IJCSIT Journal Cover

International Journal of Computer Science and Information Technology

ISSN: 3005-9682 (Print)ISSN: 3005-7140 (Online) DOI: 10.62051/ijcsit Frequency: Monthy

International Journal of Computer Science and Information Technology (IJCSIT) is a peer‑reviewed, English‑language open‑access journal published by Warwick Evans Publishing. It provides an international forum for rigorous empirical and theoretical scholarship that addresses contemporary challenges in computer science and information technology — from fundamental theory to emerging applications.

Scope: The journal covers the full spectrum of computer science, software engineering, and information technology, including computer architecture and systems, artificial intelligence and pattern recognition, computer networks and distributed computing, computer graphics and multimedia, software systems, information technology and communications security, data management and data mining, computing and networking, theory and algorithms, and emerging interdisciplinary areas.

Indexing & Abstracting: Harvard Library, Crossref, ResearchGate, Scilit, Google, Mendeley, Semantic Scholar, etc.

Latest Articles

When Does Evidence Expansion Help? A Cross-Domain Study of Retrieval Paths and Lightweight Routing for Regulatory Evidence

Abstract: Regulatory question answering often fails because the decisive evidence is not contained in the retrieved leaf passage but in its heading, parent clause, table, or cited provision. We study whether common evidence-expansion stages improve retrieval before generation under a risk-sensitive criterion that requires complete supporting evidence and rejects materially misleading additions. In a prespecified pilot of 120 questions drawn from Chinese chemical-safety standards and English pharmaceutical-regulatory documents, we compare six retrieval paths built on a shared BM25 first stage, including reranking, structural-context expansion, relation expansion, and their combinations. Human annotators labeled 10,385 evidence occurrences for completeness and harmfulness. Structural context provided the clearest benefit: adding heading, parent, and table context improved combined success by 15.83 percentage points over BM25, whereas reranking and one-hop relation expansion yielded limited independent rescue. The most complete path did not outperform the cheaper context-only path because additional evidence often introduced harmful material. Lightweight learned routers outperformed a prespecified heuristic in grouped out-of-fold evaluation, but calibration-based abstention failed to achieve usable risk and coverage. These results support selective use of structural context and separate evaluation of completeness and harm, while indicating that reliable evidence-sufficiency abstention requires substantially more calibration data. Read More

Weld Defect Detection in TOFD Images Based on Improved YOLOv8

Abstract: Small defect indications, low contrast, and background noise make weld defect detection in time-of-flight diffraction (TOFD) images challenging. This study proposes SCW-YOLO, a TOFD weld defect detector based on YOLOv8n. The model adds a high-resolution P2 detection branch, uses space-to-depth convolution (SPDConv) in the backbone and neck downsampling paths, embeds coordinate attention (CoordAtt) in the neck, and uses Wise-IoU for bounding-box regression. On the validation split of an in-house dataset of 7,463 TOFD images, SCW-YOLO achieves 80.5% mean average precision at an intersection-over-union threshold of 0.5 (mAP50) and 77.7% precision, improvements of 3.0 and 7.4 percentage points over YOLOv8n, respectively. Among the detectors compared in this study, it also has the highest validation-set recall and runs at 86 frames per second (FPS) under the reported experimental setup. Quantitative performance on an independent test set and suitability for online industrial deployment remain to be established. Read More

Research on Evaluation Method of Intelligent Chassis Based on Large Model and Multi Condition Scenarios

Abstract: The intelligent chassis is the core carrier of the executive layer of electric vehicles, and its performance directly determines the overall handling stability, ride comfort, and driving safety of the vehicle. Traditional chassis evaluation often adopts a static indicator system with fixed weights, which is difficult to adapt to the differentiated demands of urban congestion, high-speed cruising, mountainous bends, low adhesion road surfaces, and other scenarios, and the adaptability of the scenarios is limited. Therefore, this article proposes an intelligent chassis evaluation method that integrates big language models and multi condition modeling. Research and establish a four-dimensional evaluation index system that includes handling stability, driving smoothness, braking performance, and line control response. Characterize multiple operating conditions with characteristic parameters, and solve the basic weights using the Analytic Hierarchy Process and Entropy Weight Method; Further utilize the large model to analyze the semantic features of working conditions, output adaptive dynamic weight correction quantities, and construct a subjective and objective dual drive weighting framework. Simulation verification shows that compared to fixed weight schemes, this method can more accurately and sensitively characterize the true performance level of the chassis under different operating conditions. Read More

Research on Bias Drift and Evaluation of Content Generated by Large Language Models in the Chinese Context

