基于CNN与CRF的桥梁裂缝检测算法

针对实际场景中桥梁裂缝检测精度不高的问题,提出一种基于卷积神经网络与条件随机场的裂缝检测算法。使用特征提取网络对原图进行处理,提取适合裂缝检测的特征;通过区域推荐网络对原始图片中存在裂缝的候选区域进行初步定位;将得到的候选区域作为分类与回归网络的输入,利用条件随机场对该区域的空间特性进行建模,综合判定该区域是否属于裂缝。实验结果表明,该算法相较于常用的Faster-RCNN和滑窗扫描法在查准率上分别提高了9.01%和9.31%,在查全率上分别提高了7.72%和10.45%,精度均值分别提高了0.091和0.175。

Bridge crack detection algorithm based on CNN and CRF

Aiming at the problem of low crack detection accuracy in actual scenes,a crack detection algorithm based on convolutional neural network and conditional random field was proposed.The feature extraction network was used to process the original image and extract features suitable for crack detection.The candidate area of the original picture with cracks was initially located through the regional recommendation network.The candidate region was used as the input of the classification and regression network.The spatial characteristics of the region were modeled using the conditional random field,and it was comprehensively determined whether the region belonged to the crack.Experimental results show that the proposed algorithm improves the precision by 9.01%and 9.31%respectively compared with the commonly used Faster-RCNN and sliding window scanning methods,increases the recall rate by 7.72%and 10.45%,and the average precision is increased by 0.091 and 0.175.