In-vivo fish sorting by computer vision

Abstract

An image-processing algorithm, applied to images of common carp ( Cyprinus carpio), St. Peter’s fish ( Oreochromis sp.) and grey mullet ( Mugil cephalus), successfully discriminated among the species. Fish images were acquired while they were swimming in an aquarium with their side to the camera. The algorithm 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. One hundred and forty three images (47 grey mullet, 43 St. Peter’s fish and 53 carp images) were acquired and divided into two sets: 20 grey mullet, 20 St. Peter’s fish and 20 carp images in one set and the rest of the images in the other set. Each of these two sets was used as a training set for selection of feature thresholds, which were then applied to the other set as a test case (two-fold cross-validation test). Fish species identification reached 100, 91 and 91% for grey mullet, carp and St. Peter’s fish, respectively. To the best of our knowledge this is the first report on successful discrimination among fish species in vivo. We also report the results of a preliminary experiment, conducted to test the capability of fish to be trained to swim through a narrow Plexiglas channel which could be part of a sorting system, and through which fish images could possibly be acquired.