【摘要】 针对探地雷达图像人工检测速度慢，任务量重，长时间人工识别容易产生疲劳，传统图像手段检测效率低等问题，提出了一种基于改进的YOLOv7的探地雷达土石堤坝空洞检测算法。基于江西省九江市永修县和辽宁省盘锦市盘山县制作了一份当地土质的探地雷达空洞数据集，通过清洗，增强，反演增广了数据集，达到3 000张正负样本标注图像。其次在基础YOLOv7算法的基础上更换了可以自动调整网络的宽度、深度和分辨率的EfficientNet骨干网络；引入了全局注意力机制(GAM)来更好的理解输入数据中的结构，提高了土石方堤坝空洞的检测精度。实验表明，改进的YOLOv7算法在自建数据集上取得了精准度达到了80.23%,平均精度mAp@0.5达到了84.43%,较基础的YOLOv7算法上mAp@0.5提升了7.72%。召回率提升了11.9%,充分展现了算法的可行性，为堤坝空洞隐患快速检测提供了良好的技术支持。更多还原

【Abstract】 In response to the problems of slow manual detection speed, heavy workload, fatigue caused by long-term manual recognition, and low detection efficiency of traditional image methods in ground penetrating radar images, this paper proposes an improved YOLOv7 based ground penetrating radar earth-rock embankment cavity detection algorithm. Based on Yongxiu County, Jiujiang City, Jiangxi Province, and Panshan County, Panjin, Liaoning Province, a local ground penetrating radar cavity dataset was produced. Through cleaning, enhancement, and inversion, the dataset was expanded to 3 000 positive and negative sample annotation images. Secondly, based on the basic YOLOv7 algorithm, the EfficientNet backbone network that can automatically adjust the width, depth and resolution of the network is replaced. The introduction of the global attention mechanism(GAM) to better understand the structure of input data has improved the detection accuracy of cavities in earth-rock embankments. The experiment shows that the improved YOLOv7 algorithm achieved an accuracy of 80.23% on the self built dataset, with an average accuracy of 84.43% for mAp@0.5, which is 7.72% higher than the basic YOLOv7 algorithm. Recall has increased by 11.9%, fully demonstrating the feasibility of the algorithm proposed in this paper, providing good technical support for rapid detection of hidden dangers in dam cavities.更多还原