Objective: To solve the problems of poor accuracy and low sorting efficiency of the existing string fruits in robot sorting. Methods: Based on the structure of the high-speed parallel food sorting system, the improved SURF algorithm and the improved RANSAC algorithm were combined to locate the grab points of the food sorting robot, and a three-dimensional grab model was established to realize the automatic and stable grab of string fruits. Taking grapes as an example, the superiority and accuracy of the proposed method were verified by experiments. Results: Compared with the conventional method, the proposed method improved the average capture success rate by about 15.00% and the average capture time by 0.101 s. Conclusion: The food sorting robot can effectively improve the accuracy of the positioning of the holding points of string fruits, and has a high success rate.