers. [ABSTRACT FROM AUTHOR]

The principal objective of video/image captioning is to portray the dynamics of a video clip in plain natural language. Captioning is motivated by its ability to make the video more accessible to deaf and hard-of-hearing individuals, to help people focus on and recall information more readily, and to watch it in sound-sensitive locations. The most frequently utilized design paradigm is the revolutionary structurally improved encoder-decoder configuration. Recent developments emphasize the utilization of various creative structural modifications to maximize efficiency while demonstrating their viability in real-world applications. The utilization of well-known and well-researched technological advancements such as deep Convolutional Neural Networks (CNNs) and Sentence Transformers are trending in encoder-decoders. This paper proposes an approach for efficiently captioning videos using CNN and a short-connected LSTM-based encoder-decoder model blended with a sentence context vector. This sentence context vector emphasizes the relationship between the video and text spaces. Inspired by the human visual system, the attention mechanism is utilized to selectively concentrate on the context of the important frames. Also, a contextual hybrid embedding block is presented for connecting the two vector spaces generated during the encoding and decoding stages. The proposed architecture is investigated through well-known CNN architectures and various word embeddings. It is assessed using two benchmark video captioning datasets, MSVD and MSR-VTT, considering standard evaluation metrics such as BLEU, METEOR, ROUGH, and CIDEr. In accordance with experimental exploration, when the proposed model with NASNet-large alone is viewed across all three embeddings, the BERT findings on MSVD Dataset performed better than the results obtained with the other two embeddings. Inception-v4 outperformed VGG-16, ResNet-152, and NASNet-Large for feature extraction. Considering word embedding initiatives, BERT is far superior to ELMo and GloVe based on the MSR-VTT dataset. [ABSTRACT FROM AUTHOR]

Image forensics is essential for detecting image manipulation, authenticating images, and identifying sources of images. A forensic analyst can make use of various artifacts to develop a powerful forensic technique. These artifacts include JPEG blocking and quantization artifacts, streaking artifacts and contrast enhancement artifacts, etc. With the introduction of anti-forensics, it has become difficult for forensic experts to identify forged images. There are various anti-forensic methods available that try to eradicate these detection footprints/artifacts to fool the existing forensic detectors. Thus the detection of anti-forensic attacks is very crucial and plays a vital role in forensic analysis. This paper presents a review of various types of anti-forensic attacks, such as JPEG anti-forensics, Contrast enhancement anti-forensics, and Median filtering anti-forensics. Firstly a brief introduction is given about image forgery, JPEG compression, contrast enhancement, and median filtering. Then, anti-forensics is described in detail, and finally, the recent state-of-the-art anti-forensic techniques are summarized in tabular form for better understanding. This may be helpful for the forensic analyst to develop robust methods for forgery detection that can be applied in various applications such as the identification of cybercrimes, identity thefts, etc. [ABSTRACT FROM AUTHOR]

Farmers are concerned about the automatic detection of lesions and pests that threaten tomato plants. Traditional computer vision and pattern recognition technologies have limits when it comes to tackling such difficult problems. However, deep learning had gained popularity in recent years, particularly for the detection and recognition of biotic stress in diseased leaf photos of plants with varying lighting conditions, complicated backdrops, and background noise. In this paper, a system was provided that uses several "Convolutional Neural Networks (CNN)" for the automatic recognition and identification of multi-biotic tomato leaf lesions collected from PlantVillage. For instance segmentation, a "Mask R-CNN" network is used in the first phase; for semantic segmentation a Hybrid Deep Segmentation Convolutional Neural Network Model (Hybrid-DSCNN) model is compared with U-Net and Seg-Net in the second phase, and a CNN model is used for classification in the third stage. The two backbone feature extractors were employed in the Mask R-CNN network, and the ResNet50 displays average precision of 73.00% in the instance segmentation test, which is superior to other models. For the segmentation and classification tasks, the Hybrid-DSCNN 2Layer-Convo-USN has achieved an accuracy of 98.25% which is better than the pre-trained models. The precision of the proposed Hybrid-DSCNN 2Layer-Convo-USN is 95.7% and of U-Net is 94.9%. The results are positive, indicating that the entire system may be implemented in any platform that can be used in the real world. [ABSTRACT FROM AUTHOR]

