Title: Convolutional neural network optimisation for discovering plant leaf diseases with particle swarm optimiser

Authors: Vishakha A. Metre; Sudhir D. Sawarkar

Addresses: Department of Computer Engineering, Datta Meghe College of Engineering, Airoli, Navi Mumbai, Maharashtra, India; Department of Computer Engineering, D. Y. Patil College of Engineering, Akurdi, Pune, Maharashtra, India ' Department of Computer Engineering, Datta Meghe College of Engineering, Airoli, Navi Mumbai, Maharashtra, India

Abstract: The agriculture industry contributes most to expanding economies and populations, but plant diseases restrict the food production. Utilising an automatic detection method, early diagnosis of plant diseases can improve food production quality and reduce financial losses. The scope of research using deep learning and swarm intelligence in the conventional plant disease identification process is explored. Convolutional neural network (CNN) is precise for image classification problems, but its efficiency is hyper parameter selection dependent. Hence, proposed work utilises particle swarm optimisation algorithm in tuning five influential hyper parameters of a CNN architecture that optimises the process of identification and classification of plant diseases. Experimentation is conducted on 10,567 images for 10 classes including healthy and diseased plant leaves of five species from PlantVillage dataset, covering bacterial, fungal and viral diseases. An optimised CNN prototype is attained, providing 98.52% accuracy with fewer parameters and shorter training time compared to pre-trained models.

Keywords: convolutional neural network; CNN; deep learning; DL; hyper parameter optimisation; particle swarm optimisation; PSO; plant leaf disease detection; PLDD; swarm intelligence; SI.

DOI: 10.1504/IJCSE.2024.139710

International Journal of Computational Science and Engineering, 2024 Vol.27 No.4, pp.443 - 457

Received: 24 Nov 2022
Accepted: 07 Aug 2023
Published online: 05 Jul 2024 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article