Aiming at the disadvantages of CNN model for wear particle analysis,such as too many parameters,slow operation speed,and difficulty in practical application,research on lightweight of the CNN model for wear particle image classification was carried out.The parameters,computation and pruning sensitivity of each layer in the CNN model were analyzed,which determined the convolution layer 4 and 5 as the filter pruning targets.The importance of all filters in convolution layer 4 and 5 was calculated,and the filters were sorted by their importance.A lightweight model was obtained after removing the filters with low importance at 75% pruning rate and retraining.The experimental results show that after lightweight processing,the number of theoretical parameters and amount of memory usage are reduced by more than 50%,and the operation speed is increased by more than 20% while the accuracy barely drops.This study provides an idea for the application of CNN model on the portable and mobile ferrography equipment.