An Approach for Car Paint Defect Classification Integrating Channel Shuffle with MobileNetV3

To address the limitation of MobileNet where restricted information flow between channels impedes the model's ability to capture complex inter-channel feature relationships, thereby reducing detection accuracy, this paper proposes a Python-based car paint defect classification method that integrates Channel Shuffle with MobileNetV3. First, data augmentation is applied to the dataset. Within the MobileNetV3 framework, the Channel Shuffle operation is implemented via Python programming after the depthwise (DW) convolution module to enable channel recombination. Ablation experiments are conducted to evaluate the contributions of the squeeze-andexcitation (SE) and Channel Shuffle modules. Experimental results demonstrate that the proposed Python-implemented algorithm exhibits promising application potential in automotive paint defect detection, effectively improving recognition efficiency and accuracy while reducing detection costs. The model can be deployed on mobile or embedded devices.