FISH SCHOOL IDENTIFICATION IN THE BALI STRAIT USING ACOUSTIC DESCRIPTOR AND ARTIFICIAL NEURAL NETWORKS TECHNIQUES

Abstract

Accurate fish school identification is one of the crucial pieces of information for fish stock assessment. The fish stock assessment, in turn, is used as a basis for fisheries management action plan. In this paper, we discuss the development and application of acoustic descriptor (AD) and artificial neural networks (ANN) technique for fish school identification. Data (echogram) was obtained from the acoustic survey conducted in November 2000 in the Bali Strait, using SIMRAD EK500 split-beam acoustic system. In this preliminary study, the usage of AD is confined to the geometrical properties (area, perimeter, height, length, elongation, circularity, rectangular, and fractal dimension) of the echogram or acoustic backscattering images, while the ANN used back-propagation technique and a sigmoid activation function to transform the input to output. The results show that the accuracy of identifying fish school for various ANN learning rate value is about 73.3%. We observed that the school of Lemuru Sardinella lemuru, which is dominant in the Strait of Bali during the time of the survey, takes elongated geometrical formation and occupied particular water depth. Future study will incorporate the more complete set of AD, for example by employing the energetic dimension, to improve the accuracy of the school identification.