Hierarchical Classification for Live Fish Recognition

Live fish recognition in the open sea is a challenging multi-class classification task. We propose a hierarchical classification approach to recognize live fish from underwater videos. However, the hierarchical method accumulates misclassified samples into deeper layers and these accumulated errors reduce the average accuracy. We propose a set of heuristics to help construct more accurate hierarchical trees and, therefore, control the error accumulation. We create an automatically generated tree based on these heuristics and compare it to a baseline tree on a live fish image dataset. The proposed hierarchical classification method achieves about 4% better accuracy compared to state-of-the-art techniques.