Objective: To solve the problems of poor accuracy and low efficiency in target recognition methods for existing sorting robots in food production lines. Methods: On the basis of the analysis of the binocular vision food sorting system, a combination of improved particle swarm optimization algorithm and support vector machine was proposed for target recognition of food sorting robots. By improving the particle swarm optimization algorithm to optimize support vector machine parameters, an optimized support vector machine classification model was obtained. The classifier was trained for both global and local features, and feature weight coefficients were dynamically assigned to obtain the best recognition rate. Analyzed the performance of the proposed method through experiments, and verified its feasibility. Results: Compared with conventional methods, the proposed method had high recognition accuracy and efficiency in target recognition of food sorting robots, with an accuracy rate of 99.50% and an average recognition time of 0.048 s, which meet the needs of robot sorting. Conclusion: The proposed method can effectively identify canning, improved sorting accuracy and efficiency of sorting robots.