People's physical presence in smart cities is being shifted to the cyber world via social networks. Bullying that occurs through virtual devices such as phones, laptops, and tablet and PCS is known as cyberbullying. Online harassment can occur via text messaging, messages, and software, as well as online in social networking sites, discussion boards, or online games where individuals may view, participate in, or distribute content, including sending, posting, or sharing negative, harmful, false, or mean content. It can include revealing personal or private details, causing embarrassment or humiliation. This paper intends to sift through social media posts in order to identify bullying comments in text, image, and video form.We propose a graph convolutional neural network(GCN) and a pre-trained Googlenet for text and image recognition, as well as a Mel-scale filter bank speech spectrogram and CNN network model for audio post classification. This study's main results are that using graph convolution neural networks and Googlenet, as well as audio post-processing using MFCC's, produce superior results including one dimensional representation.We use in the present environment,GCN and Mel-frequency cepstrum to represent text, image, and video input, yielding an accuracy of 96%. The approach is unique in its use of these techniques and has not been previously proposed for cyberbullying detection. The study's main contribution is its ability to achieve superior accuracy in identifying bullying comments, which makes it a valuable addition to the existing literature on cyberbullying detection. [ABSTRACT FROM AUTHOR]

Intelligent Transportation System (ITS), including unmanned vehicles, has been gradually matured despite on road. How to eliminate the interference due to various environmental factors, carry out accurate and efficient traffic sign detection and recognition, is a key technical problem. However, traditional visual object recognition mainly relies on visual feature extraction, e.g., color and edge, which has limitations. Convolutional neural network (CNN) was designed for visual object recognition based on deep learning, which has successfully overcome the shortcomings of conventional object recognition. In this paper, we implement an experiment to evaluate the performance of the latest version of YOLOv5 based on our dataset for Traffic Sign Recognition (TSR), which unfolds how the model for visual object recognition in deep learning is suitable for TSR through a comprehensive comparison with SSD (i.e., single shot multibox detector) as the objective of this paper. The experiments in this project utilize our own dataset. Pertaining to the experimental results, YOLOv5 achieves 97.70% in terms of mAP@0.5 for all classes, SSD obtains 90.14% mAP in the same term. Meanwhile, regarding recognition speed, YOLOv5 also outperforms SSD. [ABSTRACT FROM AUTHOR]

This paper provides a comprehensive overview of the applications of game theory in deep learning. Today, deep learning is a fast-evolving area for research in the domain of artificial intelligence. Alternatively, game theory has been showing its multi-dimensional applications in the last few decades. The application of game theory to deep learning includes another dimension in research. Game theory helps to model or solve various deep learning-based problems. Existing research contributions demonstrate that game theory is a potential approach to improve results in deep learning models. The design of deep learning models often involves a game-theoretic approach. Most of the classification problems which popularly employ a deep learning approach can be seen as a Stackelberg game. Generative Adversarial Network (GAN) is a deep learning architecture that has gained popularity in solving complex computer vision problems. GANs have their roots in game theory. The training of the generators and discriminators in GANs is essentially a two-player zero-sum game that allows the model to learn complex functions. This paper will give researchers an extensive account of significant contributions which have taken place in deep learning using game-theoretic concepts thus, giving a clear insight, challenges, and future directions. The current study also details various real-time applications of existing literature, valuable datasets in the field, and the popularity of this research area in recent years of publications and citations. [ABSTRACT FROM AUTHOR]

Skin cancer is caused by genetic uncertainty or an irregular growth of cells, mostly grows when our skin is exposed to sun. In people, melanoma is a common type of cancer irrespective of their age, gender or race. Skin cancer has mainly two types: malignant melanoma and non-melanoma. Now days, the leading cancer type in people is melanoma among many white-skinned populations, while non-melanoma skin cancer is common in light-skinned population. The melanoma can transfer to different body parts if it is left untreated. The computerized system is needed for melanoma cancer risk evaluation, as the dermatologist monitors the condition of the patient by using skin cancer images and dermatoscope. Its identification in advance, needs the proper identification of spots on skin which is based on some features, which are extracted. Different segmentation methods play very important role in skin lesion detection from images, in which segmentation performs the image classification, into the pixels of skin and non- skin based upon the texture of skin. The convolutional neural networks (CNNs) have presented better outcomes in the skin cancer classification as compared to the findings of dermatologists. This model will help people to get help from their installed applications on mobile phone or electronic devices, for the early diagnosis of cancer. In this review study, various research papers have been presented on the skin lesions classification based on CNNs. We have discussed how CNNs have a great impact in successful skin cancer classification and methods which are implemented with success rates. Deep learning using CNN also has many advantages and, there is some vulnerability as well in misclassifying the images, under some situations. In this paper we searched IEEE, Science Direct, Google Scholar, Elsevier, PubMed and Web of Science databases to obtain original published research articles and selected papers providing necessary information related to the research and finally we have produced a systematic review on published methodologies, datasets, algorithms, results, accuracy, sensitivity etc. using CNN model. [ABSTRACT FROM AUTHOR]

