<p>Accurate identification of pika holes and bare patches in alpine meadow ecosystems is essential for ecological monitoring and degradation assessment. However, these degradation features are typically characterized by small object scales, irregular spatial distributions, and frequent overlaps, further compounded by complex backgrounds and the limited efficiency of traditional detection methods. To address these challenges, this study proposes an ST-YOLOv11–based small-object detection framework tailored for UAV imagery of alpine meadows. First, to enhance the extraction of subtle features from pika holes and bare patches, we design an attention module, C3K2_SHSA, which adaptively adjusts feature weights according to the spatial distribution of the input image, thereby strengthening discriminative small-object representations. Second, to mitigate edge information loss during the upsampling process, we introduce a Spatially Adaptive Feature Modulation (SAFM) module that integrates shallow detail features with deep semantic cues, enabling high-resolution preservation of object boundaries. In addition, to improve class separability under complex backgrounds, a high-resolution detection branch is constructed to perform small-object detection on finer-grained feature maps, effectively enhancing the model's ability to capture small-scale targets. Ablation and comparative experiments on the PB dataset demonstrate that the proposed ST-YOLOv11 model achieves improvements of 6.5%, 12.4%, and 9.9% in overall precision, recall, and mAP@0.5, respectively. Specifically, the detection precision for pika holes and bare patches increases by 9.5% and 3.5%, validating the effectiveness and robustness of the proposed method in complex natural environments.</p>