Voice and Bangla Text-Based Multimodal Disease Detection and Recommendation System for Rice Leaf Disease

Rice is one of the most significant cereal crops in the world, providing an essential food source for billions of people. Rice leaf diseases can seriously hamper rice production which leads to rice yield damage, lower production quality, or even failure entirely. Therefore, it is essential for farmers to quickly and accurately identify rice diseases in order to take the necessary precautions to safeguard their crops and guarantee food security. Machine learning has been successfully utilized for the early detection of plant diseases for some time. This study aims to develop a multimodal disease detection and recommendation system for accurate identification and treatment of illnesses of rice leaves using deep learning-based techniques. The system uses DistilBERT-base for text and voice inputs and CNN models for rice leaf image analysis to identify diseases and recommend remedies. The performance of different convolutional neural network models was compared for the task of rice leaf disease detection. Among the tested models, DenseNet121 was found to provide the best results with 100% accuracy. Our web-based system allows users to submit information using voice recordings in addition to text inputs. For additional processing, the audio input is transformed into text, making it easier for farmers who may not be competent in written language to utilize. Additionally, to further improve user accessibility, our system supports Bangla, a language that is widely spoken in the area, in addition to English text inputs and Bangla transliteration. This technology has significant potential to improve rice crop productivity and ensure a stable supply of this crucial staple meal.