Signature is a biometrics trait widely used for personal verification in financial and most other organizations where financial and relevant types of transactions are done manually with signed authorized papers. A signature verification system aims to verify the minor structural differences between genuine and forged signatures, as most skilled forgery signatures look similar to their respective genuine signatures with specific deformations. Signature verification in a multi-cultural country like India is challenging in both writer-independent and script-independent scenarios where the Indian population uses multiple scripts to write their signatures. This paper reports a writer-independent offline signature verification system that uses Convolutional Neural Network (CNN) architecture for feature extraction and classification. The objective of the proposed work is to model a CNN-based adaptable system to verify multi-scripted offline signatures. The model has been trained and tested on two publicly available databases, viz. CEDAR and BH-Sig260, which consist of Hindi, Bengali, and English signatures. Individual signature classes of unique scripts and a combination of these scripts have been considered for testing the proposed model that determines verification accuracies of 90%, 95%, 98.33%, and 93.33%, respectively. Experimental results are compelling, and the proposed model outperforms the verification accuracies of some well-known models. [ABSTRACT FROM AUTHOR]

An unusual increase of nerves inside the brain, which disturbs the actual working of the brain, is called a brain tumor. It has led to the death of lots of lives. To save people from this disease timely detection and the right cure is the need of time. Finding of tumor-affected cells in the human brain is a cumbersome and time- consuming task. However, the accuracy and time required to detect brain tumors is a big challenge in the arena of image processing. This research paper proposes a novel, accurate and optimized system to detect brain tumors. The system follows the activities like, preprocessing, segmentation, feature extraction, optimization and detection. For preprocessing system uses a compound filter, which is a composition of Gaussian, mean and median filters. Threshold and histogram techniques are applied for image segmentation. Grey level co-occurrence matrix (GLCM) is used for feature extraction. The optimized convolution neural network (CNN) technique is applied here that uses whale optimization and grey wolf optimization for best feature selection. Detection of brain tumors is achieved through CNN classifier. This system compares its performance with another modern technique of optimization by using accuracy, precision and recall parameters and claims the supremacy of this work. This system is implemented in the Python programming language. The brain tumor detection accuracy of this optimized system has been measured at 98.9%. [ABSTRACT FROM AUTHOR]

In recent years, machining learning aided diagnosis can provide non-invasive, low-cost tools to support clinicians and assist the diagnosis and monitoring of neurodegenerative disorders, in particular Parkinson's disease (PD). As an important motor symptom, disorder of the hand motion is usually used for diagnosis and evaluation of PD; moreover, majority of the patients with PD have handwriting abnormalities, which plays a special role in PD detection. In this paper, as an useful tool, we propose a novel hybrid model to learn the handwriting differences between PD patients and healthy controls, by learning and enhancing significant features from three handwriting exams, i.e., meander, circle and spiral. Based on a three-layer convolutional neural network (CNN) and a bidirectional gated recurrent unit (BiGRU), the proposed network can assess the potential of sequential information of handwriting in identifying Parkinsonian symptoms. Compared with several state of the art studies, the recognition rates of our proposed framework are 92.91%, 85.71% and 90.55% respectively in these three tests, which verifies the excellent classification effect. [ABSTRACT FROM AUTHOR]

