Target Dectection for UAV Infrared Images based on Improved YOLOv8

UAV-captured aerial images play a pivotal role across multiple domains, encompassing environmental monitoring, precision agriculture, traffic control, and disaster response. However, flight altitude and angle variations pose challenges related to multi-scale transformations, particularly hindering small target detection in infrared imagery due to pixel scarcity and subtle features. To address these issues, we present the improved YOLOv8 detection algorithm tailored for UAV infrared images. First, we design the DCN-C2f module to effectively handle geometric distortions and enhance fine-grained feature extraction for small objects. Then, we employ the ST-Det module to refine small target detection, substantially boosting accuracy. Additionally, we adopt the WIoU loss function, which not only mitigates gradient degradation from low-quality data but also accelerates network convergence. Extensive experiments on the VEDAI dataset demonstrate the algorithm's exceptional detection precision and robustness in UAV infrared image target detection, surpassing conventional approaches and the latest deep learning methods.