A Method for Target Locking in Combat Process Based on the Enhanced YOLOv8

Rapidly locking onto combat targets during the bat-tle process is an important capability for accomplishing combat missions in modern warfare. The YOLO series of algorithms can effectively meet the needs for this capability during combat operations. However, in military missions, there are complex battlefield environments, and the YOLO series of algorithms has issues with missed and false detections during night combat operations. To address this issue, this paper proposes a target locking method during combat based on an improved YOLOv8, the improved YOLOv8 model is referred to as KAN-YOLOv8n. In this method, the original Conv layer in the C2F module of the YOLOv8 algorithm is replaced with a KolmogorovArnold Networks (KAN) convolutional layer, creating a new C2F-KAN module. The KAN convolutional layer has the characteristic of a learnable nonlinear activation function, which can extract the features of targets under complex conditions with higher accuracy and speed after multiple iterations. The experimental results on the Thermal Tank Images dataset show that the method proposed in this paper can effectively improve the accuracy of target detection, with an mAP50 of 65.8%, which is 8.7%, 14.9%, and 14% higher than that of YOLOv6n, YOLOv8n-world, and YOLOv9c, respectively.