Journal of King Saud University-Computer and Information Sciences最新文献

Underwater Wireless Sensor Networks (UWSNs) are essential for a number of environmental and oceanographic monitoring applications. However, they face different and more complex challenges than terrestrial wireless sensor networks (TWSNs). The main challenges faced by UWSNs are limited include high propagation delays, poor bandwidth, low throughput, and high energy consumption. Replacing sensor batteries in such networks becomes extremely difficult as they are usually deployed in remote areas where limited human interaction is possible. The unbalanced and inefficient usage of energy by various network nodes poses another issue, as it may reduce the applicability and feasibility of the network. Therefore, proposing Energy-Efficient Routing Protocols (E-ER-Ps) is crucial to improve the performance and lifespan of these networks. Due to the challenges mentioned earlier, this research presents an extensive analysis of several different E-ER-Ps intended for UWSNs. We compare contemporary approaches that use machine learning (ML) with conventional protocols, as ML-based approaches have shown significant potential in resolving the intricate challenges faced by UWSNs. This paper aims to present a critical review of different E-ER-Ps from various prospects for UWSNs. To better comprehend the structure and uses of these protocols, we provide an innovative taxonomy for their classification. While ML-based protocols are evaluated for their flexibility, predictive power, and overall efficiency advancements, traditional protocols are evaluated based on their routing tactics and energy-efficiency improvements. A thorough comparative analysis highlights the advantages, disadvantages, and possible uses for different protocols. Furthermore, a critical analysis of ML’s function, incorporating intelligent and adaptive routing approaches, is presented, highlighting the technology’s potential to completely alter UWSN management. To formulate and implement E-ER-Ps for UWSNs, the article concludes by highlighting the present obstacles, including the need for real-time flexibility, resilience to environmental alters, and interaction with pre-existing network infrastructures. The development of ML-based approaches, hybrid approaches that combine conventional and ML-based methodologies, and the design of protocols that can adapt dynamically to the changing circumstances of underwater habitats are highlighted as future research objectives. This research provides the foundation for future advancements in this crucial field by presenting a comprehensive overview of the state-of-the-art UWSN E-ER-Ps.

In order to minimize additional computational costs in detecting forged videos, and enhance detection accuracy, this paper employs dynamic facial expression sequences as key sequences, replacing original video sequences as inputs for the detection model. A spatio-temporal dual-branch detection network is designed based on the visual Transformer architecture. Specifically, this process involves three steps. Firstly, dynamic facial expression sequences are localized as key sequences using optical flow difference algorithms. Subsequently, the spatial branch network employs the focal self-attention mechanism to focus on dynamic features of expression-relevant regions and uses Factorization Machines to facilitate feature interaction among multiple key sequences. Meanwhile, the temporal branch network concentrates on learning the temporal inconsistency of optical flow differences between adjacent frames. Finally, a binary classification linear SVM combines the Softmax values from the two branch networks to provide the ultimate detection outcome. Experimental results on the Faceforensics++ dataset demonstrate: (a) replacing whole video sequences with facial expression key sequences effectively reduces training and detection time by nearly 80% and 90%, respectively; (b) compared to state-of-the-art methods involving random sequence/frame extraction and key frame extraction based on video compression techniques, the proposed approach in this paper presents a more competitive detection accuracy.

In citrus cultivation, Anthracnose, Scab, and Greasy Spot significantly impact yield and quality. Facing challenges in detecting small targets against complex orchard backgrounds with uneven lighting and obstructions, existing models suffer from low detection accuracy. This study introduces the YOLOv8n-CDDA citrus leaf disease detection model. The Cross-Domain Dynamic Attention (CDDA) mechanism deconstructs the backbone network’s input feature maps into sections, dynamically assigning spatial and channel attention weights to reconstruct critical information and capture the variations and weak semantic features of disease textures. The proposed Adaptive Random Mix-Cut Splicing (ARMS) image augmentation technique blends diseased leaf images with healthy citrus backgrounds, enhancing the diversity and number of background targets. To reduce computational and memory consumption, the network is streamlined through channel pruning; to compensate for the loss in accuracy from pruning, a teacher–assistant–student network format is used for knowledge distillation, where the student network learns from soft knowledge to improve disease recognition accuracy. Finally, Grad-CAM++ technology generates heatmaps of the detections, facilitating the visualization of effective features and deepening understanding of the model’s focus areas. Experimental results demonstrate that the YOLOv8n-CDDA model achieves an average accuracy of 90.89% in disease detection, with an average recall rate of 81.12%, and a mean Average Precision (mAP50) of 88.36%. Compared to the original YOLO v8n and current mainstream detection models such as YOLOv5s, SSD, and Faster-RCNN, the improvements in average accuracy are respectively 2.95%, 4.78%, 14.22%, and 21.01%; in average recall, 2.36%, 3.09%, 15.74%, and 23.27%; and in mAP50, 2.38%, 3.13%, 13.45%, and 20.91%. After pruning and distillation for lightweight adaptation, the YOLOv8n-CDDA model has a parameter size of 0.8M, requires 4.2 GFLOPs, weighs 2.0 MB, and operates at 45 fps. Compared to YOLOv8n, this represents a reduction of 2.2M in parameters, 3.9 GFLOPs, and 4 MB in model weight, with an increase of 7 fps in speed. This model exhibits exceptional performance in the complex environment of citrus leaf disease detection, providing robust technical support for citrus growth monitoring studies, and offering insights for disease detection in other crops as well.

