19 October 2023 in IEEE Sensors Journal
As an essential component of the new energy vehicle battery, current collectors affect the performance of battery and are crucial to the safety of passengers. The significant differences in shape and scale among defect types make it challenging for the model detection of current collector defects. In order to reduce application costs, and conduct real-time detection with limited computing resources, we propose an end-to-end adaptive and lightweight defect detection model for the battery current collector, DGNet. Firstly, we designed an adaptive lightweight backbone network (DOS module) to adaptively extract useful features adaptively along all four dimensions of kernel space while maintaining low computational complexity. Secondly, we designed a lightweight feature fusion network (GS_FPN), which reduces parameter redundancy and fully utilizes the semantic information of the feature maps of backbone network while ensuring detection accuracy. Experimental results show that DGNet achieves a mAP50 (mean Average Precision at IoU threshold 0.5) of 91.8% on the self-made BCC surface defect database, with a model size of 4.0M, only 3.7 GFLOPs (Giga Floating-point Operations per Second), and FPS (Frames Per Second) of 181.8. To further demonstrate the capabilities of DGNet, we test it on the publicly NEU surface defect database, and the results showed that the DGNet exhibited good generalization. Compared to current advanced lightweight network models, it achieves higher detection accuracy and lower computational overhead, reaching the state-of-the-art level. Finally, we deployed DGNet on the embedded platform NVIDIA Jetson Nano for real-time detection, achieving a detection time of 0.074s per image, meeting the accuracy and real-time detection requirements for battery current collector defect detection tasks in practical industrial applications.
15 May 2024 in IEEE Sensors Journal
The future direction of global automotive development is electrification, and the battery current collector is an essential component of new energy vehicle batteries. However, the welding defects in the battery current collector during the welding process are characterized by a disorganized distribution, extensive size variations, multiple types, and ambiguous features, posing challenges for detecting welding defects in the current collector. This paper proposes a lightweight deep-learning algorithm called MGNet for detecting welding defects in the current collectors. We introduce a lightweight MDM module based on multi-scale channels, which utilizes deep dynamic convolutions as its basic structure to extract compelling features while reducing computational complexity. We also propose a lightweight feature fusion network called GS_GFPN, which fully leverages the semantic information of the backbone network feature maps while reducing parameter redundancy and maintaining detection accuracy. Experimental evaluations on both the BCC (battery current collector) surface defect database and the publicly available NEU (Northeastern University) surface defect database demonstrate that MGNet outperforms existing methods with significant improvements in detection accuracy. The mean average precision (mAP) at IoU threshold 0.5 is 93.9% and 78.0%, respectively, with Frames Per Second (FPS) of 212.8 and 238.1 and a model weights of only 3.1M. Moreover, the algorithm is successfully deployed on the NVIDIA Jetson Nano embedded device, enabling real-time defect detection for practical industrial applications.