Research Projects

Plant stress causes a poor crop production ensuing economic impediment to the farmers and leading to a global challenge for food supply. Plant stress is caused by several factors e.g., nutrition deficiency, climate change, and diseases. With the advent of machine learning (ML) and deep learning (DL), several intelligent techniques are devised for crop stress detection. Most of the methods are not generalized for multiple crops cultivated in real-fields influenced by weather and soil conditions. Existing plant disease and nutrition deficiency recognition methods tested on small-scale datasets, which are relevant for a single crop only e.g., rice, corn, etc. It is a limitation of prior works, as those are developed for a specific crop, which fail to achieve good diagnostic results for other crops. It causes a financial concern of rural farmers to use such crop-specific technology for various seasonal crops. So, a common solution is vital for multiple crops to optimize cost. Generally, stressed crops reflect structural visual variations in leaf color, texture, or inconsistent lesion compared to healthy plants. The contrastive appearance and visual symptoms provide valuable insights to identify major factors causing crop stress i.e., disease and nutrition deficiency. To solve this, though, vision-based ML/DL methods are effective in extracting complex patterns and semantic features from leaf images, yet, they lack proper interpretability to understand decision-making process, which is crucial for farmers. So, integration of ML/DL model with a large language model (LLM), called vision-language model (VLM) that boosts recognition performance endowed with proper details in natural languages is essential. In this project, a new VLM will be developed and deployed in a smartphone for assisting remote Indian farmers to overcome their challenges in situ crop conditions. As mustard and rapeseed are two most significant seed-oils in Rajasthan, both will be explored with other plants (rice, tomato, etc.) for assessing nutrition (NPK) status and diseases. Three key aspects causing agricultural stress will be studied by developing a novel VLM. A multimodal database will be developed including weather data (temperature, humidity, etc.), soil data (moisture, nutrients, etc.), and crop images at different times of day for analyzing plant growth at multiple stages. The current proposal, for the first time, combines three multimodal datasets (soil, and images) for crop health monitoring in Indian contexts. The new dataset will be used for training and validating the proposed model to predict plant stress using real-time test data, and take necessary precautions harnessing explanation provided by VLM. An IoT system will be devised using raspberry pi, data handling via cloud server and a computer. A low-cost smartphone application that will assist rural farmers to diagnose plant stress with proper explanation will be devised for improving Indian agriculture.

Agriculture, Plant disease classification, Soil-quality estimation, Convolutional neural networks, Vision language model, Thermal imaging.