An improved algorithm for regional cattle population counting under local optimal methods

As a movable collateral, live cattle need to be accurately counted and recorded on a regular basis, such as their breeds, numbers and other important information. Aiming at the problems that the existing cattle number detection algorithms have leakage detection phenomenon for dense and small targets and have many parameters and large computation volume, which are not suitable to be deployed at the end of embedded devices, a CBG-YOLOv5s cattle number detection algorithm is proposed. The experimental results show that the improved model reduces the number of parameters and computation by 37.32% and 43.04%, respectively, compared with the original model YOLOv5s, and still maintains good detection accuracy. Finally, the lightened model was successfully deployed to Nvidia Jetson Nano mobile and verified by actually running on the development board. The model in this paper is able to ensure the accuracy of cattle identification, which can provide a reference for the accuracy of detecting the number of cattle in complex farming environments and under severe occlusion conditions.