With the development of artificial intelligence, the application of AI technology in the media industry is in progress. Video auto-editing is one of the directions. In video editing, the shot scale is am important reference for shot arrangement. The existing algorithms tend to classify the shot scale based on CNN, but fail to work well on all kinds of frames with various aspect ratios. One of the focuses in this paper is to explore the relationship between pooling method and location bias in CNN, so that location features and non-location features could be treated reasonably to reach a better classification performance on kinds of frames with various aspect ratios. In a set of interesting experiments, we change the output feature maps of pooling(OFMP) to observe how CNN classify a group of images by location features and non-location features. Then, a vertical and horizontal pooling method(VH-Pooling) is proposed for a robust shot scale classification, which achieves 94.24% accuracy on a multi-aspect-ratio shot scale dataset within a high operation speed. Finally, a practical shot scale classification system is designed with a post-processing module, and successfully applied in a live news AI-editing platform. [ABSTRACT FROM AUTHOR]

In today's world, video captioning is extensively used in various applications for specially-abled and, more specifically, visually abled persons. With advancements in technology for object detection and natural processing, there has been an instant surge infusing the above mainstream tasks. One such example of this fusion resulted in the generation of Image captions when an input image is fed to the system, and it gives a short description of what is present in the image. This fusion pertained to images and was further moved to be implemented on the Videos, with some tweaking in the current methods. This paper presents the survey of the state of art techniques of various video captioning methods. There have been many inputs provided by people worldwide in this domain; thus, there was a need to compile, study and analyze all the results and present that in a comprehensive study, which we have done in this paper. The comparison of various video captioning methods on the distinct dataset was evaluated on different parameters, which were most common and mainly used for image and video analysis. This review was done for methods used from the year 2015–2019 (year by year). The most commonly used dataset and evaluation method are also pictorially represented in a bar graph and scatter plot for each year for the respective evaluation parameter. Though a lot of analysis and research has been done on video captioning, our survey shows many problems. [ABSTRACT FROM AUTHOR]

Iris recognition is a secure and best-chosen biometric application in the digital world because of its unique characteristics. Day by day, the digital world plays a significant role in human life for various applications. The applications are vastly spread over secure applications of the nation such as border control applications, criminal investigations, postmortem studies, access the digital equipment, smart homes, smart appliances, smart cars etc. Due to the digitalization of the world, all the research communities, scientists, and industries are focusing on the biometric-based secured iris recognition system. Several researchers have done much work in this domain, but there is still a scope of improvement for various reasons, i.e., less speed and accuracy of the module. The researcher has implemented various algorithms based on traditional and neural network architectures. In this scenario, this paper gives a brief on different techniques and algorithms used by researchers to predict the age of human people using the iris. This paper discussed one hundred and one papers in the literature with various image segmentation, feature extraction and classification of the iris. This paper summarizes publicly available standard databases and various evaluation parameters, i.e., accuracy, precision, recall, f-score, etc. The research community evaluated the age prediction through the iris-based state-of-the-art algorithms with secure prediction, i.e., TPR, TNR, FPR, FNR. Finally, this paper provides the strengths and weaknesses of the various state of art algorithms, respectively, and summarizes the gaps in the existing technology with the scope of improvement. [ABSTRACT FROM AUTHOR]

There is an ever-increasing number of public users of the internet just to remain learned and updated as well as distribute millions of posts, articles, and videos across platforms such as Twitter, Facebook and YouTube etc. The speedy acceptance of social media has forced to an explosive growth in information-distribution among users, with fake news becoming a fundamental intrusion of our digital day-to-day life. The spread of disinformation is in part associated to social media failing to verify the correctness of a news item. Although it is technically possible, it is challenging for a deep neural network to "attend to" the false alone portions of each news piece because the linguistics of fake news will approximate that of true news. In this paper, a Simple Shallow Perceptron Network (SSPN) has been proposed and will reduce the complications such as the curse of high dimensionality, high training time of Deep Neural Networks, over-fitting and over-thinking problem during network training. The effectiveness of the proposed SSPN model for false news identification using the benchmark LIAR dataset and the more recent FakeNewsNet dataset have also been examined. The proposed model outperformed all other existing models, with an F1-score of 0.73 for the LIAR dataset and 0.90 for the FakeNewsNet dataset. [ABSTRACT FROM AUTHOR]

