Automated Classification and Identification System for Freshwater Algae Using Convolutional Neural Networks

Abstract

Taxonomy and classification of freshwater algae is highly dependent on morpho-taxonomic characterization and molecular genetic techniques. However, these methods are considered time-consuming and tedious. This study was conducted to integrate the latest technological innovations of digital image processing and machine learning in developing an automated detection, recognition, and identification of selected algal species from the divisions of Chlorophyta and Cyanobacteria. OpenCV and Tensorflow (convolutional neural networks or CNN) were used in the development of a digital image identification system of common freshwater algal species (Chlorococcum infusionum, Chlorella vulgaris, Nostoc commune, Leptolyngbya lagerheimii, Desmodesmus abundans, Acutodesmus dimorphus, Oscillatoria proboscidea, and Oscillatoria limosa). Using OpenCV, digital microalgae images were subjected to image enhancement techniques for the removal of noise and other unwanted objects that minimizes image identification errors. TensorFlow classified these pre-processed images using CNN and gives the percentage results for the algal species in which it identifies each image. The developed automated image identification system correctly identified 75 images from a total of 80 selected freshwater algae images yielding a final test accuracy of 93.75%. This study exhibited for the first time in the Philippines the use of a CNN-based automated image identification system in the recognition and classification of freshwater algae. The developed system is applicable to algal culture collections and taxonomists for fast and easy identification of algal taxa and improved storage of algal images in the database.