Research on anti-smoke interference target recognition algorithm based on YOLOV5

To enhance target recognition in complex environments, an anti-smoke interference module based on the SDNet dehazing algorithm is designed, using NIQE for image quality assessment. The module improves image quality after antiinterference and is integrated into a serial network with YOLOv5. For faster recognition, the backbone is replaced with the MobileNetV3_small feature extractor for network lightweighting, resulting in the MSD-YOLOV5s model. The MSDYOLOV5s achieves a 3% higher recognition confidence under smoke interference, with an mAP of 0.908 and 130 FPS, meeting real-time, high-precision target recognition requirements in smoke-covered battlefield environments.