An object in the field environment usually contains two or more classes of color targets. Fast recognition of color target image is the key technology for robot positioning and operation, which is widely used in the field of military, natural disaster rescue, agricultural harvesting robot, etc. However the speed of multi-target recognition in the field environment is usually slow, which makes the visual positioning precision of robots lower at present. This paper proposed a double Otsu segmentation method based on the improved Otsu algorithm for the recognition of multiple targets. To prove the effectiveness of this method, it was used on mature litchi recognition in the field environment. First of all, in order to improve the efficiency, the traditional Otsu algorithm was improved. Then the background, stem and fruit of the target color image were respectively recognized by using the improved Otsu algorithm. Compared with the Kmeans clustering (K-means) algorithm, the fuzzy C-mean clustering (FCM) algorithm, the Otsu and K-means algorithm, and the Otsu and FCM algorithm, the double Otsu segmentation algorithm was superior to the other four algorithms on the segmentation quality and correctness rate, the running time and stability. The test results showed that the recognition time for the mature litchi by using the double Otsu segmentation algorithm was less than 0.2 s. The effectiveness of the algorithm was verified through the experiment.