Advances in Science and Technology Vol. 124

Abstract: Sign Language is a medium of communication for many disabled people. This real-time Sign Language Recognition (SLR) system is developed to identify the words of American Sign Language (ASL) in English and translate them into 5 spoken languages (Mandarin, Spanish, French, Italian, and Indonesian). Combining the study of facial expression with the recognition of Sign Language is an attempt to understand the emotions of the signer. Mediapipe and LSTM with a Dense network are used to extract the features and classify the signs respectively. The FER2013 data set was used to train the Convolutional Neural Network (CNN) to identify emotions. The system was able to recognize 10 words of ASL with an accuracy of 86.33% and translate them into 5 different languages. 4 emotions were also recognized with an accuracy of 73.62%.

Abstract: Agriculture improvement is a global economic issue and ongoing challenge in this covid-19 pandemic that is highly dependent on effectiveness. The recognition of the diseases in plant leaf performs a major role in the agriculture industry and city-side greenhouse farms approximate analysis of the leaf disease this article intends to integrate image processing techniques with the “convolutional neural network”, which is one of the deep learning approaches, to classify and detect plant leaf disease and publicly available plant the late data to help treat the leaf as early as possible, which controls the economic loss. This paper has a set that was used which consists of 10 classes of disease and three classes of a plant leaf, this research offers an effective method for detecting different diseases in plant leaf variations. The model was created to detect and recognize healthy plant kinds, such as tomato and potato, and pepper these three leaves will perform under the algorithm called a convolutional neural network. By modifying the parameters and changing the pulling combination, models that have been used to train and test these types of leaf sample images can be created. leaf disease recognition was based on these 10 different types of classes in three different species tomato, potato, and pepper the classification of sample images has reached diseases identification accuracy.

Abstract: Face masks and social distancing is essential for many infectious diseases that spread through micro-droplets. According to WHO, the preventive measure for COVID-19 is to follow social distancing. Face Detection has expanded as a widespread concern in Computer Vision and Image processing. Many unique algorithms are developed using Convolutional Architectures to curate the algorithm as accurately as possible. First, the person in the video frames is pinpointed with the aid of Deep Learning (DL). The second step is to calculate the span between any two individuals through approaches of image processing. We aim to use a binary face classifier which will help us analyze the frame and help in detecting any face present irrespective of its alignment. Our proposed technique helps in generating accurate face segmentation masks from any random-sized input picture. We start by using an RGB image of any size, the approach uses Predefined Training for feature extraction. Faster Regional Convolutional Neural Networks are used for training. Here we segment out the faces present in the given image or frame semantically. Faster Regional Convolutional Neural Network (FRCNN) processes the given data faster with higher accuracy. The precision and decision-making are very elevated in Faster RCNN compared to others.

Abstract: Sign language detection technique is a part of technology which is of extreme importance to the society. Sign languages is used by deaf and dumb people who are unable to communicate directly using sound since they lack the ability to produce or recognize sound waves which enable us to communicate easily. The proposed project aims in decreasing the distance between the sign language detection techniques which only focuses on detecting the meaning of letters like ASL and not actions provided by the users. The project detects sign languages by using key holes as the position locator and then trains the system to detect accordingly. Keyholes are used to find the position of gesture to use LSTM throughout coaching of the information. Experimental results demonstrate the efficaciousness of the planned methodology in sign language detection task

Abstract: Children and the elderly are most susceptible to brain tumors. It's deadly cancer caused by uncontrollable brain cell proliferation inside the skull. The heterogeneity of tumor cells makes classification extremely difficult. Image segmentation has been revolutionized because of the Convolution Neural Network (CNN), which is especially useful for medical images. Not only does the U-Net succeed in segmenting a wide range of medical pictures in general, but also in some particularly difficult instances. However, we uncovered severe problems in the standard models that have been used for medical image segmentation. As a result, we applied modification and created an efficient U-net-based deep learning architecture, which was examined on the Brain Tumor dataset from the Kaggle repository, which consists of over 1500 images of brain tumors together with their ground truth. After comparing our model to comparable cutting-edge approaches, we determined that our design resulted in at least a 10% improvement, showing that it generates more efficient, better, and robust results.

