基于改进YOLOv5的道路病害智能检测

Automatic detection of pavement defect based on improved YOLOv5 algorithm

针对现有道路表观病害检测识别精度低、漏判、误检率高的问题,提出了一种改进的道路表观病害检测高精度识别模型(improved pavement detection-YOLOv5,IPD-YOLOv5)。在YOLOv5的主干特征提取网络中添加由不同空洞卷积组成的ASPP模块,引入SE-Net注意力机制以加强算法从裂缝图像中提取不同尺度特征的能力,实现多尺度特征图的有效融合。结果表明:较传统检测算法,所提的IPD-YOLOv5模型在道路裂缝病害检测上的识别精度最高,其中平均精度比未改进的YOLOv5算法提高了7.47%,漏判率降低了10.29%。

The current pavement defect detection methods suffer from low recognition accuracy,high missing-detection rate,and high false-detection rate.Thus,an improved high-precision recognition model for pavement defect detection(improved pavement detection-YOLOv5,IPD-YOLOv5)was proposed.An ASPP module consisting of various void convolutions was added to the backbone feature extraction network of YOLOv5 algorithm.In addition,the SE-Net attention mechanism was introduced to enhance the ability of algorithm to extract different scale features from crack images and achieve effective fusion of multi-scale feature maps.Results show that the proposed algorithm has the highest detection accuracy for pavement crack defect detection,with an average accuracy improvement of 7.47%and a missing-detection rate reduction of 10.29%compared to the unimproved YOLOv5 algorithm.