Recognition Method of Concrete Surface Crack Detection in Complex Scenes Based on Deep Neural Network

  • Quanwei Zhao, Jincheng Liu, Leqiang Lian, Lanyun Chen


In view of the low accuracy and poor efficiency of traditional methods for detecting Concrete surface defects, which cannot meet the high-efficiency real-time requirements, a deep learning-based Concrete surface defect detection and recognition method is proposed. The YOLOv3 target detection method is applied to concrete surface defect detection and identification, and the model size and number of parameters are reduced without significantly sacrificing detection accuracy. First, this paper proposes a separable residual module based on deep separable convolution and residual network. And a network with shallower network layers and fewer channels is designed for rapid detection and identification of Concrete surface defects. Finally, the algorithm of this paper was evaluated on SD Data. The experimental results show that this method has a smaller model size, a smaller number of parameters, and a faster running time than the YOLOv3 network, while maintaining good accuracy, and compared with The detection method of Concrete surface defects at this stage has a certain improvement in accuracy, which can be better applied to the real-time detection and identification of Concrete surface defects.