Performance evaluation of hybrid CNN for SIPPER plankton image calssification

Abstract

Plankton are a diverse group of organisms that live in large bodies of water. They are a major souce of food for fishes and other larger aquatic organisms. The distribution of plankton plays an important role in the marine ecosystem. The study of plankton distribution relies heavily on classification of plankton images taken by underwater imaging systems. Since plankton are very different in terms of both size and shape; plankton image classification poses a significant challenge. In this paper we proposed the use of hybrid classification algorithms based on convolutional neural networks (CNN). In particular, we provide an in depth comparison of the experimental results of CNN with Support Vector Machine and CNN with Random Forest. Unlike traditional image classification techniques these hybrid CNN based approaches do not rely on features engineering and can be efficiently scaled up to include new classes. Our experimental results on the SIPPER dataset show improvement in classification accuracy over the state of the art approaches.