通讯机构：
[Yang, H ] W;Wuhan Polytech Univ, Coll Math & Comp Sci, Wuhan 430023, Peoples R China.
关键词：
BP neural network;hybrid optimization model;metaheuristic algorithm;renewable energy integration;wind power prediction
摘要：
The integration of intermittent wind power into modern grids necessitates highly accurate forecasting models to ensure stability and efficiency. To address the limitations of traditional backpropagation (BP) neural networks, such as slow convergence and susceptibility to local optima, this study proposes a novel hybrid framework: the Multi-Strategy Coati Optimization Algorithm (SZCOA)-optimized BP neural network (SZCOA-BP). The SZCOA integrates three innovative strategies-a population position update mechanism for global exploration, an olfactory tracing strategy to evade local optima, and a soft frost search strategy for refined exploitation-to enhance the optimization efficiency and robustness of BP networks. Evaluated on the CEC2017 benchmark, the SZCOA outperformed state-of-the-art algorithms, including ICOA, DBO, and PSO, achieving superior convergence speed and solution accuracy. Applied to a real-world wind power dataset (912 samples from Alibaba Cloud Tianchi), the SZCOA-BP model attained an R² of 94.437% and reduced the MAE to 10.948, significantly surpassing the standard BP model (R²: 81.167%, MAE: 18.891). Comparative analyses with COA-BP, BWO-BP, and other hybrid models further validated its dominance in prediction accuracy and stability. The proposed framework not only advances wind power forecasting but also offers a scalable solution for optimizing complex renewable energy systems, supporting global efforts toward sustainable energy transitions.
摘要：
Data-driven applications need advanced predictive modeling to manage nonlinear relationships and high-dimensional datasets. To address these challenges, this research presents a novel hybrid optimization model integrating an Improved Zebra Optimization Algorithm (IZOA) with a Back Propagation (BP) neural network to enhance predictive performance in complex datasets. The IZOA addresses inherent limitations in traditional optimization methods by employing a Logistic chaotic initialization technique that increases population diversity. Furthermore, a Golden Sine optimization strategy is incorporated to balance exploration and exploitation effectively. The model’s architecture leverages the strengths of both IZOA and BP neural networks, allowing for refined local adjustments while maintaining global search capabilities. Experimental evaluations demonstrate that the IZOA-BP model significantly outperforms conventional approaches, achieving superior metrics such as a Root Mean Squared Error (RMSE) of 2.271, Mean Absolute Error (MAE) of 0.891, and an R² value of 0.935. These advancements highlight the model’s robustness in capturing nonlinear interactions and adapting to high-dimensional data, positioning the IZOA-BP framework as a transformative tool for various predictive applications.
关键词：
Drone-based monitoring;Construction site safety;IoT integration;GSConv;Real-time safety detection;Linformer
摘要：
Real-time safety monitoring on construction sites is essential for ensuring worker safety, but traditional detection methods face challenges in dynamic environments with moving objects, occlusions, and complex conditions. To address these limitations, we propose GS-LinYOLOv10, an improved model based on YOLOv10, specifically designed for drone-based safety monitoring. The GSConv module introduces a lightweight feature extraction mechanism, reducing computational complexity without compromising detection accuracy. The Linformer-based attention mechanism efficiently captures global context, addressing challenges in dynamic and complex environments. The model integrates IoT sensor data for real-time feedback, incorporates the GSConv module for lightweight feature extraction, and utilizes a Linformer-based attention mechanism to efficiently capture global context. These innovations reduce computational complexity while significantly improving detection accuracy. Experimental results show that GS-LinYOLOv10 achieves a precision of 91.2% and a mean average precision (mAP) of 89.4%, outperforming existing models. The integration of IoT sensors allows the drone system to dynamically adjust its monitoring focus, improving adaptability to changing environments and enhancing hazard detection. This research provides an advanced, drone-based IoT-enhanced solution for real-time construction site safety monitoring, offering a more effective and efficient approach to safety management.
