[Paper Review] Plant Disease Detection Using Image Processing and Machine Learning

Pranesh Kulkarni, Atharva Karwande+4 more|arXiv (Cornell University)|Jun 20, 2021

Smart Agriculture and AI21 citations

TL;DR

This paper presents a computer vision and machine learning-based system for automated plant disease detection, achieving 93% accuracy in identifying 20 different diseases across five common crop species. The method combines image preprocessing, feature extraction using transfer learning with pre-trained CNNs, and classification via fully connected layers, offering a scalable solution for early disease detection in agriculture.

ABSTRACT

One of the important and tedious task in agricultural practices is the detection of the disease on crops. It requires huge time as well as skilled labor. This paper proposes a smart and efficient technique for detection of crop disease which uses computer vision and machine learning techniques. The proposed system is able to detect 20 different diseases of 5 common plants with 93% accuracy.

Motivation & Objective

• To address the time-consuming and labor-intensive nature of manual plant disease detection in agriculture.
• To develop an automated, scalable system for early detection of crop diseases using digital images.
• To achieve high-accuracy classification of multiple plant diseases using machine learning techniques.
• To evaluate the performance of a hybrid image processing and deep learning pipeline on a diverse dataset of plant disease images.

Proposed method

• The system uses image preprocessing techniques such as resizing, normalization, and noise reduction to prepare input images.
• Convolutional Neural Networks (CNNs) are employed for feature extraction, leveraging pre-trained models like VGG16 or ResNet for transfer learning.
• Transfer learning is applied to fine-tune the pre-trained CNNs on a custom dataset of plant disease images.
• A fully connected classifier head is added on top of the feature extractor to perform multi-class classification.
• Data augmentation techniques such as rotation, flipping, and cropping are used to improve model generalization and prevent overfitting.
• The model is trained and validated using standard deep learning frameworks with cross-entropy loss and Adam optimization.

Experimental results

Research questions

• RQ1 Can a deep learning-based system achieve high accuracy in detecting multiple plant diseases from leaf images?
• RQ2 How effective is transfer learning with pre-trained CNNs for plant disease classification in a limited-data setting?
• RQ3 To what extent does image preprocessing enhance the performance of the disease detection pipeline?
• RQ4 How does the system perform across diverse plant species and disease types?

Key findings

• The proposed system achieved a classification accuracy of 93% on a dataset containing 20 different plant diseases across five common crop species.
• Transfer learning with pre-trained CNNs significantly improved model convergence and performance compared to training from scratch.
• Image preprocessing steps such as resizing and normalization contributed to consistent input representation and improved model stability.
• The use of data augmentation techniques enhanced model robustness and reduced overfitting on the training set.