Sorting fish by computer vision

Abstract

Abstract In fresh-water fish farms which grow a few fish species together in a pond (polyculture fish farming), it is necessary to sort harvested fish according to species and size for optimal marketing. An image processing algorithm for discrimination between images of three fish species had been developed and tested. It was based on the method of moment-invariants (MI) coupled with geometrical considerations and was therefore insensitive to fish size, two-dimensional orientation and location in the camera’s field of view. The algorithm was applied to images of carp ( Cyprinus carpio ), St. Peter's fish ( Oreochromis sp. ) and grey mullet ( Mugil cephalus ), grabbed by a CCD camera under different lighting conditions in a lighting chamber. Three separate imaging sessions were conducted: (1) 96 images of 16 fish were acquired while they were placed individually in the illumination chamber at six different positions; (2) 140 images of 35 fish at four different positions each; (3) 146 images of 73 fish at two different orientations each. The three sessions were conducted under different lighting conditions and the fish were received from different farms. Based on the MI of their whole body, fish species identification reached 100, 94 and 86%, respectively, for grey mullet, carp and St. Peter’s fish for the first set of images and 98, 96, and 100%, respectively, for the second set. Using the shape of the whole body to identify grey mullet and the shape of the tail to differentiate between carp and St. Peter's fish, 100, 89 and 92% correct classification was achieved, respectively. Fish mass can be closely estimated from their image area. The correlation coefficients between the mass and image area of grey mullet, carp and St. Peter’s fish were 0.954, 0.986, 0.986, respectively. Manually measured fish length also correlated well with fish length when calculated from their binary image (correlation coefficients of 0.950, 0.997 and 0.983 for grey mullet, carp and St. Peter’s fish, respectively).