An Improved Sheep Counting Detection Method Based on Fusion Allocation Strategy and Multi-Objective Loss Function

Chinais a big country in animal husbandry, and sheep breeding is an important part of animal husbandry. With the scale and modernization of the sheep farm, good management of the sheep shed is an essential factor to ensure the health of the sheep, including counting the sheep number in real time. This paper proposes an object detection method based on fusion allocation strategy and multi-objective loss function for sheep number counting. The fusion allocation strategy selects the pre-selection box based on the cross-grid strategy, and then calculates the minimum allocation loss function to distinguish the positive and negative samples, and the multi-objective loss function changes the confidence loss to VarifocalLoss, which strengthens the detection performance of the model for multi-objective situation. The experimental training was carried out on a single 1080Ti GPU, and the comparative analysis was based on the attention depthwise YOLO detection model. The result shows that the mAP value of the model reached 87.81%, which is superior to the previous model, and proves the feasibility of the model in actual application scenario.