Back to Search Start Over

A convolutional neural network-driven computer vision system toward identification of species and maturity stage of medicinal leaves: case studies with Neem, Tulsi and Kalmegh leaves.

Authors :
Mukherjee, Gunjan
Tudu, Bipan
Chatterjee, Arpitam
Source :
Soft Computing - A Fusion of Foundations, Methodologies & Applications; Nov2021, Vol. 25 Issue 22, p14119-14138, 20p
Publication Year :
2021

Abstract

Medicinal plants are used to cure different common and chronic diseases in different Asian countries including India. The easy availability and planting possibility make them popular resources for alternative medicinal practices like Ayurveda. These medicinal plants possess proven healing potential without causing any side effects. The biochemical constituents of the medicinal leaves are the fundamental reasons of their healing power which considerably vary with maturity. The existing practices of maturity detection are largely based on chemical analysis of leaves through different instruments which are expensive, invasive in nature and time consuming. This paper reports a computer vision-based system to classify the medicinal leaves along with the corresponding maturity level. The process reported is advantageous in terms of comparatively easier, faster, less expensive and non-invasive operations. The presented system captures the leaf images in controlled illumination and processed leaf images are fed to the convolutional neural network (CNN) architecture-based system for classification of both the type of leaves and maturity stage. The paper also presents application of binary particle swarm optimization ((BPSO) for arriving the values of the CNN hyperparameters. The potential of the system has been assessed with three popular species of medicinal plants, namely Neem, Tulsi and Kalmegh. The classification results were verified with repeated test runs and different standard metrics including tenfold cross-validation method. The paper shows that the presented CNN-driven computer vision framework can provide about 99% classification accuracy for simultaneous prediction of leaf specie and maturity stage. Such significant performance of the presented method can be considered as a potential addition to the existing chemical and instrumental methods. [ABSTRACT FROM AUTHOR]

Details

Language :
English
ISSN :
14327643
Volume :
25
Issue :
22
Database :
Complementary Index
Journal :
Soft Computing - A Fusion of Foundations, Methodologies & Applications
Publication Type :
Academic Journal
Accession number :
153186308
Full Text :
https://doi.org/10.1007/s00500-021-06139-9