Identification of Various Osteichthyes Class of Fishes Through Gaussian Mixture Model, Kalman Filter, Pyramid Histogram of Visual Words and Support Vector Machine

Abstract

This paper proposes a prototype that can identify the species in the Osteichthyes class of fishes. This paper is mainly focused on five species, namely Glossogobius celebius (Celebes goby), Trichopodus trichopterus (Three-spot Gourami), Poe'cilia Latipinna (Sailfin Molly), Oreochromis niloticus (Nile Tilapia), and Clarias batrachus (Philippine catfish). The processes involved in the identification included detection, tracking, and identifying and classifying the fishes' respective species. Detection is performed using the GMM, which is based on background subtraction method. Tracking, on the other hand, is performed using the Kalman Filter. Identification of the species of the fish can be attained using the Pyramid Histogram of Visual Words (PHOW) and lastly, the classification process utilizes the Support Vector Machine (SVM). These four algorithms were utilized on the raw data collected to obtain the desired results. The prototype underwent controlled testing to perform initial tests and calibrations, and then went through uncontrolled testing at the Mt. Makiling Forest Reserve located at UPLB, Laguna, Philippines using the aid of the Philippine Journal of Science's paper, Freshwater Fish Fauna in Watersheds of Mt. Makiling Forest Reserve, Laguna Philippines. The proposed system can detect and track fishes with 57.5% accuracy in videos and identify the species with 92.5% accuracy in high-resolution still images. As for videos, the system can perform with 72.5% accuracy in identifying the species of the fishes at low quality videos.