An improved lightweight UAV small target detection algorithm based on YOLOv8s

UAV aerial images pose many challenges to target detection tasks because of the complexity of the context and the small size of the target. In order to overcome these issues, an improved UAV image detection algorithm is proposed in this paper. Firstly, considering the insufficient detection capability, the poor background complexity handling, and the sensitivity to scale differences and occlusion problems of the YOLOv8s model for small target detection, we propose a modified target detection head to enhance the perception of small targets. Secondly, we propose the GL-LSK attention mechanism to obtain the context information through an expanded receptive field, thereby enhancing the detection accuracy. Thirdly, we design the Inner-Wise-CIoU loss to effectively reduce the deviation of the predicted boxes, resulting improvement of the detection accuracy. The experimental results demonstrate that our algorithm not only reduces the parameters, but also improves the mAP0.5 of the validation dataset by 6.6%. The detection performance of the proposed algorithm outperforms the current SOTA, demonstrating the advantages of our proposed model in UAV detection tasks. Therefore, it is concluded that the proposed lightweight model is more suitable to deal with the UAV detection problem.