Implementation of Genetic Programming toward the improvement of acoustic classification performance for different seafloor habitats

Classification of Seafloor Habitats Using Genetic Programming

“…As in [10], the GP system used was an adaptation of GPLAB, a GP toolbox for MATLAB [14], with the main running parameters indicated in Table 1.…”

“…The function set used was shorter than in [10], containing only the most basic operators. As in [10], we have used the Lexicographic Parsimony Pressure tournament [16] and the Heavy Dynamic Limit [15] on tree depth to avoid excessive code growth, but without using any traditional static limit.…”

“…After studying the limitations of the system presented in [10], we have developed a new fitness function that allows a better learning of the feature combinations necessary to differentiate the different classes involved. The previous fitness function [10] was based on inter-class and intra-class distances, inspired by the fitness function on [11]. But it lacked the full objectivity needed to guide GP through a difficult error landscape.…”

“…One of them [9] has been the target for improvement in a work using GP [10], where GP was able to provide an improved discrimination between the different seafloor habitats. The initial motivation to use GP for this task came from a work on diesel engine diagnosis [11].…”

“…To copy otherwise, to republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee. Here we tackle the same problem studied in [10], developing a new fitness function for the GP system, testing the pairwise separability of the different classes involved in the study, and ultimately dividing the 5-class problem into several 2-class problems, whose solutions can be joined to provide a perfect discrimination of most of the seafloor habitats.…”