Sign language makes a significant contribution to enabling communication for speech-impaired individuals across the globe. It needs a physical interpreter to facilitate communication; however, these interpreters are limited in number and may not always be available when needed. Deep learning techniques can help address this challenge by creating a virtual interpreter. However, a few notable challenges such as occlusion, external lighting variations, and background subtraction are to be handled in the recognition system. Gestures performed in Indian Sign Language (ISL) involve movements of a single hand or both hands to convey signs for communication. This article proposes a Static Gesture Categorization and Interpretation of Indian Sign Language (SGCIISLang) model that employs an optimized Gated Recurrent Unit (GRU) architecture. For the experiment, we developed our own dataset titled “Static gestures of Indian Sign Language (ISL) for English Alphabet, Hindi Vowels and Numerals” which has already been published on Mendeley Data (https://data.mendeley.com/datasets/7tsw22y96w/1). The MediaPipe library is used for feature extraction, and the results are integrated into our model to classify sign motions. The region of interest is extracted using three MediaPipe approaches: holistic, holistic without pose, and holistic without face. We analyze and compare four models: Long Short Term Memory-Convolutional Neural Network (LSTM-CNN), Convolutional Neural Network (CNN), Recurrent Neural Network (RNN) and Gated Recurrent Unit (GRU) based SGCIISLang model. Based on the experimental results, the GRU-based SGCIISLang model achieves greater efficiency, faster processing and rapid convergence compared to CNN, LSTM-CNN and RNN models. The prediction precision achieved is 97.96% with loss, recall, mean square error (MSE) and F1 score values of 0.0837, 0.9742, 2.791 and 0.977, respectively. Our approach outperforms and addresses challenges. The sample prototype of our proposed approach is available on GitHub (https://github.com/AnimeshSingh777/Sample-Prototype-for-Indian-Sign-Language-Static-Gesture-Recognition-System).

Increase in Chest X-ray (CXR) imaging tests has burdened radiologists, thereby posing significant challenges in writing radiological reports on time. Although several deep learning-based automatic report generation methods have been developed, most are over-parameterized. For deployment on edge devices with constrained processing power or limited resources, over-parameterized models are often too large. This article presents a compressed deep learning-based model that is 30% space efficient compared to the non-compressed base model, while both have comparable performance. The model comprising VGG19 and hierarchical long short-term memory equipped with a contextual word embedding layer is used as the base model. The redundant weight parameters are removed from the base model using unstructured one-shot pruning. To overcome the performance degradation, the lightweight pruned model is fine-tuned over publicly available OpenI dataset. The quantitative evaluation metric scores demonstrate that proposed model surpasses the performance of state-of-the-art models. Additionally, the proposed model, being 30% space efficient, is easily deployable in resource-limited settings. Thus, this study serves as baseline for development of compressed models to generate radiological reports from CXR images.

In this article, the authors present an effort to recognize handwritten Gurumukhi place names for use in postal automation. Five feature extraction techniques (zoning, horizontal peak extent, vertical peak extent, diagonal, and centroid) have been analyzed and optimized using Principal Component Analysis (PCA). Four classification methods ( k -Nearest Neighbor ( k -NN), decision tree, random forest, and Convolutional Neural Network (CNN)) have been utilized to classify the handwritten word images. To enhance the recognition results, the authors have employed Bootstrap Aggregation (Bagging) with a majority voting scheme. The authors used a public benchmark dataset of 40,000 handwritten place-name samples in the Punjabi language for their experimental work. The experiments were conducted using a 70:30 partitioning approach, where 70% of the data was utilized for training and the remaining 30% for testing. The system achieved a maximum recognition accuracy of 96.98% by utilizing a combination of zoning, vertical peak extent, and diagonal features, and a minimum Mean Squared Error (MSE) of 0.86% based on a combination of zoning and horizontal peak extent features with a majority voting scheme through ensemble (Bagging) methodology.

Nitrogen (N) is one of the key nutrient element needed for optimum crop growth and production. Deficiency of N leads to a decrease in crop production and excess results in poor root growth and leaching into groundwater thereby causing environmental issues. Hence the optimum application of N is needed which is possible by exactly estimating the available quantities of N in the plant. In this study, an attempt has been made to estimate N in Soybean leaves using the hyperspectral and simulated Sentinel-2 observations. Spectral observations of fifteen soybean leaf samples were collected using the EKO MS-720 Spectroradiometer. The instrument operates in the spectral range of 350–1050 nm. and collects data in contiguous 213 bands. Support Vector Regression-based models were evaluated using three feature selection methods, 1) individual hyperspectral bands, 2) Normalized band ratio's and 3) simulated Sentinel-2 bands and indices. Model performance was evaluated using R2. Analysis carried out using the individual hyperspectral bands showed that bands from the red and red-edge region are performing best with R2between 0.872 and 0.876. However, NBR's estimated from band combinations in the red-edge region are performing best with R2between 0.938 - 0.956. Further, we identified a subset of wavelengths to simulate Sentinel-2 spectral bands, results showed that red-edge and narrow NIR bands provide the highest R2between 0.878 and 0.893. We observed that indices such as Canopy Chlorophyll Content Index (CCCI) and Chlorophyll Index Red Edge (CIRE) are performing better for N estimation with R2of 0.946, 0.923, respectively. Based on the observations we can conclude that red, red-edge and narrow NIR region is useful for Soybean N estimation.

Image denoising is a thoroughly studied research problem in the areas of image processing and computer vision. In this work, a deep convolution neural network with added benefits of residual learning for image denoising is proposed. The network is composed of convolution layers and ResNet blocks along with rectified linear unit activation functions. The network is capable of learning end‐to‐end mappings from noise distorted images to restored cleaner versions. The deeper networks tend to be challenging to train and often are posed with the problem of vanishing gradients. The residual learning and orthogonal kernel initialisation keep the gradients in check. The skip connections in the ResNet blocks pass on the learned abstractions further down the network in the forward pass, thus achieving better results. With a single model, one can tackle different levels of Gaussian noise efficiently. The experiments conducted on the benchmark datasets prove that the proposed model obtains a significant improvement in structural similarity index than the previously existing state‐of‐the‐art techniques.

Lung field defines a region‐of‐interest in which specific radiologic signs such as septal lines, pulmonary opacities, cavities, consolidations, and lung nodules are searched by a chest radiographic computer‐aided diagnostic system. Thus, its precise segmentation is extremely important. To precisely segment it, numerous methods have been developed during the last four decades. However, no exclusive survey consolidating the advancements in these methods has been presented till date, thus indicating a void and the need. This study fills the void by presenting a comprehensive survey of these methods with a focus on their underlying principle, the dataset used, reported performance, and relative merits and demerits. It refrains from doing a hard comparative evaluation by bringing all of them on a common platform, since the datasets used in their development and testing are of varied quality, complexity, and are not publicly available. It also provides a glimpse of deep learning, the present state of deep‐learning‐based lung field segmentation methods, expectations from it, and the challenges ahead of it.

Automated obstacle detection is required in a number of applications. It is a challenging task because there is no prior knowledge about the appearance of ground, or the locations and appearance of obstacles in the scene. In this paper, an obstacle detection system based on matching the color edge features in stereo images is presented. The paper also presents an enhanced adaptive local-cross based stereo matching method used in the system. The system has been extensively evaluated. The result of evaluation allows the usage of system for automatic detection of moderately sized obstacles within a distance range of 2-15 meters with sufficient accuracy.