SGNet:A Lightweight Defect Detection Model for New Energy Vehicle Battery Current Collectors

The quality of the current collector, an essential component in new energy vehicle batteries, is crucial for battery performance and significantly impacts the safety of vehicle occupants. However, detecting defects in battery current collector in real-time industrial applications with limited computational resources poses a major challenge. To address this, our paper proposes SGNet (ShuffleNet V2 + G_GFPN), a lightweight model for current collector defect detection. SGNet utilizes ShuffleNet V2 as the backbone feature extraction network and a designed G_GFPN as the feature fusion network, reducing parameter redundancy to ensure high detection accuracy while minimizing computational costs. Our experimental results demonstrate that the SGNet model achieves an impressive 90.7% detection accuracy. Boasting a compact model size of merely 3.5M and remarkably low GFLOPs of 4.1, SGNet achieves an outstanding FPS (Frames Per Second) rate of 178.6. In comparison to other sophisticated lightweight network models, SGNet surpasses them by delivering superior detection accuracy while demanding less computational resources. The successful deployment of the SGNet model on the embedded NVIDIA Jetson Nano platform paves the way for real-time defect detection. With a swift detection time of 0.073 seconds per image, the model meets the stringent requirements for accuracy and real-time performance in identifying battery collector tray defects within real-world industrial environments.