Species classification of individual fish using the support vector machine

Abstract

The fish species classification using echo-sounder is important for fisheries. In the case of fish school of mixed species, it is necessary to classify individual fish species by isolating echoes from multiple fish. A broadband signal, which offered the advantage of high range resolution, was applied to detect individual fish for this purpose. The positions of fish were estimated from the time difference of arrivals by using the split-beam system. The target strength (TS) spectrum of individual fish echo was computed from the isolated echo and the estimated position. In this paper, the Support Vector Machine was introduced to classify fish species by using these TS spectra. In addition, it is well known that the TS spectra are dependent on not only fish species but also fish size. Therefore, it is necessary to classify both fish species and size by using these features. We tried to classify two species and two sizes of schools. Subject species were chub mackerel (Scomber japonicas) and Japanese jack mackerel (Trachurus japonicus). We calculated the classification rates to limit the training data, frequency bandwidth and tilt angles. It was clarified that the best classification rate was 71.6 %. [This research was supported by JST, CREST.]