|
|
基于改进ResNet的电池极片毛刺分类研究
|
Abstract:
针对电池极片毛刺漏检、过杀和检测效率低问题，提出一种基于改进ResNet的电池极片毛刺分类方法。在数据预处理阶段，采用图像生成模型来增加数据的多样性和复杂性，帮助模型更好地泛化；选择ResNet18作为骨干网络，通过融合改进MSE注意力机制提高每个毛刺类别的召回率和精确率，并利用GSConv卷积实现模型轻量化加快模型推理速度。实验结果表明：改进后的网络Ours-ResNet18相比较改进前网络F1-score得分达到96.78%，提高了5.92%，单张图片CPU检测时间达到15.26 ms，能够满足实际的检测需求。
Aiming at the problems of missed detection, over-killing and low detection efficiency of burrs on battery pole pieces, a classification method for burrs on battery pole pieces based on the improved ResNet is proposed. In the data preprocessing stage, an image generation model is adopted to increase the diversity and complexity of the data, helping the model to generalize better. ResNet18 is selected as the backbone network. By integrating and improving the MSE attention mechanism, the recall rate and precision rate of each burr category are improved. And GSConv convolution is utilized to make the model lightweight and accelerate the model’s inference speed. The experimental results show that compared with the network before improvement, the F1-score of the improved network Ours-ResNet18 reaches 96.78%, an increase of 5.92%. The CPU detection time for a single image is 15.26 ms, which can meet the actual detection requirements.
| [1] | 韩任杰. 极片毛刺导致电池短路的检测方法[J]. 电池, 2023, 53(5): 538-540. |
| [2] | 杨续来, 袁帅帅, 杨文静, 等. 锂离子动力电池能量密度特性研究进展[J]. 机械工程学报, 2023, 59(6): 239-254. |
| [3] | 崔远驰, 张嵩, 唐晗, 等. 中国仪器产业发展: 现状, 挑战及展望[J]. 光学精密工程, 2024, 32(13): 1965-2003. |
| [4] | 王庆秋, 李林升, 桂久琪, 等. 基于图像处理与卷积神经网络的锂电池极片缺陷检测与分类[J]. 制造业自动化, 2023, 45(10): 50-54. |
| [5] | Mao, W., Li, L., Tao, Y. and Zhou, W. (2023) Surface Defect Image Classification of Lithium Battery Pole Piece Based on Deep Learning. IEICE Transactions on Information and Systems, 106, 1546-1555. https://doi.org/10.1587/transinf.2023edp7058 |
| [6] | Xu, C., Li, L., Li, J. and Wen, C. (2021) Surface Defects Detection and Identification of Lithium Battery Pole Piece Based on Multi-Feature Fusion and PSO-SVM. IEEE Access, 9, 85232-85239. https://doi.org/10.1109/access.2021.3067641 |
| [7] | 黄梦涛, 王露. 锂电池极片缺陷特征融合与分类[J]. 制造业自动化, 2021, 43(10): 61-63. |
| [8] | 徐文超, 张思祥, 白芳, 等. 基于改进VGG网络的单体热电池X光图像无损检测方法研究[J]. 光电子·激光, 2024, 35(2): 180-190. |
| [9] | 沈凌云, 张洁, 孙甲云, 等. 基于改进Alexnet模型的光伏电池片瑕疵分类方法[J]. 自动化与信息工程, 2022, 43(4): 42-47. |
| [10] | 任惠, 夏静, 卢锦玲, 等. 基于红外图像和改进MobileNet-V3的光伏组件故障诊断方法[J]. 太阳能学报, 2023, 44(8): 238-245. |
| [11] | 李莹, 董耀, 何自芬, 等. 基于ResNet34-TE的喷印可变彩色二维码缺陷检测[J]. 激光与光电子学进展, 2024, 61(18): 1-12. |
| [12] | 吴岸聪, 林城梽, 郑伟诗. 面向跨模态行人重识别的单模态自监督信息挖掘[J]. 中国图象图形学报, 2022, 27(10): 2843-2859. |
| [13] | 莫定涛, 俱莹, 李文进, 等. 面向卫星网络的流量分类方法研究[J/OL]. 计算机工程, 1-13. https://doi.org/10.19678/j.issn.1000-3428.0069654, 2025-02-20. |
| [14] | He, K., Zhang, X., Ren, S. and Sun, J. (2016). Deep Residual Learning for Image Recognition. 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Las Vegas, 27-30 June 2016, 770-778. https://doi.org/10.1109/cvpr.2016.90 |
| [15] | 谢宇浩, 董前民, 金尚忠, 等. 基于深度神经网络的危化品拉曼光谱识别[J/OL]. 激光与光电子学进展, 1-17. http://kns.cnki.net/kcms/detail/31.1690.tn.20240828.1425.030.html, 2025-02-20. |
| [16] | 许冲, 刘庆华. 融合多头注意力机制的探地雷达反演网络[J/OL]. 现代雷达, 1-15. http://kns.cnki.net/kcms/detail/32.1353.TN.20240725.0938.002.html, 2025-02-20. |
| [17] | Ru, Y., An, G., Wei, Z. and Chen, H. (2024) Epilepsy Detection Based on Multi-Head Self-Attention Mechanism. PLOS ONE, 19, e0305166. https://doi.org/10.1371/journal.pone.0305166 |
| [18] | Xu, H., Xiao, Y., Sun, K. and Cui, L. (2024) Improved Depthwise Separable Convolution for Transfer Learning in Fault Diagnosis. IEEE Sensors Journal, 24, 33606-33613. https://doi.org/10.1109/jsen.2024.3432921 |
| [19] | Li, H., Li, J., Wei, H., et al. (2022) Slim-Neck by GSConv: A Better Design Paradigm of Detector Architectures for Autonomous Vehicles. |
