Enhancing grape disease detection: A comparative analysis of hybrid CNN-LSTM and CNN methods

Area of Science:

• Agricultural Science
• Computer Science
• Machine Learning

Background:

• Accurate crop disease detection is vital for agricultural productivity and farmer profitability.
• Existing methods for grape leaf disease identification require improvement in accuracy and efficiency.

Purpose of the Study:

• To compare a hybrid Long Short-Term Memory (LSTM) and Convolutional Neural Network (CNN) model against a standalone CNN for grape crop disease detection.
• To develop and evaluate a deep learning system for automated, end-user deployable leaf disease classification.

Main Methods:

• Training and testing hybrid LSTM-CNN and standalone CNN models on balanced (4000 images) and imbalanced (4062 images) grape leaf datasets.
• Utilizing datasets with healthy, ESCA, leaf blight, and black rot classes for comprehensive evaluation.

Main Results:

• The LSTM-CNN hybrid method achieved 100% accuracy on both balanced and imbalanced datasets.
• The standalone CNN method achieved 99.47% accuracy on the balanced dataset and 97.89% on the imbalanced dataset.
• The LSTM-CNN method demonstrated superior performance across all evaluated metrics.

Conclusions:

• The proposed LSTM-CNN hybrid model offers a highly accurate and efficient solution for automated grape crop disease detection and classification.
• The developed deep learning system is suitable for end-user deployment, enabling immediate disease prediction without retraining.
• This technology can significantly aid farmers in minimizing crop losses and enhancing agricultural productivity.