AI-Powered PlantDoc Bot For Early Crop Disease Detection

Anish Patange , Sanket Lavhate , Soham Yadav , Chetan Aher

PlantDocBot, Plant disease classification, Computer vision in agriculture, BERT, Transfer learning, Smart farming

Every year, plant diseases cause large losses in crop yield, posing a serious threat to global food security. Effective intervention depends on a prompt and precise diagnosis of the disease, but farmers often lack access to specialists in plant pathology, particularly in rural areas. In order to help with plant disease diagnosis, this paper introduces PlantDocBot, a multimodal AI-driven chatbot that can accept both textual symptom descriptions and leaf images. The system combines Natural Language Processing (NLP) for symptom-based diagnosis and a Convolutional Neural Network (CNN) for image-based disease identification in an intuitive chat interface. The use of transfer learning (MobileNet/ResNet-based models) for effective image classification and a fine-tuned transformer (BERT-based) to understand symptoms reported by farmers are two important innovations. Once a disease has been identified, a treatment recommendation module suggests evidence-based treatments based on a carefully curated knowledge base. With a Streamkit-built frontend prototype that allows image upload and real-time communication, we deployed PlantDocBot using HuggingFace Transformers for natural language processing and TensorFlow for CNN model training. A brief user study reveals the chatbot's responses are pertinent and useful, and evaluation on benchmark datasets demonstrates high disease detection accuracy (up to 95% top-1 accuracy). Small-scale farmers can now access real-time, affordable diagnosis thanks to PlantDocBot's scalable approach to democratizing plant disease expertise. To further close the knowledge gap in agriculture, future work will investigate voice interfaces, multilingual support, and increased coverage of disease.

Volume 14 Issue 4, April-2026

Pages : d479-d486