Abstract: A single-image-super-resolution (SISR) is the process of converting a single low-quality (LR) image to a high-quality (HR) image. This technology is utilised in a variety of industries, including medical and satellite imaging, to retrieve quality and required information from blurred or overexposed photos. Because of the lack of ability to extract important data and images due to poor quality surveillance photographs, this method can be utilised in the field of surveillance to produce high-quality images. We'd like to use General Adversarial Networks to handle low-quality photos because existing methods have resulted in slightly fuzzy and greasy images that look like oil paintings (GAN). We'd like to introduce Super Resolution General Adversarial Networks in particular (SRGAN). This method employs perceptual losses. In this case, PSNR, MSE, and SSIM values are shown to be superior to those obtained by standard approaches in this case. The SRGAN-processed photos are of excellent quality, allowing the images to be seen through hazy and misty areas.

Abstract: The rapid growth in Covid-19 cases increases the burden on health care services all over the world. Hence, a quicker and accurate diagnosis of this disease is essential in this situation. To get quick and accurate results, X-ray images are commonly used. Deep Learning (DL) techniques have reached a high position since they provide accurate results for medical imaging applications and regression problems. However the pre-processing methods are not successful in eliminating the impulse noises and the feature extraction technique involving filtering methods did not yield good filter response. In this paper, Covid-19 X-ray images were classified using the Fuzzy Gabor filter and Deep Convolutional Neural Network (DCNN). Initially the Chest X-ray images are pre-processed using Median Filters. After pre-processing, a Fuzzy Gabor filter is applied for feature extraction. Local vector features were first extracted from the given image using the Gabor filter, taking these vectors as observations. The orientation and wavelengths of the Gabor filter were fuzzified to improve the filter response. The extracted features are then trained and classified using the DCNN algorithm. It classifies the chest X-ray images into three categories that includes Covid-19, Pneumonia and normal. Experimental results have shown that the proposed Fuzzy Gabor-CNN algorithm attains highest accuracy, Precision, Recall and F1-score when compared to existing feature extraction and classification techniques.

Abstract: The hasty spread of the perilous coronavirus has resulted in a significant loss of human life and unprecedented public health challenges around the world. Early screening of COVID-19 followed by an immediate medical response can halt the spread of the infection. The deep learning algorithms coupled with chest X-ray images provide fast and accurate results. This study aims to fine-tune seven pre-trained models for COVID-19 detection using chest X-ray radiographs. A sample of 3428 chest X-ray images comprising 1626 COVID-19 images was used to train and validate the models. The Inception V3 model outperformed the other models with an accuracy of 99.42%.

Abstract: Detection of surface imperfections in planar surfaces in the modern assembling process is a continuous area of examination. It has generally been difficult to ensure the outright faultlessness of surfaces of carbon and iron combinations, or any surface so far as that is concerned. Nature of appearance, or rather, the outer layer of modern items like metal sheets is being held at an especially elevated expectation particularly of late, to satisfy these guidelines and to guarantee that client necessities are met, we use PC vision based arrangements, these arrangements basically involve utilizing a 2D or 3D deformity recognition procedure like for example edge identification, these calculations joined with an AI model will recognize and distinguish absconds on the different planar surfaces. The model will be prepared with a dataset of pictures that will contain harmed as well as unharmed surfaces, this is to ensure that the model can distinguish a deformity and without any imperfection the model ought to perceive the surface as homogeneous. However long the model is prepared appropriately, these techniques have been demonstrated to be particularly vigorous, and as an answer they have been common of late and throughout the course of recent years. CNN and brain networks in view of CNN designs were utilized to prepare the model to group the imperfections.

Abstract: In light of the growing number of vehicles, automated license plate recognition (ALPR) systems are much needed. The ALPR system is a widely used technology for various vehicle management processes such as law enforcement, surveillance, toll booth operations, parking lots, etc. We propose a license plate recognition system, where a neural network concept is applied. This system includes image pre-processing which helps to quickly and easily locate, segment and recognize the license plate characters, so image pre-processing is one of the important factors that affect total system performance. As we are performing the character segmentation of the license plate, the accuracy of the character recognition increases. In India, license plates are not only different in shape and size but also have different colours according to the registration or license number in India. There are 8 types of license plates in total issued by the RTO. In this effort, we identify the type of license plate by detecting the colour of the license plate. Thus vehicle registration types are recognized from the colour of the license plate detected.