ResNeXt convolution neural network topology-based deep learning model for identification and classification of Pediastrum

Abstract

For identification of different Pediastrum species in a sample, the determination of microscopic feature and colony morphology are the preliminary steps before sending them to the higher genomic and proteomic level. Great efforts with high expertise are required for the time-consuming manual process. In the present study, the first time an effort has been done to address the problem for identification and classification of Pediastrum species with the help of convolutional neural networks (CNNs). The modified ResNeXt CNN (Convolution Neural Network) model is used for training and validation of the data set consisting of 42,000 algal images. Modified ResNeXt CNN topology differentiates cells based on the formation of coenobia, cell arrangement and feature and particularly the sculptures on the outer sporopollenin cell-wall layers. An experimental result of 98.45% classification accuracy and F1-score more than 0.98 demonstrates the effectiveness of the proposed method. In the future, such time and cost-effective facilities can be used as promising sources for phycological studies.