An Empirical Performance Analysis of Modified Convolutional Neural Networks Model for Edible Plant Leaf Disease Detection

Abstract

Plant diseases cause low agricultural productivity and are difficult to control and identify due to limited knowledge and lack of resources available with the farmers. The late identification of tomato leaf disease may cause future losses and crop damage. This work presents a deep learning approach for detecting tomato leaf diseases using upgraded Convolutional Neural Networks (CNNs) that have been developed with the layered structure change of convolutions. In this work, the dataset is taken from Kaggle, and in total 11,000 RGB images of tomato leaf diseases were used and the training was carried out for 20 epochs, 40 epochs and 60 epochs with or without adding one more convolution layer to the traditional CNN structure. The accuracies that are obtained without structural change of CNN have been found as 77%, 86% and 88%, whereas, the modified CNN resulted in accuracies of 77%, 93% and 96% with epochs 20, 40, and 60, respectively. These outcomes prove that the modified CNN model improves the prediction accuracy of tomato leaf disease detection over 40 and 60 epochs. This model could further be useful in the early diagnosis of the diseases affecting the leaves and consequent improvement in yield and quality of crops.

Plant phenotyping relevance

トマト葉の病徴を画像から分類する改良CNNを開発・比較し、疾患検出精度を評価しているため、植物病害状態の画像ベース表現型解析手法が中心である。

Code and data availability

公開本文の所在を確認できませんでした。非公開または購読が必要な可能性があります。

No evidence-backed public reproduction asset is currently recorded.