The LETBP feature descriptor based fish species classification using Kepler optimization with Extreme Learning Machine

Abstract

Fish species classification plays a crucial role in underwater environments, serving to audit ecological balance, monitor fish populations, and preserve endangered species. However, the interaction of light with ocean water results in scattered and absorbed light, leading to hazy, low-contrast and low resolution images. This, in turn, makes fish classification a challenging and arduous task. Hence, in order to address the issue in this paper an automatic fish classification technique is proposed. To improve the quality of the images a basic CLAHE image enhancement technique is applied. Then, the novel feature descriptor method called local energy triangular binary pattern (LETBP) is proposed to extract features from the images, which effectively extracts the pixel information from all directions. The extracted unique feature values are given to the Extreme Learning Machine (ELM) for the final classification. The ELM network randomly selects the bias and weights and in order to overcome this issue an optimization technique called Kepler Optimization Algorithm (KOA) is adopted. The KOA algorithm tries to improve the search space of exploration and exploitation ratio. The augmented dataset is given to ELM classifier for the classification fish species. The proposed KOA-ELM achieves the high classification rate of 99.23 on fish (F4K) dataset.

Keywords

• feature descriptor
• fish classification
• optimization algorithm
• elm classifier