Human action recognition has been a talked topic since machine vision was coined. With the advent of neural networks and deep learning methods, various architectures were suggested to address the problems within a context. Convolutional neural network has been the primary go-to architecture for image segmentation, flow estimation and action recognition in recent days. As the problem itself is an extended version of various subproblems, such as frame segmentation, spatial and temporal feature extraction, motion modeling and action classification as a whole, some methods reviewed in this paper addressed sub-problems and some tried to address a single architecture to the action recognition problem. While being a success, convolution neural networks have drawbacks in its pooling methods. CapsNet, on the other hand, uses squashing function to determine the activation. Also it addresses spatiotemporal information with the normalized vector maps while CNN-based methods extracts feature map for spatial and temporal information and later augment them in a fusion layer for combining two separate feature maps. Critical review of papers provided in this work can contribute significantly in addressing human action recognition problem as a whole.

Crop root development, and in turn crop growth, is strongly influenced by soil strength and the mechanical impedance of compacted layers, which restrict root elongation and exploration. Because the depth and thickness of compacted layers vary across a field, their identification is essential for site-specific tillage and sustainable root-zone management. A sensing approach that can support future real-time identification of compacted layers after soil-specific calibration, which would enable variable-depth tillage, reducing mechanical impedance and improving energy-use efficiency while maintaining crop yields. This study aimed to develop and evaluate prediction models that can support future real-time identification of compacted soil layers using soil electrical conductivity (EC) and moisture content as non-destructive indicators. A sandy clay soil (48.6% sand, 29.3% clay, 22.1% silt) was tested in a soil-bin laboratory under controlled conditions at three moisture levels (13, 18, and 22% db.) and six depth layers (C1–C6, 0–30 cm) identified from the penetration-resistance profile to measure penetration resistance, shear resistance, and EC. Penetration and shear resistance increased toward the most resistant depth layer and decreased with increasing moisture content, whereas EC generally increased with both depth layer and moisture content. Linear regression models relating penetration resistance ($R^2=0.893$) and shear resistance ($R^2=0.782$) to EC and moisture content were developed and evaluated. Field validation in a paddy field of similar texture showed that predicted penetration resistance differed from measured values by 3–6% across the three compaction treatments evaluated. Root length density and root volume decreased with increasing machine-induced compaction, confirming the agronomic relevance of the modeled patterns and supporting the suitability of the proposed indicators. Together, these results demonstrate that EC and moisture content can potentially be used as non-destructive proxies for compacted-layer identification and provide a calibration basis for future on-the-go sensing systems to support site-specific, variable-depth tillage in agricultural fields.

Recent years have seen a rise in the use of various machine learning techniques in computer vision, particularly in posing feature-based human action recognition which includes convolutional neural networks (CNN) and recurrent neural network (RNN). CNN-based methods are useful in recognizing human actions for combined motions (i.e., standing up, hand shaking, walking). However, in case of uncertainty of camera motion, occlusion, and multiple people, CNN suppresses important feature information and is not efficient enough to recognize variations for human action. Besides, RNN with long short-term memory (LSTM) requires more computational power to retain memories to classify human actions. This research proposes an extended framework based on capsule network using silhouette pose features to recognize human actions. Proposed extended framework achieved high accuracy of 95.64% which is higher than previous research methodology. Extensive experimental validation of the proposed extended framework reveals efficiency which is expected to contribute significantly in action recognition research.

This study introduces a multimodal approach designed to detect anomaly in satellite imagery. The approach is comprised of three main components: the image processing models (ResNet and Regulated Network), the text processing models (BERT and a graph model), and a combination model that acts as a classifier, seamlessly integrating features extracted from both image and text models. Our work utilizes the SpaceNet8 Challenge dataset, focusing on regions in Louisiana USA, damaged by Hurricane Ida in 2021, and the Ahrweiler district in Germany, impacted by flooding in Western Europe during the same year. The experimental results demonstrate significant performance improvements when integrating image processing models and text processing models, surpassing the baseline CNN models. This study offers insights for future work in anomaly detection and semantic segmentation, emphasizing the effectiveness of a multimodal approach with heterogeneous data.

Convolutional neural networks have become ubiquitous in image classification tasks. The state-of-the-art models for image classifications use convolutional layers in one way or another. There is a need for deploying deep learning models, especially the real-time vision models, in the edge devices to get better latency. But deploying such models in edge devices are becoming critical as the networks are becoming deeper and more dense. An overparameterized network is not necessarily required in many of the use cases of such deployment. This led researcher to develop technique for optimizing smaller and shallower networks, network architecture search techniques, and deep learning model compression techniques. In this research, we proposed a framework that utilizes deep determinisitic policy gradient, a class of deep reinforcement learning algorithm, to the learn the best set of filters considering the intrinsic dimensionality of the dataset, feature of each layer and the criteria based on which the filters of a convolutional layer will be ranked. By learning this relationship, we can prune off unnecessary filters which will reduce both computational and memory requirement for the model without losing too much accuracy. Our method showed that the model can prune off 66% filters overall.