Detection of sealed and unsealed cracks with complex backgrounds using deep convolutional neural network

摘要: Abstract Crack Deep Network (CrackDN) is proposed in this research with the purpose of detecting sealed and unsealed cracks complex road backgrounds. CrackDN based on Faster Region Convolutional Neural (Fast-RCNN) architecture by embedding a sensitivity detection network parallel to feature extraction (CNN), both which are then connected Proposal Refinement (RPRN) for classification regression. The state-of-the-art aspect lies fusion network, facilitated being able detect sever background. Four kinds background conditions considered crack analysis: normal unbalanced illuminations, markings shadings. raw pavement images first processed simultaneously formed batch line filters that each differ angles sensitive region extraction, CNN, utilized ZF-Net. Then extracted maps applied as input RPRN, thereby prediction scores together bounding box can be obtained. performance compared Faster-RCNN SSD300 architectures detection. Results demonstrate achieve mean average precision higher than 0.90, outperforms SSD300. speed (around 6 fps) slightly lower but significantly Faster-RCNN. Meanwhile, better most conditions. Moreover, difficult However, still obtain accuracy above 0.85.