Deep learning-based ResNeXt model in phycological studies for future

• Published in: Algal research (Amsterdam) 2020-09, Vol.50, p.102018, Article 102018
• Main Authors: Yadav, D.P., Jalal, A.S., Garlapati, Deviram, Hossain, Kaizar, Goyal, Ayush, Pant, Gaurav
• Format: Article
• Language: English
• Subjects: Algal genera; Convolution neural network; Data augmentation; ResNeXt topology
• ISSN: 2211-9264; 2211-9264
• DOI: 10.1016/j.algal.2020.102018

Algae are photosynthetic eukaryotes that may range from unicellular to multicellular forms. Algae have been reported from almost all the ecological systems, including terrestrial, marine, and aquatic ecosystems. The manual classification of algae is a time-consuming method and requires great efforts with expertise due to the numerous families and genera. In the present study, an automated system is developed for the identification and classification of the 16 algal families with a data set of 80,000 images by a modified ResNeXt CNN (Convolution Neural Network) model. Cell differentiation by modified ResNeXt CNN topology is based on cell arrangement and morphological features including area, width, shape, and length of the cell. An experimental result of 99.97% classification accuracy demonstrates the effectiveness of the proposed method. The present investigation may open a new path in the future for the development of a time and a cost-effective, highly sensitive computer-based system for the identification and classification of different algae. •Conventional algal identification techniques are time and cost consuming.•Automated computer-based recognition approaches are on the cutting edge.•Highly sensitive algorithms in phycological studies are future necessity.