For the safety and reliability of the system, Remaining Useful Life (RUL) prediction is considered in many industries. The traditional machine learning techniques must provide more feature representation and adaptive feature extraction. Deep learning techniques like Long Short-Term Memory (LSTM) achieved an excellent performance for RUL prediction. However, the LSTM network mainly relies on the past few data, which may only capture some contextual information. This paper proposes a hybrid combination of Convolution Neural Network (CNN) and LSTM (CNN+LSTM) to solve this problem. The proposed hybrid model predicts how long a machine can operate without breaking down. In the proposed work, 1D horizontal and vertical signals of the mechanical bearing are first converted to 2D images using Continuous Wavelet Transform (CWT). These 2D images are applied to CNN for key feature extraction. Ultimately, these key features are applied to the LSTM deep neural network for predicting the RUL of a mechanical bearing. A PRONOSTIA data is utilized to demonstrate the performance of the proposed model and compare the proposed model with other state-of-the-art methods. Experimental results show that our proposed CNN+LSTM-based hybrid model achieved higher accuracy (98%) with better robustness than existing methods. [ABSTRACT FROM AUTHOR]

Frame deletion detection is a challenging task in the field of digital video forensics. This paper proposes a deep-learning-based frame deletion detection method for single-shot videos. We capture traces of frame deletion forgery from both adjacent and long-range continuous frames. Specifically, we propose a novel multi-scale difference convolutional neural network (MSDCNN) structure, which models different levels of inter-frame variations. Then, we use the long-short-term memory network (LSTM) to capture the long-term variation pattern of multi-scale differential features. The proposed method is a simple and principled frame deletion detection framework with a small computational cost. According to the experiments, the proposed framework can achieve a more advanced performance of frame deletion detection than traditional methods and methods based on 3D convolutions. [ABSTRACT FROM AUTHOR]

In this digital age, 3D data interpretation has emerged as a significant research area in which videos are the most extensively utilized electronic medium for data transfer. The appropriate classification of video data is critical for storing and broadcasting. In this paper, we have introduced a novel attention-based two-stream deep neural network (A2SN) for sports video classification. The first stream of A2SN is a transfer learning-based model that transfers the pre-trained weights from base to head model for spatial feature learning. In contrast, the second stream is an attention-based feature extractor module used to learn the spatiotemporal features of video data. After that, the features obtained from the transfer learning-based and attention-based models are concatenated together and passed through the dense layer for the classification task. An extensive experiment has been performed on UCF50 and SVW datasets, and an accuracy of 99.26% and 97.3% is achieved, respectively, that validate the efficacy of the proposed model. We have also compared the performance of A2SN on the UCF101 dataset and achieved an accuracy of 97.1% to check the model's generalizability. [ABSTRACT FROM AUTHOR]

Deep learning algorithms have shown exceptional effectiveness in a wide range of supervised and unsupervised learning tasks in a variety of fields, including image processing, computer vision, natural language processing, and speech or voice processing. In this paper, a comprehensive analysis is conducted to assess the impact of deep learning on user authentication using both physiological and behavioural biometrics. This work encompasses the diverse deep learning approaches employed in authentication schemes tailored for smart devices. Meticulous scrutiny of commonly used datasets in these authentication studies is undertaken, accompanied by a comparative assessment of performance. The deep learning models under consideration span a spectrum of architectures, including deep neural networks, convolutional neural networks, deep auto-encoders, recurrent neural networks, and their variants. To enhance the clarity and categorization of authentication techniques for smart devices, a taxonomy is proposed based on the specific authentication metrics employed: (1) Knowledge-based Authentication (KBA), (2) Physiological Biometric-based Authentication (PBBA), (3) Behavioural Biometric-based Authentication (BBBA), (4) Physiological and Behavioural Continuous Authentication (PBBCA), and (5) Multi Modal Authentication (MMA). Furthermore, potential contributions of deep learning techniques to the realms of physiological and behavioural biometrics are discussed. Significance is placed on performance metrics, including accuracy, stability, and robustness, in evaluating these deep learning-based authentication systems. The challenges and limitations that deep learning approaches must surmount when dealing with real-world biometric data in the context of biometric identification systems are examined. This work not only underscores the transformative role of deep learning in user authentication but also offers valuable insights into the evolving landscape of biometric identification on smart devices. An examination of performance metrics provides a holistic view of the strengths and areas for improvement in deep learning based authentication solutions. [ABSTRACT FROM AUTHOR]

