Species Classification by the Support Vector Machine Using the Broadband Split-Beam

Species classification using an acoustic sounder is important for fisheries. With schools of mixed species, it is necessary to be able to classify individual fish species from echoes, and to isolate individual fish echoes. A broadband signal, which offered the advantage of high-range resolution, was applied for this purpose, and the positions of fish were estimated using the split-beam system. The target strength (TS) spectrum of individual fish echoes was computed from isolated echoes and estimated positions. In this paper, these TS spectra were used as features of fish classification for machine learning. Also, it is well known that the TS spectra are dependent on not only fish species but on fish size. Therefore, it is necessary to classify both fish species and size using these features. We attempted to classify two species and two fish sizes using the Support Vector Machine (SVM) and Nearest Neighbor Algorithm (NNA) as machine learning. Subject species were chub mackerel (Scomber japonicas) and Japanese jack mackerel (Trachurus japonicus). The classification rates using the SVM were superior to those using the NNA. These rates were dependent on the frequency bandwidth and tilt angle. The classification rate was 71.6% with limitation of the tilt angles.