第 5 期 组合机床与自动化加工技术 No． 5 2022 年 5 ， Abstract Aiming at the difficulty in detecting surface defects of steel plates an improved Faster Ｒ-CNN ， model was used to detect 8 kinds of surface defects of two types of steel plates． Firstly the data were en- ， hanced to get the data set of steel plate surface defects． Secondly three different feature extraction net- ， ， ， ， works VGG16 MobileNet-V2 and ＲesNet-50 were used to train and test the model on the data set and the model accuracy was compared to determine the optimal feature extraction network under the task of this pa- per． Then cluster analysis of the defect data using the K-means algorithm to customize a more suitable an- ， chor scheme for surface defects on steel plates． Finally the feature pyramid network is added to the back- bone network to further
