Abstract:  Mouse hole density is an important indicator of grassland degradation and one criterion for assessing the severity of rodent damage. To address the laborintensive and inefficient nature of traditional mouse hole detection methods, this paper proposes a mouse hole detection method based on Yolov8n+ and UAV images. First, the UAV equipped with a visible light sensor collects RGB images, which are annotated to create a mouse hole dataset. Then, the scale intervals of the mouse holes are analyzed to optimize the YOLOv8n network structure by eliminating redundant detection heads and streamlining the network, while retaining the information interaction channels between deep and shallow layers. Finally, the object detection evaluation metrics are used to assess both detection accuracy and model complexity. The experimental results show that YOLOv8n+ achieves an average detection accuracy for mouse holes of 98.6%, with the model size of 3.2MB and the detection speed of 172.42FPS. Compared with YOLOv8n and other mainstream algorithms, YOLOv8n+ outperforms both in terms of detection accuracy and speed, proving its capability for accurate and efficient detection of mouse holes. This method satisfies the requirements for lowaltitude UAV remote sensing detection in real scenarios, makes up for the shortcomings of traditional methods, and improves the realtime monitoring and flexibility of mouse hole detection.

Key words: desert grassland, mouse hole, UAV, object detection, rodent damage monitoring

摘要： 鼠洞密度是草原退化的重要参考指标，也是鼠害程度的标准之一。针对传统鼠洞检测劳动强度大、工作效率低的问题，提出基于Yolov8n+和无人机图像的鼠洞检测方法。首先，无人机搭载可见光传感器采集RGB图像，对图像进行标注制作鼠洞数据集。然后，统计鼠洞尺度区间优化YOLOv8n网络结构：剔除冗余的检测头，精简网络结构；保留深层与浅层之间的信息交互通道。最后，采用目标检测评价指标从检测精度和模型复杂度两个方面对YOLOv8n+进行评估。结果表明：YOLOv8n+对鼠洞的平均检测精度为98.6%，模型体积为3.2MB，检测速度为172.42FPS。相比于YOLOv8n和其他主流算法，YOLOv8n+在检测精度和检测速度上表现最好，表明YOLOv8n+能够实现对鼠洞精确、高效的检测，满足现实场景下无人机低空遥感对鼠洞的检测要求，弥补传统方法的不足，提高鼠洞监测的实时性和灵活性。

关键词: 荒漠草原, 鼠洞, 无人机, 目标检测, 鼠害监测