An Improved YOLOv8 Target Detection Algorithm and Its Application in Tennis Ball Picking Robot

The design and development of a tennis ball pickup robot can greatly improve the efficiency of ball pickup work on tennis courts. Based on this, combined with the related research on small target recognition, we propose an improved YOLOv8 algorithm for tennis ball recognition and detection that can be deployed on embedded devices. First, GAM is introduced into the backbone network to combine global and local features to improve the perception of key information. Secondly, the feature fusion method in the neck network is improved to BiFPN, which effectively solves the problems such as scale change and improves the detection ability of small targets. Finally, the detection head of the network is extended by adding a P2 layer to make the algorithm more suitable for small target detection. In this paper, we created our own tennis ball dataset and designed detailed ablation and comparison experiments to evaluate the detection performance of the model before and after the improvement. The results show that the improved model detects effectively and accurately and outperforms the original model.