【摘要】 为提升沥青路面病害自动化识别的准确率，提出一种特征网络增强算法(YOLO-EH)。该网络包含一种可以与CBAM注意力机制进行结合的新型特征增强模块(FEM)以及一种可以对FPN添加反馈链接的新型逆向二次循环特征金字塔网络(RCFPN)。实验结果表明，与原YOLOv5算法相比，YOLO-EH对于同一批路段数据在平均病害识别准确率上提高了2.6个百分点，验证了其准确性与有效性。更多还原

【Abstract】 To improve the accuracy of automatic identification of asphalt pavement diseases, a feature network enhancement algorithm(YOLO-EH) was proposed. The network included a new feature enhancement module(FEM) that could be combined with CBAM attention mechanism and a new reverse quadratic cyclic feature pyramid network(RCFPN) that could add feedback links to FPN. Experimental results show that compared with the original YOLOv5 algorithm, YOLO-EH improves the average disease recognition accuracy by 2.6 percentage points for the same batch of road data, which verifies the accuracy and effectiveness of the algorithm.更多还原