Phishing is an attack that attempts to replicate the official websites of businesses, including government agencies, financial institutions, e-commerce platforms, and banks. These fraudulent websites aim to obtain sensitive information from users, such as credit card numbers, email addresses, passwords, and personal identities. In response to the increasing number of phishing assaults, several anti-phishing strategies have been developed. However, existing techniques often fail to extract the most crucial features, leading to potential misclassification. Additionally, the complex algorithms employed result in high response times. To address these challenges, this paper proposes a novel approach called Bidirectional Long Short-Term Memory based Gated Highway Attention block Convolutional Neural Network (BiLSTM-GHA-CNN) for detecting phishing URLs. The BiLSTM captures contextual features, while the CNN extracts salient features. The integration of the highway network into the BiLSTM-CNN architecture enables the capture of significant features with rapid convergence. Furthermore, a gating mechanism is employed to weigh the output features of the CNN and BiLSTM. Five datasets from diverse sources such as Phish Tank and Open Phish were created for experimentation. The results demonstrate that BiLSTM-GHA-CNN achieves superior detection accuracy, precision recall, and F1-score compared to state-of-the-art techniques. Moreover, the proposed system significantly reduces the response time to a remarkable 12.46 ms. [ABSTRACT FROM AUTHOR]

Crop diseases are highly severe in agriculture which affect crop production. Apple is one of the most important fruit worldwide for economic and nutritional reasons. Even though much state-of-the-art has been done on crop disease detection and recognition, fungal disease-affected crops are still untouched. Fungal diseases can produce mycotoxins, which are toxic compounds that can contaminate food and feed, causing illness or death in humans and animals. This paper mainly focuses on detecting fungal diseases in apple crops. The improved ResNeXt, a deep learning model is proposed for the detection of apple crop diseases. The benchmark dataset has been used in the proposed model which contains 9395 images of diseases having 4 classes of apple. Initially, the transfer learning approach such as Inception-v7, ResNet on crops having fungal disease has been applied. Both approaches were not giving desired results. Further, an improved ResNeXt model variant of the Convolution neural network (CNN) for Fungal disease prediction has been proposed. Moreover, pre-processing step have been done for handling imbalanced dataset and also segmentation has been done to extract region of interest in crop. Comparison of the proposed model with the state-of-the-art approaches has been done. The proposed model attained an accuracy of 98.94%, recall of 99.2%, precision of 99.4%, and 99.2% of F1 score. The findings of this study can be utilized by the community for improving the crop production, as it demonstrates the potential for deep learning techniques to aid in the early detection of crop diseases, leading to improved crop yields and quality. [ABSTRACT FROM AUTHOR]

Video capsule endoscopy (VCE) is a non-invasive procedure to examine the human bowel. The VCE technology generates thousands of images from different parts of the gastrointestinal tract. Since the examination of these images is a tedious and time-consuming task for doctors, automated diagnosis of digestive diseases from VCE images is highly desired. The majority of the existing studies are based on CNN methods, which are not efficient enough in learning invariant global features in VCE images. Therefore, this paper presents a new framework that combines the learning of global and local features from VCE images. The proposed method utilizes a specific attention mechanism within a convolutional neural network to extract local features, while a vision transformer captures global features. Both local and global features are fused for final classification. Extensive experiments were performed on the public Kvasir Capsule Endoscopy dataset, revealing a promising accuracy of 97%. These results not only highlight the model's capabilities but also demonstrate its favorable standing when compared to the state-of-the-art methods. Additionally, achieving a recall of 85%, the proposed system demonstrated robust generalization capabilities, performing impressively on an unseen dataset. [ABSTRACT FROM AUTHOR]

