To address the problems of low optical precision and high false alarm rate in detecting unexploded ordnance (UXO) on the ground in various vegetation environments with drone-based optical detection, a UXO detection method based on the infrared features of unexploded ordnance using YOLOv5 is proposed in this paper. Firstly, the target data of unexploded ordnance is reconstructed, and the ECA attention mechanism is introduced to improve the recognition accuracy. At the same time, the ASPP hole space pyramid pooling is introduced to improve the recognition efficiency, and the CIoU_NMS is used as the prediction box selection criterion. The experimental results show that on the bird's eye view UXO target infrared data set in multiple groups of different vegetation environments, the SAE-YOLOv5 algorithm has an improvement in UXO target precision from 83% to 87%, and the average precision mean is improved from 83.6% to 85%, compared with the original YOLOv5 algorithm model. The algorithm is effective in detecting UXO targets in the four complex backgrounds mentioned in the paper, and with a low false alarm rate.