A Study on Performance Enhancement Techniques for Object Detection Models in Bird Collision Prevention Systems

As global energy demand continues to rise, wind power has established itself as a key source of clean energy. However, bird collisions near wind turbines pose risks such as turbine malfunction and harm to avian wildlife. To address this issue, bird collision prevention systems have been developed using object detection technology, with the YOLO model being widely adopted due to its effectiveness. Among them, Yolov8n has demonstrated high performance in object detection with fast inference speed, but it requires a large number of parameters and significant computational resources. In this study, we propose the LSCD module to enhance the performance and reduce the complexity of Yolov8n. We further introduce a modified Yolov8n model incorporating both LSCD and MLCA techniques, and evaluate its effectiveness through experiments. The results show that, compared to the original Yolov8n, the proposed model reduces parameters by 28%, decreases inference cost by 25%, and improves accuracy by 4%. The proposed object detection model, when applied to bird collision prevention systems, can contribute to a more balanced advancement between energy development and environmental protection