Neural Network Architectures for Multimodal Applications: Convolutional Networks for Plant Disease Classification and Transformer Models for Conversational Interaction

This research presents an artificial intelligence system designed to support precision agriculture by addressing two critical challenges: automated plant disease diagnosis and accessible agricultural information retrieval.The system comprises two synergistic deep learning models.First, a Convolutional Neural Network (CNN) is developed for the image-based classification of plant diseases, trained on a curated dataset containing leaf images of various crops (e.g., tomato, potato, corn) affected by common blights, molds, and deficiencies.The CNN model, leveraging transfer learning with architectures like EfficientNet, achieves high accuracy in localizing and classifying disease symptoms, providing a rapid diagnostic tool for farmers.Second, a Natural Language Processing (NLP) chatbot, fine-tuned on a specialized crop-based Question-Answering (QnA) dataset, serves as a conversational agricultural assistant.This dataset, compiled from agricultural extension resources and expert knowledge, covers topics such as pesticide recommendations, treatment protocols, cultivation best practices, and symptomology.Experimental results demonstrate the effectiveness of the CNN model, which achieves a test accuracy of over 94% on held-out data.This end-to-end system demonstrates the practical potential of multimodal AI in democratizing expert-level agricultural knowledge, promoting early disease intervention, and contributing to sustainable farming practices.