Nuclear power plants' instrumentation and control systems (I&Cs) are essential for guaranteeing secure and dependable operations. If these systems fail, there could be serious accidents, radiation exposure risks, and environmental damage. To avoid catastrophic outcomes, the dependability of I&C systems in NPPs is of the utmost importance. With the help of the suggested methodology, the instrumentation and control system's dependability are examined efficiently. This paper aims to employ deep learning models to analyze the I&C system's reliability. A novel fault monitoring system with reliability analysis of Safety–Critical Systems (SCS) is introduced based on the hybrid deep learning (HDL) Model. The hybrid deep learning model is developed based on a convolutional neural network (CNN) and an optimized Recurrent Neural Network (RNN) (CNN-ORNN). In addition to that, a Weighted Kurtosis and Skewness (WKS)-based Aquila Optimizer Algorithm (AOA) is developed to enhance the feature learning strategy. The automated feature learning is achieved in the HDL, where a new Self-Improved Aquila Optimizer (SIAO) is utilized for an optimized Recurrent Neural Network. SIAO is a conceptual enhancement of the standard Aquila optimizer. To demonstrate the assessment, a nuclear power plant's isolation condenser system is utilized and implemented in Matlab. Moreover, the performance of the proposed model is evaluated based on the performance matrices, i.e., accuracy, specificity, sensitivity, False Positive Rate (FPR), and False Negative Rate (FNR). In addition, reliability is also utilized because it is a major parameter for evaluating the performance of the proposed model for I&C systems in NPPs. [ABSTRACT FROM AUTHOR]

Most solitary finger impression check and acknowledgment frameworks / methods are based on the minutiae feature points. Feature Extraction is a fundamental advance in solitary finger impression based acknowledgment frameworks. In this paper, a CNN based finger impression affirmation strategy is proposed without preprocessing an image. The framework fuses two phases include extraction and coordinating. Feature elicitation is realized by different filters with different parameter set; matching juncture relates extracted features and creates a corresponding score. Recognition attainment of the preferred system has been tested by utilizingFVC2004 database. The inference is very favoring for implementing a CNN based self-regulating fingerprint recognition system. Our method achieves an overall rate of 99.1% of accurately classified samples. [ABSTRACT FROM AUTHOR]

Recently, the speaker recognition techniques have been widely attractive for their extensive use in many fields, such as speech communications, domestic services, security and access control and smart terminals. Today's interactive devices like smart-phone assistants and smart speakers need to deal with short duration speech segments. However, existing speaker recognition applications perform poorly when short utterances are available and require relatively long speech to perform well. Aiming at solving this problem, we introduce in this paper, a novel method to enhance the speaker recognition capability with short utterance speaker recognition applications. For this purpose, we considered new deep neural network architectures based on convolutional neural network (CNN) and recurrent neural network (RNN). The proposed method is evaluated with the standard i-vector based on Probabilistic Linear discriminant analysis (PLDA) approach. The experimental results show that our model could outperform the i-vector -PLDA baseline system and enhance the speaker recognition capability when significant and short utterance duration are used. [ABSTRACT FROM AUTHOR]

Rehabilitation robots are gaining significant popularity for impaired gait rehabilitation. However, to make the recovering individual feel natural while walking and restore their original gait pattern, adapting the rehabilitation system according to the individual's need and walking characteristics becomes imperative. In this paper, we have compared four deep learning models for their ability to generate a personalized gait trajectory at different gait speeds. The first three models are primitive and are the basic implementations of long short term memory (LSTM), convolutional neural network (CNN) and gated recurrent unit (GRU). The fourth model is our proposed model, which is a sequential combination of LSTM and CNN. We considered hip, knee and ankle joints data as human gait is represented as the joint angle trajectories of these joints in the sagittal plane. We trained these models on a benchmark public human walking dataset consisting of treadmill walking data of 42 healthy individuals at eight different walking speeds. Anthropometric and demographic data along with gait speeds were given as input to the models. Our proposed LSTM-CNN sequential model is able to generate stable gait trajectories in the speed range of 0.49-1.76 m/s with a high correlation of 0.98 between the actual and the predicted trajectories, and an R2 Score of 0.94 is obtained. This work can be utilized for providing personalized gait reference trajectories for the rehabilitation of amputees and stroke patients using rehabilitation systems such as exoskeleton robots and prosthetic legs. Also, this work can be utilized for generating stable walking trajectories for bipedal robots. [ABSTRACT FROM AUTHOR]

Human posture recognition is an important task for intelligent systems specially those performing action recognition. In this paper, we propose a novel multispectral corner detector and a new HOG-based multispectral local descriptor. First, we select salient features which are extracted from an edge image obtained by picking the maximum eigenvalue of the jacobian matrix. Second, we extract for each feature point a local descriptor which combines both the Lab colour channels and depth information in a well-posed way using the Jacobian matrix. Last, we conduct a one-against-all learning strategy using both an incremental Covariance-guided One-Class Support Vector Machine (iCOSVM) and a Convolutional Neural Network (CNN). Experimental results show that we outperform the state-of-the-art methods whether our descriptor is combined with iCOSVM and with CNN. [ABSTRACT FROM AUTHOR]

