Lightweight sensitive target identification method based on network pruning and transfer learning

The invention discloses a lightweight sensitive target identification method based on network pruning and transfer learning, and the method comprises the steps: 1, collecting an actual image set of a related battlefield, and carrying out the cross-domain amplification through a cyclic adversarial network, and obtaining a target image set; 2, training the YOLO-F by taking the target image set as input until a loss function of the YOLO-F is converged, obtaining a trained YOLO-F, and realizing target recognition by utilizing the trained YOLO-F; according to the method, when battlefield target detection is carried out through embedded equipment in related technologies, under the condition that the precision tends to a stable interval, the model size and the operation frequency are greatly reduced, and the method is suitable for light-weight platforms and recognition scenes with relatively harsh operation requirements. 本发明公开了一种基于网络剪枝与迁移学习的轻量化敏感目标识别方法，包括：步骤1：采集相关战场实际图像集，并利用循环对抗网络进行跨域扩增得到目标图像集；步骤2：将目标图像集作为输入对YOLO-F进