Enhancing Automation with Label Defect Detection and Content Parsing Algorithms

Abstract

The stable operation of power transmission and distribution is closely related to the overall performance and construction quality of circuit breakers. Focusing on circuit breakers as the research subject, we propose a machine vision method for automated defect detection, which can be applied in intelligent robots to improve detection efficiency, reduce costs, and address the issues related to performance and assembly quality. Based on the LeNet-5 convolutional neural network, a method for the detection of character defects on labels is proposed. This method is then combined with squeezing and excitation networks to achieve more precise classification with a feature graph mechanism. The experimental results show the accuracy of the LeNet-CB model can reach up to 99.75%, while the average time for single character detection is 17.9 milliseconds. Although the LeNet-SE model demonstrates certain limitations in handling some easily confused characters, it maintains an average accuracy of 98.95%. Through further optimization, a label content detection method based on the LSTM framework is constructed, with an average accuracy of 99.57%, and an average detection time of 84 milliseconds. Overall, the system meets the detection accuracy requirements and delivers a rapid response. making the results of this research a meaningful contribution to the practical foundation for ongoing improvements in robot intelligence and machine vision.