A Hybrid Inception V3-SVM Based Framework for Steel Surface Defect Classification

The classification and identification of steel defects are critical in the manufacturing process, as accurate and rapid classification can significantly enhance product quality. This paper proposes a hybrid model combining the Inception V3 convolutional network and Support Vector Machine (SVM), based on transfer learning, and validates it on the NEU and X-SDD datasets. First, the Inception V3 model is employed for transfer learning to extract features of steel surface defect(SSD), which are then passed to the SVM for classification. On the NEU and X-SDD datasets, the accuracy, macro precision, macro recall, and macro F1-score achieved were 0.9944, 0.9946, 0.9944, 0.9944, and 0.9761, 0.9748, 0.9769, 0.9757. Compared to the standard model, these metrics are improved by 2.58%, 2.42%, 2.58%, 2.59%, and 2.08%, 1.84%, 2.41%, 2.19%. Regarding inference time, the proposed model achieved inference times of 0.083s and 0.066s on the NEU and X-SDD datasets, representing improvements of 88.47% and 87.69% compared to the standard model. Additionally, a performance comparison was conducted with classic convolutional network models, including MobilenetV2, VGG16, ResNet50, and NASNetMobile. The results demonstrate that the Inception V3-SVM hybrid model offers high accuracy and efficiency, showing great potential for application in SSD classification.