YOLO-PAD: A Lightweight Network Based on YOLO for Infrared Anti-UAV Detection

In the field of anti-UAV identification, infrared detection technology is important for timely localization and detection of UAVs. However, since infrared UAV targets are often in environmental conditions with small pixel occupancy, low contrast and complex backgrounds, it poses a great challenge to the detection and localization tasks of UAV targets. Therefore, to address these issues, we propose a YOLO-PAD model based on YOLOv8n. First, We introduce the convolutional SPD-Conv to replace the step-length convolution and pooling layer in the YOLOv8 algorithm to increase the depth of the feature map, which ensures the original spatial information of the image and improves the recognition performance of long-range infrared UAV targets. Secondly, we constructed a C2f-PPA network module instead of the original C2f module using the PPA (Parallelized Patch Perception Attention) module to further improve the feature extraction depth. Finally, the DyHead (Dynamic Head) module is introduced to fuse scale, space and task perception into a unified framework. The effective application of the attention mechanism enhances the performance of the model in recognizing infrared UAVs. The experiments showed that the accuracy, recall, mAP50mAP50-95 and were improved by 1.1%, 8.6%, 4.2% and 5.9%, respectively, compared to the traditional method.