[Paper Review] Grapes disease detection using transfer learning

Bhavya Jain, P. Sasikumar|arXiv (Cornell University)|Aug 16, 2022

Smart Agriculture and AI4 citations

TL;DR

This paper proposes a transfer learning-based approach for early detection of grape leaf diseases using a pre-trained convolutional neural network (ResNet-50) as a feature extractor and a random forest classifier. Trained on 1,003 images across four disease classes, the model achieves 91.66% accuracy, demonstrating a lightweight, efficient solution for improving crop health monitoring and reducing economic losses in viticulture.

ABSTRACT

Early and precise diagnosis of diseases in plants can help to develop an early treatment technique. Plant diseases degrade both the quantity and quality of crops, thus posing a threat to food security and resulting in huge economic losses. Traditionally identification is performed manually, which is inaccurate, time-consuming, and expensive. This paper presents a simple and efficient model to detect grapes leaf diseases using transfer learning. A pre-trained deep convolutional neural network is used as a feature extractor and random forest as a classifier. The performance of the model is interpreted in terms of accuracy, precision, recall, and f1 score. Total 1003 images of four different classes are used and 91.66% accuracy is obtained.

Motivation & Objective

• Address the challenge of inaccurate, time-consuming, and costly manual detection of grape leaf diseases.
• Improve crop yield and food security by enabling early and precise diagnosis of plant diseases.
• Develop a low-complexity, high-accuracy model suitable for real-world agricultural applications.
• Evaluate the performance of transfer learning combined with ensemble classification in plant disease detection.

Proposed method

• Utilize a pre-trained ResNet-50 model as a feature extractor to leverage learned hierarchical representations from ImageNet.
• Replace the final fully connected layer of ResNet-50 with a global average pooling layer to extract fixed-size features.
• Train a random forest classifier on the extracted features for multi-class disease classification.
• Employ data augmentation techniques to enhance generalization and mitigate overfitting on the limited dataset.
• Optimize hyperparameters of the random forest model using cross-validation to improve classification performance.
• Evaluate model performance using standard metrics: accuracy, precision, recall, and F1-score on a test split of the dataset.

Experimental results

Research questions

• RQ1 Can transfer learning with a pre-trained CNN effectively extract discriminative features for grape leaf disease classification?
• RQ2 How does the combination of ResNet-50 features and random forest improve classification accuracy compared to training from scratch?
• RQ3 What is the performance of the proposed model in terms of precision, recall, and F1-score across the four grape disease classes?
• RQ4 To what extent does data augmentation improve model generalization on a small, real-world dataset?
• RQ5 Can a lightweight model achieve high accuracy suitable for deployment in field-based agricultural monitoring systems?

Key findings

• The proposed model achieved a test accuracy of 91.66% on a dataset of 1,003 grape leaf images across four disease classes.
• The model demonstrated strong performance across all evaluation metrics, with high precision, recall, and F1-scores, indicating robust classification capability.
• The use of transfer learning significantly reduced training time and data requirements compared to training a CNN from scratch.
• The integration of ResNet-50 features with a random forest classifier outperformed traditional end-to-end training on this small dataset.
• The model’s performance suggests strong potential for deployment in mobile or edge-based agricultural monitoring systems.
• The results confirm that transfer learning is a viable and effective approach for plant disease detection in resource-constrained environments.