CDS-YOLO: a real-time infrared small target detection model

In contemporary computer vision, You Only Look Once (YOLO) stands as a significant advancement in object detection. Despite its outstanding performance in real-time monitoring and recognition across various applications, YOLO encounters limitations in detecting small infrared targets. Traditional YOLO architectures struggle with capturing intricate details, accommodating the unique properties of infrared imagery, and optimizing the detection of small-scale objects. To address these challenges, we propose a real-time infrared small target detection model integrating a Content-Aware Cross-Channel Feature Pyramid Network (CACCFPN), Dual-Channel Downsampling Convolution (DCDConv), and Shape Intersection over Union (ShapeIoU), specifically designed for real-time monitoring in challenging environments. Experimental results demonstrate that the proposed model achieves a mean Average Precision (mAP) of 93.6% on the Infrared Small Aircraft Target Detection (ISA-TD) dataset using only 5.8M and 11.9 GFLOPs, and an mAP of 94.9% on the Infrared Small Aircraft Detection (ISAD-6) dataset. Compared to YOLOv8, our model improves mAP by 2.1% on ISA-TD and 1.6% on ISAD-6, respectively, and outperforms the Detection Transformer (DETR) by 5.0% on ISA-TD. The model enhanced frame rate further supports its applicability for detecting and tracking small, low-altitude aircraft targets in infrared imagery.