In order to detect litchi fruit under nature scenes, an identification method was studied. A simple process was proposed to segment the image and remove noises. Its key steps include image converting; segmenting by Two-Cluster K-means clustering; labeling by white and black; small noises removing by open-close combined morphology; and big noises removing by blob area thresholding. A new contribution has been made to weighing by ultrasonic sensor and machine vision which allows for identifying litchi weight under nature scenes. A max error of 27% was recorded in tests of such weighing method.