Tire damage image recognition based on improved convolutional neural network
Sushi Zhang, Yuhong Wu, Jun Chang · 2020
In the application of tire damage image recognition, the use of convolutional neural network to train the model needs to extract the feature information of the tire image, and when judging the feature information, it is prone to overfitting, which reduces the training accuracy of the algorithm. To solve this problem, an image recognition algorithm based on convolutional neural network and ridge regression analysis is proposed. A new regular term is introduced into the original loss function of the algorithm, in order to change the ratio of the two parts in the new loss function, and reduce the jitter of the fitted curve of the feature information. Finally, it is verified through experiments that the improved convolutional neural network algorithm can improve the training accuracy and image recognition rate of tire damaged images.