Based on computer vision technology, the data of beef cattles’ feeding behavior can be obtained with the help of the existing system, and carrying out the correlation analysis with weight change, health status, etc. is of great significance for scientific beef breeding. A method of beef cattles’ feeding behavior recognition based on machine vision was proposed. YOLOv3 was used to detect the beef cattle targets in the observation range, and convolutional neural network was used to recognize the feeding behavior of single target, and then the recognition of feeding behavior of multitarget beef cattle was realized. In convolution operation, padding was used to enhance the network’s ability to extract the edge features of the target; the corrected linear units (ReLU) was used as the activation function to prevent the gradient from disappearing; the dropout method was used to improve the generalization ability of the network. Taking the actual beef cattle farm monitoring video as the research object, the experiment was carried out on eight test sets. The average precision of beef cattle target detection within the observation range was 83.8%, the average precision of feeding behavior recognition was 79.7%, the average recall rate was 73.0%, and the average accuracy rate was 74.3%, which can meet the monitoring of beef cattle feeding behavior. The multiobjective recognition method of beef cattle feeding behavior based on YOLOv3 and convolutional neural network had good accuracy, and provided a new way for noncontact monitoring of beef cattle behavior.