Real-time safety monitoring on construction sites is essential for ensuring worker safety, but traditional detection methods face challenges in dynamic environments with moving objects, occlusions, and complex conditions. To address these limitations, we propose GS-LinYOLOv10, an improved model based on YOLOv10, specifically designed for drone-based safety monitoring. The GSConv module introduces a lightweight feature extraction mechanism, reducing computational complexity without compromising detection accuracy. The Linformer-based attention mechanism efficiently captures global context, addressing challenges in dynamic and complex environments. The model integrates IoT sensor data for real-time feedback, incorporates the GSConv module for lightweight feature extraction, and utilizes a Linformer-based attention mechanism to efficiently capture global context. These innovations reduce computational complexity while significantly improving detection accuracy. Experimental results show that GS-LinYOLOv10 achieves a precision of 91.2% and a mean average precision (mAP) of 89.4%, outperforming existing models. The integration of IoT sensors allows the drone system to dynamically adjust its monitoring focus, improving adaptability to changing environments and enhancing hazard detection. This research provides an advanced, drone-based IoT-enhanced solution for real-time construction site safety monitoring, offering a more effective and efficient approach to safety management.
作者机构：
[Hua Yang; Zhonger Li; Zhan Shu; Junda Liu; Ming Zhao; Mingzhi Mu] College of Mathematics and Computer Science, Wuhan Polytechnic University, Wuhan, China
会议名称：
2025 8th International Conference on Advanced Electronic Materials, Computers and Software Engineering (AEMCSE)
会议时间：
09 May 2025
会议地点：
Nanjing, China
会议论文集名称：
2025 8th International Conference on Advanced Electronic Materials, Computers and Software Engineering (AEMCSE)
摘要：
To enhance the accuracy and convergence speed of electricity load forecasting, this paper proposes an optimized electricity load forecasting model that integrates a Modified Frilled Lizard Optimization (MFLO) algorithm with a Backpropagation (BP) neural network. The MFLO addresses the traditional challenges of slow convergence and local optima by incorporating Lévy flight mechanisms and self-weight factors, which enhance global search capabilities. By integrating information entropy into fitness adjustments, the model improves the quality of initial solutions, leading to better predictions. The MFLO-BP model demonstrates significant performance improvements, achieving a Mean Absolute Percentage Error (MAPE) of 1.09%, a Root Mean Square Error (RMSE) of 152.5, and an $\mathbf{R}^{2}$ value of 0.97, outperforming conventional BP models with MAPE of 1.57% and RMSE of 205.4. Our findings contribute to more effective energy management strategies and offer a robust framework for evolving energy demands.
摘要：
To solve the missed and wrong detection problems of the object detection model in identifying soybean companion weeds, this paper proposes an enhanced multi-scale channel feature model based on RT-DETR (EMCF-RTDETR). First, we designed a lightweight hybrid-channel feature extraction backbone network, which consists of a CGF-Block module and a FasterNet-Block module working together, aiming to reduce the amount of computation and the number of parameters while improving the efficiency of feature extraction. Second, we constructed the EA-AIFI module. This module enhances the extraction of detailed features by combining the in-scale feature interaction module with the Efficient Additive attention mechanism. In addition, we designed an Enhanced Multiscale Feature Fusion (EMFF) network structure, which first differentiates the inputs of the three feature layers and then ensures the effective flow between the original and enhanced features of each feature layer by two multiscale feature fusions as well as one diffusion. The experimental results demonstrate that the EMCF-RTDETR model improves the average precision mAP50 and mAP50:95 by 3.3% and 2.2%, respectively, compared to the RT-DETR model, and the FPS is improved by 10%. Moreover, our model outperforms other mainstream detection models in terms of accuracy and speed, revealing its significant potential for soybean weed detection.
摘要：
To enhance the accuracy of PM2.5 concentration predictions amidst inherent randomness and complexity, this paper introduces a novel prediction method called the Integrated Black-winged Kite Algorithm with Backpropagation (IBKA-BP). This approach improves the traditional Backpropagation (BP) neural network by optimizing its weights and thresholds, effectively addressing common issues such as slow convergence and the tendency to get trapped in local optima. Comparative analyses of prediction errors demonstrate that the IBKA-BP model outperforms other advanced PM2.5 concentration prediction models. Notably, it achieves a Mean Absolute Error (MAE) of 5.51, a Root Mean Square Error (RMSE) of 7.29, and an $\mathbf{R}^{2}$ value of $\mathbf{0. 9 3 6 4}$, indicating superior predictive accuracy. These findings confirm that the IBKA-BP model significantly enhances PM2.5 concentration prediction, offering a robust framework for future applications in air quality assessment.
