Classification of Car Scratch Types Based on Optimized BP Neural Network

Abstract

At present, the detection of scratch types on the surface of automobiles still adopts manual inspection, which has the disadvantages of high leakage detection rate and low efficiency. In order to realize automatic detection, this paper puts forward a kind of car scratch classification method based on optimized BP neural network (H-IGA-BP). The feature vector which extracts obvious scratch characteristic from the texture is served as the input of BP neural network. Aiming at the difficulty in determining the number of hidden layer nodes in BP neural network, the golden section algorithm is used to find the ideal value. The traditional BP neural network has long training time and easily falls into local extremum. By improving the adaptive genetic algorithm (IGA), the selection operator, crossover operator and mutation operator are modified to optimize the weights and thresholds of BP neural network. The experimental results show that this method can effectively improve the accuracy and robustness of scratches classification. It provides a new method for the automatic detection of car scratch types.