Application of Faster‐Rcnn based on Resnet50 in Appearance Inspection of Industrial Products

Ao Duan ； Yuefan Cong

《International Core Journal of Engineering》 8卷1期 (2022/01) Pp. 102-113

https://doi.org/10.6919/ICJE.202201_8(1).0015

摘要

Aiming at the problems of low recognition rate of surface defects of industrial products by traditional detection algorithms and inaccurate localization of small defects, an improved Faster RCNN deep learning network was proposed to detect defects. Firstly, after data enhancement, the traditional Faster RCNN feature extraction network is improved to make the network layer deeper to enhance the feature extraction capability of small defects. Then, the ROI Align algorithm is used to replace the rough ROI Pooling algorithm to obtain more accurate defect location information and obtain anchor frames that are more suitable for defects. Experimental results show that the recognition effect of the improved network on surface defect detection is up to 99%, which is more than 10% higher than the original Faster RCNN network, and the detection ability of small defects is also significantly improved.

關鍵字

Deep Learning ； Target Detection ； Convolutional Neural Network ； Faster RCNN ； Resnet50

參考文獻

PERNKOPF F. Detection of surface defects on raw steel blocks using Bayesian network classifiers [J]. PatternAnalysis& Applications,2004, 7(3) : 333-342.

ZHOU P, XU K, LIU SH H. Metal surface defectrecognition method based on the feature fusion of shearwave and wavelet [J]. Chinese Journal of MechanicalEngineering, 2015 ,51(6) : 98-103.

WANG T, CHEN Y , QIAO M, et al. A fast and robust convolutional neural network-based defect detection modelin product quality control [J]. International Journal ofAdvanced Manufacturing Technology, 2017, 94 (9) :3465-3471.

MEI S, YANG H, YIN Z. An unsupervised-leaming-based approach for automated defect inspection ontextured surfaces[ J ]. IEEE Transactions onInstrumentation and Measurement,2018: 1266-1277.

Wang P, Liu R, Xin X J, Liu PD. Multiscale Feature Fusion-Based Object Detection Algorithm[J/OL]. Laser & Optoclectronics Progress,2021, 58(2): 0210001.