Abstract: In this research an improved computer vision detection technology and image fusion method had been proposed so as to create a navel orange surface damage detection system, i.e. firstly, digitized data in detection was collected from image collection card to be processed through image binarization, edge detection and grey stretch.Then the change curve of image grey- scale average was analyzed. Finally, feature image was extracted after weighted filtering so as to improve the precision and speed of navel orange sorting and packaging. The experiment showed that edge feature detection had strong ability in blurred image processing, and that damage location in algorithm design had speeded up the system processing with 10.5 oranges per second, which was characterized by high precision, generality and good stability.