The widespread integration of software into all parts of our lives has led to the need for software of higher reliability. Ensuring reliable software usually necessitates some form of formal methods in the early stages of the development process which requires strenuous effort. Hence, researchers in the field of software reliability introduced Software Reliability Growth Models (SRGMs) as a relatively inexpensive approach to software reliability prediction. Conventional parameter estimation methods of SRGMs were ineffective and left more to be desired. Consequently, researchers sought out swarm intelligence to combat its flaws, resulting in significant improvements. While similar surveys exist within the domain, the surveys are broader in scope and do not cover many swarm intelligence algorithms. Moreover, the broader scope has resulted in the occasional omission of information regarding the design for reliability predictions. A more comprehensive survey containing 38 studies and 18 different swarm intelligence algorithms in the domain is presented. Each design proposed by the studies was systematically analyzed where relevant information including the measures used, datasets used, SRGMs used, and the effectiveness of each proposed design, were extracted and organized into tables and taxonomies to be able to identify the current trends within the domain. Some notable findings include the distance-based approach providing a high prediction accuracy and an increasing trend in hybridized variants of swarm intelligence algorithms designs to predict software reliability. Future researchers are encouraged to include Mean Square Error (MSE) or Root MSE as the measures offer the largest sample size for comparison.

The efficient and timely execution of tasks is a fundamental challenge in the realm of future Internet of Things (IoT) networks. To address this challenge, fog devices are often deployed close to end devices to facilitate task processing on behalf of IoT nodes. One strategy for improving task computational delay is to employ parallel task offloading, in which tasks are subdivided into subtasks and sent to different fog devices for execution in parallel. However, allocating computational resources to fog nodes and mapping these resources to IoT subtasks is a key challenge in this area. This work models the parallel task offloading problem as a graph-matching problem and utilizes a many-to-many matching technique to achieve a stable mapping of IoT subtasks to fog node resources. Unfortunately, the proposed solution is subject to the problem of externalities due to the dynamic preference profiling of fog nodes. To address this issue, we employ an iterative algorithm to resolve any blocking pairs that may arise. Our results demonstrate that the proposed technique reduces the task latency by 29% as compared to other matching-based techniques available in the literature.

Accurate diagnosis of kidney disease is crucial, as it is a significant health concern that demands precise identification for effective and appropriate treatment. Deep learning methods are increasingly recognized as valuable tools for disease diagnosis in the biomedical field. However, current models utilizing deep networks often encounter challenges of overfitting and low accuracy, necessitating further refinement for optimal performance. To overcome these challenges, this paper proposes the introduction of two ensemble models designed for kidney stone detection in CT images. The first model, called StackedEnsembleNet, is a two-level deep stack ensemble model that effectively integrates the predictions from four base models: InceptionV3, InceptionResNetV2, MobileNet, and Xception. By leveraging the collective knowledge of these models, StackedEnsembleNet improves the accuracy and reliability of kidney stone detection. The second model PSOWeightedAvgNet, leverages the Particle Swarm Optimization (PSO) algorithm to determine the optimal weights for the weighted average ensemble. Through PSO, this ensemble approach assigns optimized weights to each model during the ensembling process, effectively enhancing the performance by optimizing the combination of their predictions. Experimental results conducted on a large dataset of 1799 CT images demonstrate that both StackedEnsembleNet and PSOWeightedAvgNet outperform the individual base models, achieving high accuracy rates in kidney stone detection. Furthermore, additional experiments on an unseen dataset validate the models’ ability to generalize. The comparison with previous methods confirms the superior performance of the proposed ensemble models. The paper also presents Grad-CAM visualizations and error case analysis to provide insights into the decision-making processes of the models. By overcoming the limitations of existing deep learning models, StackedEnsembleNet and PSOWeightedAvgNet offer a promising approach for accurate kidney stone detection, contributing to improved diagnosis and treatment outcomes in the field of nephrology.

