A novel lightweight blue sheep target real-time detection algorithm

The utilisation of advanced algorithms in detecting blue sheep populations offers significant real-time statistical data for blue sheep protection. However, numerous classical algorithms employ convolution structures as fundamental units for data processing, resulting in a considerable number of redundant channel calculations and a lack of a holistic understanding of the image characteristics. Addressing these challenges, this paper maximises the complementary advantages of convolution and involution. It integrates a learnable weighted splicing structure to construct a lightweight real-time network model, reducing computational complexity to a certain extent and ensuring space efficiency. The model considers both global and local information of the feature map. Experiments demonstrate that the detection speed of this method is 38.3% faster than YOLOv5, showcasing improved realtime performance. Moreover, the AP50 is 0.07 higher than YOLOv5, and the generated model capacity is 85% of YOLOv5.