In order to improve the detection accuracy of surface imperfections on filamentary materials, multi-scale image analysis method is used for feature extraction and imperfection recognition. Through image preprocessing, feature enhancement and multiscale feature fusion, different types of defects are classified using SVM classifier, and the accuracy rate reaches 97.2%, the recall rate is 95.6%, and the F1 value is 0.964. The experimental results show that the use of multiscale fusion feature method can effectively improve the detection precision and robustness, especially in the performance of the stability under the complex environmental conditions, and the computation time is controlled within 21.8 ms. The detection system has high classification accuracy and localization accuracy for different defect types.