摘要：
Corn is an important source of renewable energy, which can be converted into ethanol through fermentation and distillation. Ethanol, as a clean and renewable energy source, can not only be used as an additive and alternative to gasoline but also can be used to manufacture chemicals such as acetaldehyde, ethylene glycol, ethylamine, ethyl acetate, acetic acid, chloroethane, etc. However, after infection, corn leaves may rot, turn yellow, and produce a large number of viruses, leading to a decrease in corn yield. Timely and accurate detection of infected corn leaves is an important measure for the prevention and treatment of corn leaf infection. The existing target detection algorithms have unsatisfactory effects on the detection and classification of infected corn leaves. To quickly and accurately detect corn leaves and classify disease-infected leaves to achieve ideal detection results in practical corn leaf detection situations, this paper proposes a new algorithm called yolo-SDW based on the yolov5 algorithm. The yolo-SDW algorithm introduces a spatial depth conversion convolution (SPD-Conv) into the backbone network of the original yolov5 algorithm, replacing the traditional stride convolution with SPD-Conv. Very paramount for operational efficiency, this will enhance the adaptability and usability of the model. A vision Transformer with deformable attention (DAT) is introduced. This attention automatically adjusts the attention distribution according to the data needs when processing images of different scales, thereby improving the accuracy and performance of the model. Meanwhile, a novel Wise-IOU V3 loss function is used as the bounding box loss function, resulting in a lower false positive rate when dealing with dense targets. The experimental results show that the improved algorithm has a 6.4 % increase in average precision mAP compared to the original yolov5 algorithm, reaching 83.5 %. The speed has increased by 3.2 %, while precision and recall rates have also been significantly improved.
Corn is an important source of renewable energy, which can be converted into ethanol through fermentation and distillation. Ethanol, as a clean and renewable energy source, can not only be used as an additive and alternative to gasoline but also can be used to manufacture chemicals such as acetaldehyde, ethylene glycol, ethylamine, ethyl acetate, acetic acid, chloroethane, etc. However, after infection, corn leaves may rot, turn yellow, and produce a large number of viruses, leading to a decrease in corn yield. Timely and accurate detection of infected corn leaves is an important measure for the prevention and treatment of corn leaf infection. The existing target detection algorithms have unsatisfactory effects on the detection and classification of infected corn leaves. To quickly and accurately detect corn leaves and classify disease-infected leaves to achieve ideal detection results in practical corn leaf detection situations, this paper proposes a new algorithm called yolo-SDW based on the yolov5 algorithm. The yolo-SDW algorithm introduces a spatial depth conversion convolution (SPD-Conv) into the backbone network of the original yolov5 algorithm, replacing the traditional stride convolution with SPD-Conv. Very paramount for operational efficiency, this will enhance the adaptability and usability of the model. A vision Transformer with deformable attention (DAT) is introduced. This attention automatically adjusts the attention distribution according to the data needs when processing images of different scales, thereby improving the accuracy and performance of the model. Meanwhile, a novel Wise-IOU V3 loss function is used as the bounding box loss function, resulting in a lower false positive rate when dealing with dense targets. The experimental results show that the improved algorithm has a 6.4 % increase in average precision mAP compared to the original yolov5 algorithm, reaching 83.5 %. The speed has increased by 3.2 %, while precision and recall rates have also been significantly improved.
摘要：
With the continuous development of artificial intelligence technology, deep learning methods have been widely used in smart agriculture. With the continuous progress of object detection algorithms, it is a future trend to introduce computer vision methods into smart agriculture. This paper proposes an improved YOLOv8 network model for detecting whether apple is still in a healthy state in smart agriculture systems. By introducing a better backbone network EfficientNet, features can be extracted from the data efficiently. In addition, by introducing a novel WIOU calculation function, the rectangular box can be computed better. In this experiment, the average accuracy of the improved YOLOv8-Enet is mAP0.5 and mAP 0.5:0.95, which are 7.1% and 6.5% higher than that of YOLOv8-base, respectively. The proposed YOLOv8-Enet model can effectively detect apple surface defect and provide theoretical and technical support for future research on vision of smart agriculture
期刊：
2024 International Conference on New Trends in Computational Intelligence (NTCI),2024年:91-95
作者机构：
[Hua Yang; Chengwu Peng; Shenyang Sheng; Qi Wang; Jie Xiao; Shi Cao; Zhaoqi Meng; Tianwei Tang; Rou Fu; Xiaomei Huang] School of Mathematics and Computer Science, Wuhan Polytechnic University, Wuhan, China
会议名称：
2024 International Conference on New Trends in Computational Intelligence (NTCI)
会议时间：
18 October 2024
会议地点：
Qingdao, China
会议论文集名称：
2024 International Conference on New Trends in Computational Intelligence (NTCI)
关键词：
Tomato leaf;SCConv;PSA;DIoU;YOLOv5
摘要：
To address the challenges of complex backgrounds and low accuracy in detecting tomato leaves, we propose an improved tomato leaf detection model, YOLOv5s-SPD, based on the YOLOv5 network, for identifying tomato leaf diseases. First, the SSConv module is introduced into the backbone network of the YOLOv5 algorithm, replacing the C3 module with the SSConv module to reduce feature redundancy in both spatial and channel dimensions, thus lowering the computational load of the model. Additionally, a PSA attention mechanism is incorporated to enhance the model's feature extraction capability for tomato leaf boundaries in harsh environments, reducing the miss detection rate. The bounding box loss function is also replaced with DIoU, taking into account the distance and overlap between bounding boxes, thereby enabling more precise measurement of detection box accuracy and improving the positioning accuracy of tomato leaves. Experimental results show that compared to the original YOLOv5 algorithm, the improved algorithm increases the mean precision by 2.4%, reaching 85.9%. The recall rate also shows a significant improvement.
