Multi-level residual network VGGNet for fish species classification

The development of an image-based fish classification system using Convolutional Neural Network (CNN) has the advantages of no longer directly conducting features extraction and several features analysis. These steps has been involved by cascading convolution from initial to final block, where the initial, middle, and final blocks produce low-, middle-, and high-level features, respectively. Due to cascading convolution, CNN produces only high-level features. However, fish classification requires not only high-level features but also low-level features such as points, lines, and textures for representing edge spines, gill covers, fins, and skin textures in order to achieve higher performance; furthermore, CNN generally has not yet incorporated low-level features in the last block. In this paper, we proposed Multi-Level Residual (MLR) as a new residual network strategy by combining low-level features of the initial block with high-level features of the last block by applying Depthwise Separable Convolution. We also proposed MLR-VGGNet as a new CNN architecture inherited from VGGNet and strengthened it using Asymmetric Convolution, MLR, Batch Normalization, and Residual features. Our experimental results show that MLR-VGGNet achieved an accuracy of 99.69%, outperformed original VGGNet relative up to 10.33% and other CNN models relative up to 5.24% on Fish-gres and Fish4-Knowledge dataset.(c) 2021 The Authors. Published by Elsevier B.V. on behalf of King Saud University. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

Automated summary

This paper proposes an image-based fish classification system using a Convolutional Neural Network (CNN) that combines low-level and high-level features using a multi-level residual network strategy, and introduces new techniques in the CNN architecture to improve performance. Experimental results show that the system performs well on fish image datasets.