Plant Leaf Disease Detection and Classification Using Deep Learning: A Review and A Proposed System on Bangladesh’s Perspective

Md. Jalal Uddin Chowdhury, Zumana Islam Mou, Rezwana Afrin, Shafkat KibriaInternational Journal of Science and Business 2023 · DOI

TL;DR

This paper addresses the critical challenge of plant disease detection in Bangladesh's agriculture, which significantly impacts food security and livelihoods. The authors propose a Convolutional Neural Network (CNN) model for automated leaf disease detection, achieving an accuracy of 85.31% on a dataset of bell pepper, tomato, and potato crops. This work aims to provide a more precise and timely solution than manual inspection, potentially boosting crop yields and farmer income.

What Problem Does This Paper Address?

Agriculture is the backbone of Bangladesh's economy, employing a large portion of the population and contributing significantly to its GDP. However, plant diseases pose a major threat to agricultural production, leading to reduced crop quality, increased production costs due to pesticide use, and diminished farmer income. Traditional methods of disease detection rely on manual inspection by farmers, which is often inaccurate and time-consuming, especially for early-stage detection. This delay in identification can lead to the ineffective use of resources and significant crop loss. The paper aims to bridge this gap by leveraging deep learning for automated, accurate, and early detection of plant diseases, specifically tailored to crops important in Bangladesh.

Methods

The core of this research is the development and evaluation of a Convolutional Neural Network (CNN) model for plant leaf disease detection. The methodology follows a standard machine learning pipeline:

  • Dataset Collection: The authors utilized the publicly available "Plant Village Dataset" from Kaggle, which contains images of plant leaves.
  • Data Preprocessing: Images were processed, cleaned, and labeled. This involved normalizing images and applying data augmentation techniques to enhance robustness and improve feature learning.
  • Feature Extraction: Unlike traditional machine learning, the CNN model automatically learns discriminative features directly from the image data during the training phase.
  • Model Training: A CNN architecture was employed for training. This architecture consists of an input layer, multiple hidden layers (including Convolutional, Rectified Linear Unit (ReLU), and Pooling layers), and an output layer. Specifically, they used a 2x2 max pooling layer for better precision.
  • Classification and Recognition: The trained CNN model analyzes input leaf images to classify them as healthy or unhealthy and identify specific diseases.

The proposed CNN model was compared against other machine learning algorithms, including Support Vector Machine (SVM), Logistic Regression, Decision Tree, and K-Nearest Neighbors (KNN).

Datasets

The primary dataset used in this study is the Plant Village Dataset (available on Kaggle). This dataset comprises 17,430 images of plant leaves, specifically focusing on three crops: bell pepper, tomato, and potato. The images are labeled into 14 distinct classes, representing various types of damage and healthy leaves. The dataset was divided into training and testing sets.

Key Insights and Ideas

  • Deep Learning's Advantage in Feature Learning: The paper highlights how deep learning models like CNNs can automatically learn complex features from raw image data, eliminating the need for manual feature engineering, which is a significant advantage over traditional machine learning approaches.
  • Importance of Data Augmentation: Preprocessing steps like normalization and augmentation are crucial for improving the robustness and accuracy of the CNN model, especially when dealing with variations in image quality and lighting.
  • CNNs Outperform Traditional ML for this Task: The study reinforces the general finding that CNNs are highly effective for image-based classification tasks, significantly outperforming traditional algorithms like SVM and KNN in this specific context of plant disease detection.
  • Tailoring Solutions for Local Needs: While many existing studies use datasets from various countries, this research emphasizes the importance of focusing on crops prevalent in Bangladesh to provide a more practical and beneficial solution for local farmers.

Key Results

The proposed CNN model achieved a testing accuracy of 85.31%. This is a significant improvement compared to the alternative models evaluated:

  • SVM: 66.15%
  • Logistic Regression: 42.43%
  • KNN: 41.25%

The paper also presents figures illustrating the testing accuracy across different models, epochs, and batch sizes, further supporting the superior performance of the CNN. The model was trained with 150 epochs and utilized enhanced datasets.

Limitations and Future Work

The authors acknowledge several limitations and propose future research directions:

  • Reliance on Online Datasets: The study was limited by the lack of readily available Bangladeshi-specific datasets. The authors plan to collect their own data from Bangladeshi fields in the future to train models from scratch.
  • Computational Resources: The current work was performed using a regular CPU. The authors intend to utilize highly configured GPUs in the future to further improve overall accuracy and processing speed.
  • Limited Crop Scope: The current model focuses on bell pepper, tomato, and potato. Future work will include data for rice, a staple food in Bangladesh, to expand the model's applicability.
  • Offline Mobile Application: To address the limited internet access in rural Bangladesh, the authors plan to develop a mobile application based on their model that can function offline, making it more accessible to farmers.
  • Transfer Learning: The authors express interest in using pre-trained transfer learning models for potentially more accurate performance compared to training from scratch.

Summary updated

Abstract

A very crucial part of Bangladeshi people’s employment, GDP contribution, and mainly livelihood is agriculture. It plays a vital role in decreasing poverty and ensuring food security. Plant diseases are a serious stumbling block in agricultural production in Bangladesh. At times, humans can’t detect the disease from an infected leaf with the naked eye. Using inorganic chemicals or pesticides in plants when it’s too late leads in vain most of the time, deposing all the previous labor. The deep-learning technique of leaf-based image classification, which has shown impressive results, can make the work of recognizing and classifying all diseases trouble-less and more precise. In this paper, we’ve mainly proposed a better model for the detection of leaf diseases. Our proposed paper includes the collection of data on three different kinds of crops: bell peppers, tomatoes, and potatoes. For training and testing the proposed CNN model, the plant leaf disease dataset collected from Kaggle, is used which has 17430 images. The images are labeled with 14 separate classes of damage. The developed CNN model performs efficiently and could successfully detect and classify the tested diseases. The proposed CNN model may have great potency in crop disease management.
BibTeX
@article{chowdhury2023plant,
  author = {Md. Jalal Uddin Chowdhury and Zumana Islam Mou and Rezwana Afrin and Shafkat Kibria},
  title = {{Plant Leaf Disease Detection and Classification Using Deep Learning: A Review and A Proposed System on Bangladesh’s Perspective}},
  journal = {International Journal of Science and Business},
  year = {2023},
  doi = {10.58970/ijsb.2214},
}
Figure