The existing methods for monitoring the cows dietetic behavior do not allow monitoring of multiple cows simultaneously through a single device. A multi-objective cow dietetic behavior identification method was proposed based on the YOLO v3 algorithm. According to the difference in the goals, the cows to be monitored were classified to three groups to achieve dietetic behavior monitoring of multiple cows with a single device. However, the YOLO v3 algorithm has some disadvantages, such as high computational cost, large energy consumption, and strong equipment dependence. So the lottery ticket hypothesis was referred to apply this approach. And an iterative magnitude pruning algorithm for the identification of cow dietetic behavior based on the YOLO v3 network was proposed. Using this approach, the number of parameters was decreased by 87.04%, the mean average precision (mAP) value reached 79.9%, which was increased by 4.2 percentage points. Nevertheless, results proved that through the iterative magnitude pruning technique, the cow behavior monitoring task was feasible at a reduced cost. The effectiveness of screening out the optimal sparse subnetwork from the cow dietetic behavior identification model based on the lottery ticket hypothesis was verified.