COVID-19 has engulfed over 200 nations through human-to-human transmission, either directly or indirectly. Reverse Transcription-polymerase Chain Reaction (RT-PCR) has been endorsed as a standard COVID-19 diagnostic procedure but has caveats such as low sensitivity, the need for a skilled workforce, and is time-consuming. Coronaviruses show significant manifestation in Chest X-Ray (CX-Ray) images and, thus, can be a viable option for an alternate COVID-19 diagnostic strategy. An automatic COVID-19 detection system can be developed to detect the disease, thus reducing strain on the healthcare system. This paper discusses a real-time Convolutional Neural Network (CNN) based system for COVID-19 illness prediction from CX-Ray images on the cloud. The implemented CNN model displays exemplary results, with training accuracy being 99.94% and validation accuracy reaching 98.81%. The confusion matrix was utilized to assess the models' outcome and achieved 99% precision, 98% recall, 99% F1 score, 100% training area under the curve (AUC) and 98.3% validation AUC. The same CX-Ray dataset was also employed to predict the COVID-19 disease with deep Convolution Neural Networks (DCNN), such as ResNet50, VGG19, InceptonV3, and Xception. The prediction outcome demonstrated that the present CNN was more capable than the DCNN models. The efficient CNN model was deployed to the Platform as a Service (PaaS) cloud. [ABSTRACT FROM AUTHOR]

The most dangerous type of skin cancer in the world is Melanoma. Early diagnosis of this cancer in primary stages can increase the chance of surviving death. In recent years, automatic skin cancer detection systems have played a significant role in increasing the rate of cancer diagnosis. Although deep convolutional neural networks presented advantages over traditional methods and brought tremendous breakthroughs in many image classification tasks, accurate classification of skin lesions remains challenging due to the complexity of choosing appropriate architecture for deep neural networks and hyper-parameter tuning. The aim of this paper is to increase the performance of skin lesion classification system through optimizing hyper-parameters and architecture of deep neural network using metaheuristic optimization algorithms. For this purpose, three optimization algorithms are employed to find an optimal configuration for the convolutional neural network either in pre-trained models or model that are trained from scratch. Then the deep features extracted from the optimized models were fused together in pairs and used to train a KNN classifier. The effect of applying hyper-parameter optimization is evaluated on ISIC 2017 and ISIC 2018 datasets. The accuracy of the deep neural network produced by our method reaches to 81.6% and F1-score of 80.9% on ISIC 2017 dataset and accuracy of 90.1% and F1-score of 89.8% on ISIC 2018. The results of the present study indicate that the proposed method outperforms similar methods in classifying seven and three classes of images, without requiring heavy preprocessing and segmentation steps. [ABSTRACT FROM AUTHOR]

Image forgery detection is the basic key to solve many problems, especially social problems such as those in Facebook, and court cases. The common form of image forgery is the copy-move forgery, in which a section of the image is copied and pasted in another location within the same image. In this type of image forgery, it is easy to perform forgery, but more difficult to detect it, because the features of the copied parts are similar to those of the other parts of the image. This paper presents an approach for copy-move forgery detection based on block processing and feature extraction from the transforms of the blocks. In addition, a Convolutional Neural Network (CNN) is used for forgery detection. The feature extraction is implemented with serial pairs of convolution and pooling layers, and then classification between the original and tampered images is performed with and without transforms. A comparison study between different trigonometric transforms in 1D and 2D is presented for detecting the tampered parts in the image. This comparison study is based on the completeness rate for the detection. This comparison ensures that the DFT in 1D or 2D implementations is the best choice to detect copy-move forgery compared to other trigonometric transforms. In addition, the paper presents a comparison study between ten cases using the CNN learning technique to detect the manipulated image. The basic idea is to use a CNN to detect and extract features. The proposed CNN approach can also be used for active forgery detection because of its robustness to detect the manipulation of digital watermarked images or images with signatures. [ABSTRACT FROM AUTHOR]