Papaya Fruit And Leaf Ringspot Virus Detection Using Convolutional Neural Network (Cnn) And Transfer Learning

Abstract

This article uses the application of Convolutional Neural Networks (CNN) and transfer learning models to discover Papaya Ringspot Virus (PRSV) in papaya plants, a critical concern for papaya cultivation in Malaysia. The study compares the performance of CNN with three pre-trained models: VGG16, ResNet50, and InceptionV3. Each model employs the Adam optimizer with a learning rate of 0.001. Models were trained and evaluated utilizing a dataset comprising 1,524 images of papaya fruits and leaves affected by PRSV. CNN, trained for 10 epochs without fine-tuning, achieved a final accuracy of 99.22%, starting from an initial accuracy of 81.91%. In contrast, the pre-trained models demonstrated superior performance. InceptionV3 consistently achieved a perfect final accuracy of 100% across various configurations, with initial accuracies ranging from 92.52% to 96.02%. VGG16 also attained 100% accuracy after fine-tuning. ResNet50 showed robust performance, achieving 100% accuracy in all configurations, with initial accuracies improving with extended training periods. The experimental results indicate that pre-trained models significantly outperform the custom CNN in PRSV detection, highlighting the benefits of transfer learning. InceptionV3, with its efficient architecture and fewer parameters than VGG16, demonstrated the highest performance, making it the most suitable model for this task. The results highlight to ability of deep learning models to enhance and discover crop disease and support sustainable agriculture in Malaysia.