A Lightweight Model Solution for Aluminum Material Defect Detection at the Edge

Zhengzhou Chen, Jinhao Liang, Yueling Zhang, Jiangtao Wang, Qing Zhang, Gang Xu

Abstract: Defect detection models have been widely applied in industrial edge platform. However, in the field of aluminum material defect detection still have some challenges: (i) limited availability of aluminum material defect datasets and (ii) the high computational cost associated with large-scale models. To address these challenges, we propose a novel lightweight model for aluminum material defect detection. For the first challenge, we propose Positive Sample Augmentation(PSA) to effectively increase the number of samples. Additionally, we employ model pruning, knowledge distillation and model quantization to reduce inference latency and model size. We have successfully deployed this solution on real-world industrial platforms, double the running speed and reduce approximately 75% hardware costs, thereby validating the feasibility of our approach.

External IDs:dblp:conf/sies/ChenLZWZX24