A Hybrid Memetic Algorithm (Genetic Algorithm and Tabu Local Search) with Back-Propagation Classifier for Fish Recognition

Abstract

In this research our goal is to develop a hybrid approach for optimizing and enhancing back- propagation classifier (BPC) performance using a Memetic Algorithm (Genetic algorithm and Tabu Local Search), thus; Memetic Algorithm used to tune the parameters of the BPC. The proposed hybrid approach (HGATS-BPC) was tested based on fish images recognition. To recognize the pattern of interest (fish object) in the image based on extracted features from color signature. Histogram technique and Gray Level Co-occurrence Matrix (GLCM) used to extract 20 features from fish images based on color signature. The general BPC has several issues to be optimized by Memetic Algorithm such as speed, and easy for running into local minimum. In our study we used 800 fish images for 20 different fish families; each family has a different number of fish types. These images are divided into two datasets: 560 training images and 240 testing images. HGATS-BPC successfully optimized and enhanced the performance of the BPC in term of accuracy and outperformed the traditional BPC and previous methodologies by obtaining more accurate results but with a high cost of computational time compared to the BPC. The overall accuracy obtained by BPC was 85%, while the HGATS-BPC obtained 94% based on 800 fish images. Finally; the HGATS-BPC classifier is able to classify the given fish images into poisonous or non-poisonous fish and classify the poisonous and non-poisonous fish into its family.