Research Projects

This project is about helping India’s medicinal plant farmers, who face big problems with their produce. Many batches get rejected because active compounds like rosmarinic acid, withanolides, and berberine don’t meet quality standards or because diseases show up too late. Current testing methods, like HPLC and GC-MS, destroy the plants and only give results after harvest, causing huge income losses for small farmers. Hyperspectral imaging (HSI) can check these compounds without harming the plants, but the equipment is expensive and needs heavy computing power, so it’s out of reach for most farmers. Neural Spectral Reconstruction (NSR) technology can estimate full spectra from a single RGB image, but existing models trained on urban scenes or vegetables lack accuracy for medicinal plants’ key absorption bands. The aim is to create a robust RGB–HSI dataset for Ocimum sanctum, Withania somnifera, and Tinospora cordifolia, cultivated under controlled stresses like powdery mildew, nitrogen deficiency, and water stress. The objective is to develop MedSpecNet, an NSR model designed to prioritize accuracy in critical absorption bands for these plants’ compounds. The proposed model will be compressed into a lightweight version for fast operation on standard Android phones, enabling farmers to use it easily. A mobile app will be developed to allow farmers to monitor plant health and compound levels in real time, supported by a cloud system that continuously improves the model without requiring new hardware. The hypothesis is that a lightweight model emphasizing key wavelengths can reconstruct hyperspectral data from RGB images with high accuracy and predict compound levels comparable to lab results, regardless of plant type or stress condition. The investigation will involve growing seedlings of these three species in a hydroponic setup with varied stress combinations, monitoring conditions like pH, temperature, and light using IoT sensors. A hyperspectral imaging camera and an RGB camera will capture images to build a dataset with lab-verified compound labels weekly. MedSpecNet will be trained using advanced deep learning techniques, starting with a large model and compressing it for mobile use. The app will be tested on real farms, comparing predictions with lab tests to assess improvements in rejection rates and cost savings. This project will transform an expensive HSI system into an affordable AI-driven tool, enabling non-destructive plant testing for farmers. The open dataset and code will support global researchers, and the approach can extend to spices, tea, and vegetables, strengthening the herbal industry and advancing precision agriculture worldwide. The focus is on empowering farmers and promoting smarter, sustainable farming practices.