Journal of Computer Science Research

Optimization of Mobile Robot Delivery System Based on Deep Learning

2024-09-03T16:49:28+08:00

In modern logistics and delivery systems, mobile robot delivery systems have garnered significant attention due to their efficiency and flexibility. However, existing robot delivery systems still face numerous challenges in complex and dynamically changing environments. For instance, traditional algorithms exhibit low efficiency when processing high-dimensional and unstructured data, making it difficult to adapt to real-time changing environments, which results in reduced accuracy and efficiency in path planning and task execution. Additionally, the lack of effective perception and decision-making mechanisms makes it challenging for robots to handle complex scenarios and variable delivery demands. To address these issues, this paper proposes an optimization method for mobile robot delivery systems based on deep learning. Firstly, this study introduces a spatial attention mechanism into the model. By focusing on key areas in the environment and dynamically adjusting the attention points, robots can better recognize and avoid obstacles in complex environments, thus improving the accuracy of navigation and path planning. Secondly, the Deep Deterministic Policy Gradient (DDPG) algorithm is employed for policy optimization, facilitating efficient learning in high-dimensional continuous spaces, enabling robots to learn effective delivery strategies in complex environments. Finally, through an end-to-end optimization approach, the system can directly convert sensor data inputs into control command outputs, reducing the complexity and error accumulation of intermediate steps and simplifying the system structure. Experimental results demonstrate that the proposed method significantly enhances the overall performance of the delivery system. It performs exceptionally well on several key indicators, including the accuracy of path planning, task execution efficiency, and system robustness. The effectiveness of combining the spatial attention mechanism with the deep policy gradient algorithm has been fully validated, providing new insights and methods for future optimization of robot delivery systems.

The Stronger and Weaker Bull-bear Point Algorithm to Find the Opportunity of Buying Stocks Based on Big Data

2024-08-13T11:42:50+08:00

Study and develop the machine learning algorithm to find the opportunity of profiting in high probability for the investor to buy a stock based on the analysis of big data. In the first place, a preparative algorithm to find the bull-bear points is established. And then, basing on the preparative algorithm, a weaker bull-bear point algorithm and a stronger algorithm bull-bear point algorithm to find the opportune time of buying stock are respectively developed. What we have done in the present work can advance the development of the stock study and the quantitative technology.

Human-centered Artificial Intelligence Development

2024-07-16T16:43:33+08:00

Few researchers provide a wider vision of artificial feet, hands, mouths, eyes, ears, and brains. This limits our vision of them and their significant impacts on the modern Industrial Revolution and Artificial Intelligence (AI) history. This article presents a novel perspective on human-centered social development starting from artificial feet. After briefly reviewing AI, this article explores the age of AI and artificial feet, hands, mouths, eyes, ears, and brains. It also applies AI to artificial feet and artificial brains. The research reveals that artificial feet are one of the origins of the Industrial Revolution and a real foundation of AI. The study demonstrates that artificial feet and brains liberate our body and society, whereas from artificial brains to artificial feet is control of our body and society. This article also looks at AI's trends and challenges. The approach in this article will facilitate the research and development of big data, analytics, and intelligences.

Improving U-Net Performance for Tumor Segmentation Using Attention Mechanisms

2024-09-12T16:43:03+08:00

U-Net is a widely recognized neural network model for medical image segmentation, renowned for its efficiency in extracting features from both current and past input data. However, traditional U-Net models exhibit limitations in extracting edge features, particularly in medical CT images characterized by complex gray distributions and close pixel intervals. This leads to suboptimal performance, with low accuracy, recall, intersection over union (IoU), and F1-score. This research proposes an improved U-Net model incorporating an attention mechanism to enhance tumor segmentation accuracy and efficiency. The attention mechanism strategically weights important features, directing the network to focus on task-relevant areas. Experimental results demonstrate that our proposed attention-based U-Net model significantly improves tumor segmentation performance, achieving notable enhancements in accuracy, recall, IoU, and F1-score. Further validation across diverse datasets confirms the model's generalization ability and superiority compared to the original U-Net method. This research contributes to the advancement of medical image segmentation techniques, highlighting the potential of attention mechanisms in optimizing deep learning models for clinical applications.

Improvement and Research on Object Detection Algorithm for Wind Turbine Blade Defect Scenarios Based on YOLOv8

2024-08-19T14:17:18+08:00

Wind turbine blades are vital for energy generation, where defects can cause efficiency loss and costly maintenance. This paper proposes an improved object detection algorithm based on YOLOv8 for detecting defects in wind turbine blades. Enhancements include network architecture modifications and advanced attention mechanisms, which boost detection accuracy while maintaining real-time processing. Our approach is tested on a custom dataset, showing better performance than the standard YOLOv8 model. These improvements can enhance automated defect detection in wind turbines, reducing downtime and operational costs, and contributing to more efficient renewable energy maintenance.

Learning Dominant Urban Flows around High-Rise Buildings with Data-Driven Balance Models

2024-07-31T17:31:38+08:00

This thesis develops a data-driven dominant balance model to recognise and cluster the flow pattern blowing through a high-rise building in an urban area under neutral atmospheric conditions. To be consistent with the governing equation used in simulations, the Reynolds-Averaged Navier-Stokes (RANS) equation is selected as the governing equation. It is divided into six sub-parts based on the physical meanings of each term in RANS. The time-averaged simulation results are used as the data set basis for further machine learning and clustering. The approach used to achieve the final dominant balance models consists of knowledge from fluid mechanics, statistics and programming. Knowledge from fluid mechanics is mainly used for proposing governing equations and interpreting the final outcomes, whereas the knowledge from programming is used for script writing and program running. Finally, the knowledge from statistics is the key for algorithms to achieve the clustering and dominant balance model acquirement. This approach includes the finite difference method, Gaussian mixture models (GMM), singular value decomposition and sparse principal component analysis (SPCA). The finite difference method is used for approximating the derivatives in RANS, which works as a post-processing step. GMM are trained by using randomly subsampled points and applied for the clustering of the processed data points. A drawback of yielding overlapping and trivial clusters of GMM is spotted and SPCA is applied as the solution to trivial results, using regularisation to proceed with a sparse approximation for excessive cluster elimination. The final data-driven dominant balance models are obtained and visualised by generating two tables for two cases.

Research and Model Library Construction in Teacher-Student Learning Architectures for Knowledge Transfer

2024-09-24T09:29:34+08:00

This paper summarizes and replicates multiple classical and cutting-edge knowledge transfer methods, including Factor Transfer (FT), Knowledge Distillation (KD), Deep Mutual Learning (DML), Contrastive Representation Distillation (CRD), and Born-Again Self-Distillation (BSS). Additionally, we studied three advanced knowledge transfer methods: Relational Knowledge Distillation (RKD), Similarity-Preserving (SP), and Attention-based Feature Distillation (AFD), successfully replicating an optimized version of KD, namely RKD. Based on these methods, a flexible model library was constructed in Pycharm, allowing the quick integration of multiple knowledge transfer strategies. The experimental results are visualized through a user-friendly interface, enabling intuitive comparisons of model training speed and performance across different methods. This research provides valuable insights into the challenge of building a reusable framework that efficiently integrates various knowledge transfer strategies into deep neural networks.