SensiYOLO: YOLO-based Detection Method for Solid Waste Stockpiles in Remote Sensing Images

The intelligent detection of solid waste stockpiles plays a critical role in environmental management, particularly against the backdrop of rapid urbanization and industrialization, which have led to a continuous increase in solid waste generation. Efficient and accurate monitoring of these stockpiles has become an urgent necessity. Although remote sensing technology has been widely adopted for such tasks due to its wide coverage and cost-effectiveness, existing detection methods still face significant limitations in terms of feature representation and adaptability to complex environments. To address these challenges, this study proposes a novel object detection algorithm, SensiYOLO, built upon the YOLOv11 framework and enhanced by two key modules: the Dynamic Spatial-Channel Attention (DSCA) module and the Adaptive Context Perception (ACP) module. The DSCA module leverages multi-head attention to dynamically enhance feature responses across spatial and channel dimensions, thereby improving feature discrimination. Meanwhile, the ACP module combines depthwise separable convolution with strip pooling to achieve lightweight, multi-scale context fusion, effectively capturing semantic information with minimal parameter overhead. Without significantly increasing model complexity, the proposed method demonstrates improved capability in recognizing target information under complex conditions. Experimental results on the Solid Waste Stockpiles (SWS) dataset show that SensiYOLO outperforms the baseline YOLOv11 model, achieving increases of 3.2% in precision, 3.7% in recall, and 3.1% in mAP. The synergistic design of the DSCA and ACP modules enhances feature representation while maintaining computational efficiency, significantly improving the model’s applicability in large-scale and high-precision detection scenarios. This work offers a practical deep learning-based solution to the technical challenges of solid waste stockpile monitoring in complex environments, with strong potential for scalability and real-world deployment in environmental governance.