ABSTRACT
At present, species identification of freshwater fish mainly relies on manual work in China. The shortage of this form is because of poor working environment, high labor intensity and low efficiency [8,9]. In view of this situation, Zhang [10] proposed a species identification method of freshwater fish based on color
components and ratio of long axis and short axis (called RLS in this paper). They first extracted the blue, green and red components and statistically analyzed the threshold in each color component for all fish species. Then they extracted the contour of the fish body, and calculated RLS, which reflected the proportion of body length and body width. And lastly, the classification was achieved through multiple conditions judgment, and the species identification of silver carp, bream, cyprinoid and crucian was successfully accomplished. In this method, the effect of color component is dependent on the image acquisition hardware. Different acquisition equipments may lead to different color component features. And RLS only takes the overall characteristics of the fish body in account, so, considering the different fish which own little variety in body contours, such as crucian and cyprinoid, RLS cannot obtain ideal identification effect. Wan [11] put forward a species identification method of crucian and cyprinoid based on RLSs of the segment fish body. They divided the fish body into 5 segments in the length direction of freshwater fish after the body contour was extracted. Then the radio of length and averaged width of each body segment were employed to be the features which represent the fish body. Species identification for crucian and cyprinoid was finally executed by using BP neural network. In the method, RLSs of the segment fish body overcome the shortage of RLS of the whole fish
body. But they still consider only the features based on fish contour, and have not fully utilized the more abundant characteristics of the fish surface texture.
