Abstract Facial emotion recognition (FER) is a significant topic of research in computer vision, yet it is quite challenging to identify facial emotions in a complex environment. This study delivers a comparison of whether the convolutional neural network (CNN) architectural model functions better when given solely the facial landmarks for training or when given extra information, such as raw pixels of facial images. To address this, two FER approaches have been initially investigated, involving (i) a Davis Library based deep convolution network (Dlib-DCN) model to explore the impact of employing facial landmarks on an aliased deep convolution network (DCN) model, and (ii) an anti-aliased DCN (A-DCN) model to improve the aliasing problems. Then, an innovative hybrid DA-DCN approach that is based on facial landmarks extracted from facial data has been proposed. These models have been implemented, evaluated on three widely used FER datasets, and then compared with each other to detect eight distinct emotions from image data, including happiness, neutral, disgust, contempt, fear, sadness, anger, and surprise. Furthermore, to measure the proposed method’s quality and efficiency, numerous performance metrics have been evaluated, including accuracy, specificity, sensitivity, Jaccard Coefficient, and training time. The experimental results demonstrate that the anti-aliased facial landmark-based approach (DA-DCN) significantly outperforms both the anti-aliased pixel-based (A-DCN) approach and the aliased facial landmark-based (Dlib-DCN) approach in terms of precision and reliability while reducing the dimensionality of the input data. The suggested DA-DCN model achieves an overall accuracy of 99.3% on the Extended Cohn-Kanade (CK +) datasets, 98.12% on the Japanese female facial expressions (JAFFEE), and 84.44% on the Real-world Affective Face (RAF) dataset, one of the most difficult FER datasets.

Abstract This paper proposes a novel convolutional neural network (CNN) architecture designed for semantic segmentation in remote sensing images. The proposed W13 Net model addresses the inherent challenges of segmentation tasks through a carefully crafted architecture, combining the strengths of multistage encoding–decoding, skip connections, combined weighted output, and concatenation techniques. Compared with different segmentation models, the suggested model performs better. A comprehensive analysis of different segmentation models has been carried out, resulting in an extensive comparison between the proposed W13 Net and five existing state-of-the-art segmentation architectures. Utilizing two standardized datasets, the Dense Labeling Remote Sensing Dataset Termed (DLRSD), and the Mohammad Bin Rashid Space Center (MBRSC) Dubai Aerial Imagery Dataset, the evaluation entails training, testing, and validation across different classes. The W13 Net demonstrates adaptability, generalization capabilities, and superior results in key metrics, all while displaying robustness across a variety of datasets. A number of metrics, including accuracy, precision, recall, F1 score, and IOU, were used to evaluate the system’s performance. According to the experimental results, the W13 Net model obtained an accuracy of 87.8%, precision of 0.88, recall of 0.88, F1 score of 0.88, and IOU of 0.74. The suggested model showed a significant improvement in segmentation IOU, with an increase of up to 18%, when compared to other with the recent segmentation models taking into consideration the model’s comparatively low number of parameter (2.2 million) in comparison with the recent models.

Abstract This article proposes robust features fusion methodology for supervised palmprint recognition. The process of features fusion has been formulated according to the hybridization between robust morphological features of principal lines and the following features extractors: principal component analysis (PCA), transformed domain extractors, and invariant moments. The region of interest (ROI) has been extracted based on the identification of the valley points locations. Moreover, both morphological operations and edge detection have been applied in order to formulate the correct ROI. First, both length and slope of each extracted principal line have been estimated morphologically. Second, different features extraction techniques have been applied for the resultant ROI including (PCA), transformed domain, and invariant moments. Finally, seven supervised machine learning methods have been manipulated in order to achieve the highest palmprint recognition accuracy. The accuracy of the recognition was evaluated by measuring sensitivity, dice, precision, Jaccard coefficients, correlation, accuracy, and recognition time. Experimental findings show that, among all tested machine learning methods, the feed forward neural network with back propagation based on features vector of the fusion between invariant moments and principal lines morphology has achieved approximately 99.9% of successful palmprints recognition.

Abstract Human activity recognition has become an expansive field of interest in recent years, both in academic and industrial research. Human Activity recognition (HAR) is concerning the prediction of person's movement or action such as walking, standing, sitting, up and downstairs, etc. Convolutional neural network (CNN) is a key component with deep learning. The main objective of this study is to design and implement an activity recognition algorithm in the state of the art of deep learning systems which accomplish superlative performance to detect human activities. Accordingly, an extensive comparison has been developed between different deep learning algorithms such as classical (CNN) models and Recurrent Neural Network (RNN) models with respect to the major human activities. Furthermore, a robust (CNN) deep learning model has been built up and proposed in order to enhance the recognition precision of human activities. This proposed model uses raw data acquired from a set of inertial sensor and exploring numerous human achas been built up and proposedtivities; sitting, standing, jogging, walking, and etc. Experimental results show that the precision of the proposed deep structure has achieved 97.5% with respect to the NAdam optimizer which would be considered as the most effectively recognizer compared to other deep learning architectures.

Abstract A multiclass 3D object recognition has perceived a numerous evolution with respect to both accuracy and speed. This study introduces the implementation of modern YOLO algorithms (YOLOv3, YOLOv4, and YOLOv5) for multiclass 3D object detection and recognition. All YOLO algorithms have been tested according to a very large scaled dataset (Pascal VOC dataset). Performance evaluation has targeted the calculation of the following metrics; mAP (mean average precision), recall, F1‐score, IOU (intersection over union), and the running time. Experimental results demonstrate that the YOLOv3 has targeted mAP of 77%, IOU of 0.41 and the total running time was almost 8 h. Moreover, in YOLOv4, it has targeted mAP of 55%, IOU of 0.035 and the total running time nearly 7 h. In addition, YOLOv5 has established the mAP of 48%, IOU of 0.045, and the total running time was about 3 h. Finally, a modified version of YOLOv5 has been proposed in the state‐of‐the‐art of optimizing its hyperparameters and layering system. Accordingly, the mAP scored about 55% with 3 h running time. The final conclusions of this study have demonstrated that YOLOv3 has scored the highest recognition accuracy, however, the proposed modified YOLOv5 has scored the least processing time.