Industrial Surface Defect Detection Based on Privacy Protection Federal Learning
Aiming at the needs of multi-party data centralization and security privacy in industrial defect detection,this paper proposes a federated learning industrial defect detection framework based on privacy pro-tection.This framework uses multiple distributed devices to capture product images to identify defects,and then adds Gaussian noise to the central server to make up for the data insecurity and increased communication burden of transferring the original data to the server in the traditional way.The experimental results confirm that after using the multi-site global model and Gaussian differential privacy encryption processing,the test re-sults are 0.7 mAP higher than the classical FRCN method,realizing the defect detection accuracy that can be compared with the traditional centralized method,and provide a solution for industrial information collaboration and individual data protection.
NETL
NSTL
万方数据
