关键词: 裂痕检测, 支持向量机（SVM）, 基于区域的全卷积网络（R-FCN）, 残差网络（ResNet）, 感受野

Abstract: To solve the problem of the detection accuracy and recognition using traditional machine learning method to detect texture with transparent plastic crack rate is not high, an improved detection method based on Region-based Fully Convolutional Networks（R-FCN） is proposed. It makes up for the shortcomings of the original network’s low sensitivity to tiny cracks by using mixed-scale receptive field fusion procession in Residual Network（ResNet）. Experimental results show that the crack detection accuracy based on improved R-FCN is about 20% higher than that based on Support Vector Machine（SVM）, and is about 8% higher than that based on R-FCN without improvement. The validity of the method is proved.

Key words: crack detection, Support Vector Machine（SVM）, Region-based Fully Convolutional Networks（R-FCN）, Residual Network（ResNet）, receptive field