ENHANCING CROP RESILIENCE: A HYBRID MODEL OF VGG16, INCEPTION AND XCEPTION FOR PLANT DISEASE PREDICTION

Abstract

Timely identification of plant diseases poses a significant challenge in the agricultural sector. Various image processing techniques are currently being utilized to distinguish the type of plant disease from a single image. Detecting illnesses promptly can help minimize the impact of crop diseases on food production. The objective of this research is to develop a CNN-based model that takes an image of a leaf as input, enabling the trained system to recognize the disease by analyzing distinct patterns on the leaf. Once the disease is identified, the system itself can suggest remedies to mitigate the disease. Hybrid Convolutional Neural Networks outperform other CNN architectures. By leveraging deep learning to analyze image features, a hybrid CNN model (HCM) can effectively detect plant diseases. Initially, the CNN module simultaneously extracts various features using the pre-trained models VGG16, Inception V3, and Xception, which are fine-tuned through transfer learning after training on the ImageNet dataset. Subsequently, the extracted features are concatenated to yield comprehensive results, as pre-trained models extract unique features. Finally, the fully connected layers of the HCM are trained to differentiate between different types of plant diseases. The proposed HCM demonstrates an 85% accuracy rate in detecting pathogens in leaf images, suggesting that farmers could benefit from maintaining secure and disease-free agricultural systems.