Plant IVRNet- A deep transfer Learning Model with stacked pre-trained models for plant leaf disease detection

DOI: https://doie.org/10.0802/Jbse.2024378504

Suma S H , K. Prabhushetty

Keywords:

Plant disease, CNN, VGG-16, IVR, Machine Learning, Deep Learning, ResNet-50, VGGNet.

Abstract:

Detection of plant leaf disease detection always remains a critical area of research in agricultural science, as prompt and precise identification of diseases can significantly enhance the crop management and yield. Current advancements in deep learning (DL) demonstrate that deep transfer learning models can enhance classification performance across various tasks. In this work, we introduce a deep transfer learning-based model for plant leaf disease classification. The proposed Plant IVR Net architecture integrates convolutional neural networks (CNNs) with VGG-16 and ResNet-50 models to leverage the strengths of these pre-trained networks. The model applies a loss function to minimize overall system loss and improve classification accuracy. The key components of the proposed Plant IVR Net include combination of convolution layer for feature extraction, pooling layers to address the spatial size issues and fully connected layer to obtain the final classification. To address class imbalance, data augmentation procedures such as rotation and noise addition, flipping and cropping are employed. Fine-tuning the pre-trained models further enhances performance. This approach effectively classifies plant leaf diseases, providing a robust tool for precision agriculture and plant health monitoring. The performance of Plant IVR Net approach is measured on publically available Plant village dataset where the Plant IVR Net approach has obtained the overall accuracy as 99.98% which shows significant improvement when compared with the existing transfer learning and deep learning models.

PDF