Image captioning, the task of generating descriptive sentences for images, has seen significant advancements by incorporating semantic information. However, previous studies employed semantic attribute detectors to extract predetermined attributes consistently applied at every time step, resulting in the use of irrelevant attributes to the linguistic context during words’ generation. Furthermore, the integration between semantic attributes and visual representations in previous works is considered superficial and ineffective, leading to the neglection of the rich visual-semantic connections affecting the captioning model’s performance. To address the limitations of previous models, we introduced a novel framework that adapts attribute usage based on contextual relevance and effectively utilizes the similarities between visual features and semantic attributes. Our framework includes an Attribute Detection Component (ADC) that predicts relevant attributes using visual features and attribute embeddings. The Attribute Prediction and Visual Weighting module (APVW) then dynamically adjusts these attributes and generates weights to refine the visual context vector, enhancing semantic alignment. Our approach demonstrated an average improvement of 3.30% in BLEU@1 and 5.24% in CIDEr on MS-COCO, and 6.55% in BLEU@1 and 25.72% in CIDEr on Flickr30K, during CIDEr optimization phase.

Corn diseases significantly impact crop yields, posing a major challenge to agricultural productivity. Early and accurate detection of these diseases is crucial for effective management and mitigation. Existing methods, mostly relying on analyzing corn leaves, often lack the precision to identify and classify a wide range of diseases under varying conditions. This study introduces a novel approach to detecting corn diseases using image processing and deep learning techniques, aiming to enhance detection accuracy through pre-processing, improved feature extraction and selection, and classification algorithms. A new deep Convolutional Neural Network (CNN) model named TreeNet, with 35 layers and 38 connections, is proposed. TreeNet is pre-trained using the Plant Village imaging dataset. For image pre-processing, the YCbCr color space is utilized to improve color representation and contrast. Feature extraction is performed using TreeNet and two pre-trained models, Darknet-53, and DenseNet-201, with features fused using a serial-based fusion method. The Entropy-coded Sine Cosine Algorithm is applied for feature selection, optimizing the feature set for classification. The selected features are used to train Support Vector Machine (SVM) and K-Nearest Neighbor (KNN) classifiers, with extensive experiments conducted using both 5-fold and 10-fold cross-validation, and feature sizes ranging from 200 to 1150. The proposed method achieves classification accuracy, precision, recall, and F1-score of 99.8%, 99%, 100%, and 99%, respectively, surpassing existing benchmarks. The integration of TreeNet with Darknet-53 and DenseNet-201, along with robust pre-processing and feature selection, significantly improves corn disease detection, highlighting the potential of advanced CNN architectures in agriculture.

Learning multi-scale feature representations is essential for medical image segmentation. Most existing frameworks are based on U-shape architecture in which the high-resolution representation is recovered progressively by connecting different levels of the decoder with the low-resolution representation from the encoder. However, intrinsic defects in complementary feature fusion inhibit the U-shape from aggregating efficient global and discriminative features along object boundaries. While Transformer can help model the global features, their computation complexity limits the application in real-time medical scenarios. To address these issues, we propose a Cross-scale Fusion Network (CFNet), combining a cross-scale attention module and pyramidal module to fuse multi-stage/global context information. Specifically, we first utilize large kernel convolution to design the basic building block capable of extracting global and local information. Then, we propose a Bidirectional Atrous Spatial Pyramid Pooling (BiASPP), which employs atrous convolution in the bidirectional paths to capture various shapes and sizes of brain tumors. Furthermore, we introduce a cross-stage attention mechanism to reduce redundant information when merging features from two stages with different semantics. Extensive evaluation was performed on five medical image segmentation datasets: a 3D volumetric dataset, namely Brats benchmarks. CFNet-L achieves 85.74% and 90.98% dice score for Enhanced Tumor and Whole Tumor on Brats2018, respectively. Furthermore, our largest model CFNet-L outperformed other methods on 2D medical image. It achieved 71.95%, 82.79%, and 80.79% SE for STARE, DRIVE, and CHASEDB1, respectively. The code will be available at https://github.com/aminabenabid/CFNet

The ECG signal is often accompanied by noise, which can affect its shape characteristics, so it is important to perform signal de-noising. However, the commonly used signal noise reduction methods, such as wavelet or filter transformation, often prioritize high-frequency signals over low-frequency ones, leading to the loss of low-frequency band features or difficulties in capturing them. We propose a fusion reconstruction framework that combines hash autoencoder and margin semantic reinforcement to enhance low-frequency band features. Specifically, for labeled samples, margin semantic reinforcement identifies and corrects weight discrepancies among bands with similar waveforms but different labels to amplify the low-frequency signals associated with the label and reduce irrelevant ones. Meanwhile, hash autoencoder utilizes a semantic hash dictionary to reconstruct the original signal and mitigate noise pollution. For unlabeled samples, the hash autoencoder is utilized to generate pseudo-labels, followed by the reproduction of the aforementioned enhanced reconstruction process. The final step involves weighting the two types of signals, enhanced with margin semantics and hash autoencoder reconstruction, to achieve the reconstruction objective of the original signal, facilitating recognition and detection tasks. Experiments conducted on different classical classifiers demonstrate that the reconstructed ECG signals can significantly improve their performance.