Abstract: Bias in content generated by large language models is not static; it may shift as models evolve, affecting the fairness and reliability of their outputs. Evaluating the presence and patterns of bias drift is therefore important for supporting model safety and fairness. This study examined models from different versions and models with different parameter sizes within the same model family. Using a Chinese-language bias evaluation dataset and metrics designed to assess bias drift, it analyzed changes in bias in model-generated content across versions and parameter sizes. The results showed that both model-version updates and changes in parameter count led to bias drift in generated content, and that the direction of drift varied. A single evaluation or assessments conducted only at fixed intervals may not detect bias drift promptly as large language models evolve. Version updates and increases in parameter count should therefore not, by themselves, be taken as evidence of improved model safety; ongoing bias evaluation and verification of alignment effectiveness are needed. Value alignment is not a one-time process, and its effectiveness cannot be assumed to persist over time. Alignment effectiveness should be assessed throughout model-version updates and changes in parameter count, with verification serving as a necessary checkpoint after a version update or an increase in parameter count. This study provides empirical evidence for evaluating and analyzing bias drift in content generated by large language models and may inform research on model alignment and the development of regulatory frameworks. Read More

Optimization Design and Simulation Analysis of a 2.4 GHz Microstrip Parallel-Coupled-Line Directional Coupler

Abstract: The 2.4 GHz ISM band is one of the principal operating frequency bands for short-range wireless communications and radio-frequency testing equipment, including Internet of Things (IoT), Bluetooth, and Wi-Fi systems. As a key passive component in RF front ends, the microstrip parallel-coupled-line directional coupler commonly involves a trade-off among coupling accuracy, port isolation, and device miniaturization. These performance indicators are difficult to optimize simultaneously using conventional structures, thereby limiting the further integration and miniaturization of portable wireless terminals. In this study, a low-cost FR4 substrate is employed to investigate the multi-objective coordinated optimization of a 2.4 GHz parallel-coupled-line directional coupler. Based on odd- and even-mode microwave transmission theory and four-port impedance-matching principles, theoretical parameter calculation, iterative ADS electromagnetic simulation, and comparative analysis of multiple design schemes are combined. A tapered coupling section and an isolation-port matching and absorption structure are introduced to improve the conventional quarter-wavelength coupling topology. The simulation results demonstrate that the optimized coupler achieves a coupling deviation within ±0.4 dB, an isolation of 22.7 dB, and an insertion loss as low as 0.78 dB. Its overall routing area is reduced by 17.2% compared with that of a conventional uniformly spaced coupled-line structure. All performance indicators satisfy the requirements for the mass production of commercial RF modules. This study establishes an integrated optimization procedure that considers electrical performance, physical dimensions, and PCB fabrication constraints, providing a theoretical basis and simulation reference for the miniaturized development of planar passive components operating in the same frequency band. Read More

Multi-Dimensional Perspectives on Intelligent Driving Research: Technological Evolution, Industrial Transformation, and Institutional Challenges

Abstract: Intelligent driving technology is at a critical turning point, transitioning from laboratory validation to large-scale commercial application. Its development involves not only technological breakthroughs but also deeper issues such as industrial chain restructuring, data governance, and legal and ethical considerations. Based on a systematic review of recent literature, this paper presents the research landscape across three dimensions: technological evolution, industrial transformation, and institutional governance. The findings reveal that: technologically, intelligent driving is undergoing a paradigm shift from modular rules to “end-to-end” data-driven approaches, with hardware energy efficiency, algorithm generalization capability, and long-tail scenario handling remaining core bottlenecks, while traditional control methods such as optimized fuzzy control still hold practical value in local obstacle avoidance scenarios; industrially, the industrial chain is trending toward networked restructuring, with the China-U.S. competitive landscape each possessing distinct advantages; institutionally, data governance faces the dual challenges of the failure of the “informed consent” model and national security risks, and the legal liability framework urgently needs updating. Based on a systematic integration of existing research, this paper identifies research gaps and proposes directions for future inquiry. The findings may serve as a reference for continued technological breakthroughs and institutional development in intelligent driving. Read More

MSFA-Net: A Low-Light Image Enhancement Method Based on Multi-Scale Feature Fusion and Attention Mechanism

Abstract: Low-light images are degraded by uneven illumination, sensor noise, color distortion, and weak texture boundaries. To address insufficient detail recovery, noise amplification, and inadequate coordination between global illumination and local texture, this paper proposes MSFA-Net, a compact residual enhancement network based on multi-scale feature fusion and hybrid channel–spatial attention. Each MSFA block extracts complementary information through standard 3×3, dilated 3×3, and 5×5 convolutional branches. A 1×1 fusion layer integrates local edges and large-receptive-field context, after which channel attention recalibrates global feature responses and spatial attention emphasizes informative underexposed regions. A learnable residual gate stabilizes optimization and limits over-enhancement. The network is trained with a composite objective combining pixel, structural, gradient, color-consistency, and total-variation losses. With four MSFA blocks and 32 feature channels, the baseline configuration contains approximately 0.192 million trainable parameters. A reproducible evaluation framework is defined on LOL and LOL-v2 using PSNR, SSIM, LPIPS, NIQE, edge preservation, noise statistics, and computational complexity. The proposed formulation focuses on a single algorithmic problem and provides an explicit mechanism for the coordinated improvement of illumination, texture detail, and visual fidelity. Read More