摘要：
Relation extraction (RE) is vital in natural language processing (NLP) for Analyzing the relationships among entities within unstructured text, supporting applications like knowledge graphs and question-answering systems. Existing methods often utilize graph neural networks (GNNs) and pretrained language models such as BERT, but they struggle with long-range dependencies and class imbalance in sparse relationships. In this paper, we introduce AxU-Doc, an innovative model that leverages axial attention within a U-shaped architecture to effectively acquire comprehensive information and promote logical inference between entities. Additionally, we adopt the HingeABLoss function, replacing the conventional cross-entropy loss, focusing on tackling class imbalance and enhance model efficiency. Our experiments on two publicly available datasets shows that AxU-Doc achieves competitive results, confirming its effectiveness in document-level RE tasks.
会议名称：
18th International Conference on Bio-Inspired Computing: Theories and Applications (BIC-TA)
会议时间：
DEC 15-17, 2023
会议地点：
Changsha, PEOPLES R CHINA
会议主办单位：
[Yang, Hua;Li, Jian;Liu, Neng;Yi, Kecheng;Wang, Jing;Fu, Rou;Zhang, Jun;Xiang, Yunzhu;Yang, Pengcheng;Hang, Tianyu] Wuhan Polytech Univ, Sch Math & Comp Sci, Wuhan 430040, Peoples R China.^[Zhang, Tiancheng;Wang, Siyi] Wuhan BisiCloud Technol Co Ltd, Wuhan 430015, Peoples R China.
会议论文集名称：
Communications in Computer and Information Science
关键词：
GA-BP neural network;Prediction model;Rice processing loss
摘要：
Food is closely related to national economy and people's livelihood. Rice is the largest grain crop in China, it is crucial to predict the loss rate of rice during processing to reduce food waste and ensure food security. This study first obtained the loss rate of rice processing through the recovery survey form of enterprises. Then, prediction was carried out using two common models: the BP neural network and multiple linear regression. Finally, the genetic algorithm was applied to optimize the BP neural network for further prediction and com-pared with the original models. The experimental results showed that the GA-BP model had higher prediction accuracy and smaller error compared to the first two models. It is valuable in reducing processing losses and maintaining food security.
摘要：
In this paper, in order to improve the search ability and adaptability of the seagull optimization algorithm, a multi-strategy collaborative improvement-based seagull optimization algorithm (MI-SOA) is proposed. Firstly, LogisticsTent chaotic mapping with is introduced to enhance the global search ability of the algorithm; secondly, the algorithm's search path is optimized by filtering out the superior and inferior solution locations through the inverse learning strategy to improve the quality of the solution; the algorithm combines the Lévy flights and the positive cosine operator, which enhances the algorithm's local search accuracy and flexibility; the improved seagull optimization algorithm has been developed in the mathematical model and parameter settings are optimized, and by adjusting the iterative process and parameter selection of the algorithm, the algorithm can be better adapted to various complex optimization problems. The experimental results show that the MI-SOA algorithm has better performance and faster optimization ability than other algorithms in solving numerical optimization problems; the improved seagull optimization algorithm is applied to three-dimensional path planning problems in practical engineering applications.
通讯机构：
[Yang, H ] W;Wuhan Polytech Univ, Sch Math & Comp Sci, Wuhan 430040, Peoples R China.
