Novel deep learning assisted plant leaf classification system using optimized threshold-based CNN

Abstract

In general, plant classification systems can be a beneficial tool in agriculture, especially when identifying plant types in a systematic what's more, sensible way. Already, plant breeders relied on observation and experienced personnel to distinguish plant varieties. However, some plants, such as leaves and branches, have nearly identical characteristics, making identification difficult. Therefore, there is a need for a system that can solve this problem. Therefore, this study focuses on the characterization of plant leaves using convolution neural network (CNN) techniques. The main idea of this paper is to propose a new deep learning-based model for plant leaf classification. Initially, preprocessing is done using RGB-to-grayscale conversion, histogram equalization, and median filtering to improve the image quality required for further processing. The results show that with the activation layer of the algorithm, 15-layer network design and a trial-training ratio of 70-30, the plant leaf classification system can achieve 90% classification accuracy of coriander and parsley with an error rate of 0.1. Furthermore, due to its high accuracy, the system can be extended to other uses such as identifying plant diseases and species.