Energy-Constrained Tensile Damage Modelling and Parameter Identifiability of Recycled Aggregate Shotcrete

Abstract: A reliable constitutive description of recycled aggregate shotcrete must distinguish tensile resistance, deformation capacity, and post-peak energy dissipation. This study develops a reproducible numerical framework based on eleven reported steel-fiber-reinforced shotcrete mixtures and uses eighteen independent impact-test summaries to assess the limits of transferability. A four-parameter additive tensile-strength model is benchmarked against grade-only, logarithmic, and quadratic alternatives by leave-one-out cross-validation. An energy-constrained tensile damage law is subsequently reconstructed from peak stress, peak strain, the one-third-peak secant modulus, and finite-interval tensile work. The ascending branch satisfies the reported secant definition exactly, whereas the descending branch satisfies an explicit work constraint. The additive model achieves a calibration coefficient of determination of 0.9892 and a leave-one-out root-mean-square error of 0.0861 MPa. Within the evaluated mixtures, increasing recycled aggregate replacement from 0% to 100% reduces tensile strength by 15.4%, while increasing steel-fiber content from 0% to 2% at 50% replacement increases strength by 120%. An arithmetic consistency audit identifies a compression-denominator mismatch, an approximately twofold difference in toughness normalization, and incompatible entries in the independent impact dataset. The proposed formulation delineates an admissible family of energy-equivalent softening responses and quantifies the non-identifiability of residual strength from summary measurements alone. The findings provide a transparent basis for constitutive parameter screening, with predictions explicitly confined to the available evidence rather than interpreted as new laboratory measurements or validated thermal–dynamic responses. Read More

Deep Learning-Based Optimization Design for Vectorial Holographic Metasurfaces

Abstract: Metasurface-based vectorial holography, featuring high information capacity and device integration advantages, represents an important research direction in the field of light field manipulation. Traditional design methods face challenges such as low efficiency and limited degrees of freedom in polarization control, making it difficult to meet practical application requirements. This paper proposes a generative adversarial network-based optimization design method for vectorial holographic metasurfaces. An attention-enhanced network architecture is constructed to achieve end-to-end mapping from target light fields to metasurface structures. A channel attention mechanism is introduced to improve polarization feature learning capability, and a partitioned weighted loss function is designed to balance image quality and noise suppression. Experimental results show that while maintaining polarization fidelity above 0.95, the proposed method improves design efficiency by more than two orders of magnitude compared to traditional iterative algorithms, with holographic reconstruction efficiency reaching over 85%. This research provides an effective approach for the rapid design of high-performance vectorial holographic metasurfaces. Read More

Vision-Based Safety Early Warning System for Human–Robot Collaboration

Abstract: With the large-scale popularization of intelligent industrial robots, human–robot collaboration (HRC) has become the core production mode of flexible manufacturing workshops, which breaks the isolation barrier between humans and industrial equipment and significantly improves production efficiency. However, frequent cross-space coexistence between operators and robotic arms brings prominent collision risks, and traditional hardware isolation fences and fixed safety sensors have poor adaptability to dynamic and changeable production scenes, failing to realize flexible, real-time and full-range safety monitoring [2]. This paper designs a vision-based safety early warning system oriented to human–robot collaborative scenes, which takes industrial binocular cameras as the perception carrier and relies on computer vision algorithms to complete real-time target detection, human-robot spatial coordinate calculation and multi-level dangerous area division. The system continuously tracks the three-dimensional spatial distance between human bodies and robot bodies in real time, automatically triggers graded safety early warning signals once workers invade the early warning zone or dangerous zone of the robot working space, and can output linkage deceleration and stop signals to the robot controller to avoid collision accidents [6]. This study establishes a multi-dimensional evaluation index system including detection delay, early warning accuracy and scene anti-interference ability, and divides test scenes into static robot fixed operation area, dynamic robot continuous movement and multi-person mixed operation scenarios for grouped comparative verification. Core mediating variables restricting early warning reliability are extracted, including image occlusion resistance, environmental light interference suppression, human body key point extraction accuracy and robot three-dimensional model matching precision [5]. The paper further summarizes typical application paradoxes … Read More