Detection of Potential Breeding Sites Based on AAV Remote Sensing Imagery and MultiDCCSP-YOLO Network

Detection and control of potential breeding sites (PBS) for Aedes albopictus are considered among the most effective methods for managing mosquito-borne diseases. Due to the high adaptability and rapid reproduction of mosquitoes, PBS are widely distributed. Traditional manual investigation methods are inefficient and would increase the risk of exposure. Autonomous Aerial Vehicle (AAV) remote sensing technology, with its speed and broad coverage, can capture large-scale images quickly. However, current PBS detection methods primarily focus on the detection of large, regular containers. To address the issue for detecting diverse PBS in complex environments, we propose a method called MultiDCCSP-YOLO. This approach integrates several modules to improve the PBS detection accuracy from AAV remote sensing images with challenging backgrounds. We introduce the Multi-DC Block and use it to build the CSPDCStage module. This module captures multi-scale features while reducing computational demands without compromising the detection accuracy. The FasterELAN module minimizes overfitting risks and enhances the detection of PBS targets with limited-sample classes. Additionally, we incorporate the Adaptive Threshold Focal Loss (ATFL) function to emphasize small, hard-to-detect targets, thereby improving multi-scale PBS detection performance. The MultiDCCSP-YOLO method enables efficient and accurate PBS detection in complex environments, facilitating large-scale monitoring efforts.