Research on Defect Detection of Castings Based on Deep Residual Network

In this study, we proposed a method for detecting the appearance defect of castings based on deep residual network, which is used to solve the problems of low accuracy, difficult application conditions and insufficient robustness of traditional defect detection methods. This method divides the casting into multiple regions, preprocesses the image of each region, and then inputs the processed image into the convolutional neural network to extract the features, and finally determines whether the sample has defects. The deep residual network ResNet-34 was chosen as the network model, and its activation function was improved. The ASoftReLU function was proposed to alleviate the neuron-death problem and improve the accuracy and fitting speed of the network. Finally, the improved defect detection system was tested on the data set of castings. Through the comparison and analysis of the experimental results, the network model with the highest accuracy and the most generalization ability was obtained. Experimental results show that the accuracy of this method is much higher than the traditional method.