关键词：
text classification;convolutional neural networks (CNN);attention mechanism;multi-layer feature fusion
摘要：
Text classification is one of the fundamental tasks in natural language processing and is widely applied in various domains. CNN effectively utilizes local features, while the Attention mechanism performs well in capturing content-based global interactions. In this paper, we propose a multi-layer feature fusion text classification model called CAC, based on the Combination of CNN and Attention. The model adopts the idea of first extracting local features and then calculating global attention, while drawing inspiration from the interaction process between membranes in membrane computing to improve the performance of text classification. Specifically, the CAC model utilizes the local feature extraction capability of CNN to transform the original semantics into a multi-dimensional feature space. Then, global attention is computed in each respective feature space to capture global contextual information within the text. Finally, the locally extracted features and globally extracted features are fused for classification. Experimental results on various public datasets demonstrate that the CAC model, which combines CNN and Attention, outperforms models that solely rely on the Attention mechanism. In terms of accuracy and performance, the CAC model also exhibits significant improvements over other models based on CNN, RNN, and Attention.
摘要：
In recent years, the domain of diagnosing plant afflictions has predominantly relied upon the utilization of deep learning techniques for classifying images of diseased specimens; however, these classification algorithms remain insufficient for instances where a single plant exhibits multiple ailments. Consequently, we view the region afflicted by the malady of rice leaves as a minuscule issue of target detection, and then avail ourselves of a computational approach to vision to identify the affected area. In this paper, we advance a proposal for a Dense Higher-Level Composition Feature Pyramid Network (DHLC-FPN) that is integrated into the Detection Transformer (DETR) algorithm, thereby proffering a novel Dense Higher-Level Composition Detection Transformer (DHLC-DETR) methodology which can effectively detect three diseases: sheath blight, rice blast, and flax spot. Initially, the proposed DHLC-FPN is utilized to supersede the backbone network of DETR through amalgamation with Res2Net, thus forming a feature extraction network. Res2Net then extracts five feature scales, which are coalesced through the deployment of high-density rank hybrid sampling by the DHLC-FPN architecture. The fused features, in concert with the location encoding, are then fed into the transformer to produce predictions of classes and prediction boxes. Lastly, the prediction classes and the prediction boxes are subjected to binary matching through the application of the Hungarian algorithm. On the IDADP datasets, the DHLC-DETR model, through the utilization of data enhancement, elevated mean Average Precision (mAP) by 17.3% in comparison to the DETR model. Additionally, mAP for small target detection was improved by 9.5%, and the magnitude of hyperparameters was reduced by 324.9 M. The empirical outcomes demonstrate that the optimized structure for feature extraction can significantly enhance the average detection accuracy and small target detection accuracy of the model, achieving an average accuracy of 97.44% on the IDADP rice disease dataset.
通讯机构：
[Chong Wang; Yinsheng Qiu] A;Authors to whom correspondence should be addressed.<&wdkj&>Hubei Key Laboratory of Animal Nutrition and Feed Science, School of Animal Science and Nutrition Engineering, Wuhan Polytechnic University, Wuhan 430023, China<&wdkj&>Authors to whom correspondence should be addressed.<&wdkj&>College of Animal Science and Technology & College of Veterinary Medicine, Zhejiang A & F University, Hangzhou 311300, China
摘要：
Aflatoxin M1 (AFM1), a group 1 carcinogen, is a risk factor to be monitored in milk. This study aimed to investigate the occurrence of AFM1 in milk in Xinjiang, China, and to assess the risk of exposure for milk consumers in different age-sex groups. A total of 259 milk samples including pasteurized milk (93 samples), extended-shelf-life (ESL) milk (96), and raw donkey milk (70) were collected in Xinjiang from January to March in 2022. The AFM1 content of the milk samples was detected using a validated ELISA method. Of the 259 total samples analyzed for AFM1, 84 (32.4%) samples were contaminated at levels greater than the detection limit of 5 ng/L, with the maximum level of 16.5 ng/L. The positive rates of AFM1 in pasteurized milk and ESL milk were 43.0% (n = 40) and 45.8% (n = 44), respectively, and AFM1 was undetectable in donkey milk. The estimated daily intakes of AFM1 in each age group were lower than the hazard limits and were similar between male and female milk consumers. Therefore, the AFM1 contamination of milk in Xinjiang is low but still needs to be continuously monitored considering that children are susceptible to AFM1.
