Infrared small target detection algorithm improved based on YOLOv8

To address the challenges of small imaging area, low resolution, occlusion susceptibility, and detection inaccuracy in infrared small target recognition, this paper proposes an SSE-YOLO-based infrared small target detection algorithm. The key innovations include: (1) Introducing a deep non-strided convolution module into YOLOv8s to preserve fine-grained details and enhance feature learning efficiency; (2) Adding a dedicated detection layer during feature extraction to improve small target capture capability; (3) Designing an Efficient Dual-Attention Mechanism (EDAM) to adaptively learn channel and spatial importance, thereby emphasizing critical image regions; (4) Adopting the Shape_IoU loss function to refine bounding box regression by focusing on shape and scale constraints. Extensive experiments on the FLIR dataset and a proprietary dataset from Iraytek demonstrate that the proposed method achieves mean average precision (mAP) scores of 89.8% and 92.1%, outperforming the baseline model by 3.3% and 2.9%, respectively. These results validate its effectiveness in improving detection accuracy and robustness for infrared small targets under complex scenarios.