Research on Surface Defect Classification Algorithm of Steel Plate Based on Improved BP Neural Network

Abstract

This article aims to study a steel plate surface defect classification algorithm based on an improved BP neural network. Firstly, we analyzed the classification of surface defects on steel plates and proposed an image segmentation algorithm based on shape features, which divides the steel plate surface image into multiple regions and extracts shape features from different regions. Then, we used traditional BP neural networks and improved BP neural network models to classify these shape features to determine the type of surface defects on the steel plate.

Specifically, we propose an improved BP neural network model to address the issues of low classification accuracy and slow training speed that traditional BP neural network models face in dealing with multi class problems. The model uses momentum term and learning rate annealing technology to accelerate the network training process, and uses Sigmaid function instead of the traditional step function to improve the fitting ability of BP neural network.

Through a large number of experiments, we compared and analyzed the performance of traditional BP neural network and improved BP neural network in the classification accuracy and training speed of steel plate surface defects. The results show that the steel plate surface defect classification algorithm based on improved BP neural network has significantly improved classification accuracy and training speed compared to traditional BP neural network. This algorithm has important application value for automatic recognition and classification of surface defects on steel plates.

In summary, this article studies a steel plate surface defect classification algorithm based on an improved BP neural network. In the future, we will further optimize the performance of the algorithm and improve its application scenarios to improve its practicality and universality.