Research on Few-Shot Defect Detection Algorithm Based on Federated Learning

The algorithm based on deep learning has been widely used in defect detection in all walks of life, but the performance of the deep learning model depends mainly on rich annotation data. However, in the actual scene, obtaining large-scale, high-quality data to ensure users' privacy and safety is challenging, which limits its further promotion in specific application fields. To solve this problem, we propose a federated few-shot defect detection framework, which uses the privacy protection of the federated framework to jointly train independent few-shot tasks distributed on different clients to obtain a few-shot model that can quickly adapt to new tasks with limited data. We have done many experiments to evaluate our framework's effectiveness, and the results show that our framework is superior to the baseline and achieves the same performance as the model trained with a lot of data.