Intellectual Property India Publishes Patent Application for 'AI-Driven Multispectral Crop Stress Detection Detection Using Vegetation Indices And Machine Learning' Filed by Mohan Kandasamy; M/S Sri Shanmugha College Of Engineering And Technology; Dr. G. Vijayakumar; Abhinithi M; Jayashree V; Kavibharathi B; Sudha R; Dr. V. Dharmaraj; Mr. M. Sowndhararajan; and Ms. A. Harini

MUMBAI, India, May 29 -- Intellectual Property India has published a patent application (202641062218 A) filed by Mohan Kandasamy; M/S Sri Shanmugha College Of Engineering And Technology; Dr. G. Vijayakumar; Abhinithi M; Jayashree V; Kavibharathi B; Sudha R; Dr. V. Dharmaraj; Mr. M. Sowndhararajan; and Ms. A. Harini, Velur, Tamil Nadu, on May 15, for 'ai-driven multispectral crop stress detection detection using vegetation indices and machine learning.'

Inventor(s) include Mohan Kandasamy; M/S Sri Shanmugha College Engineering And Technology; Dr. G. Vijayakumar; Abhinithi. M; Jayashree V; Kavibharathi B; Sudha R; Dr. V. Dharmaraj; Mr. M. Sowndhararajan; and Ms. A. Harini.

The application for the patent was published on May 29, under issue no. 22/2026.

According to the abstract released by the Intellectual Property India: "Agricultural productivity worldwide is increasingly threatened by a wide range of crop stress conditions, many of which remain invisible to the naked eye until considerable change has already done.This paper proposes an AI-based crop stress detection framework that merges multispectral and hyperspectral remote sensing data with supervised machine learning algorithms to achieve early, automated, Identification of three primary stress categories such as Nutrient deficiency,Water stress and Pest or Disease infection. The system computes 3 Vegetative indices such as NDVI,NDRE,SAVI as Engineered features, and evaluates classifiers a one dimensional Convolutional Neural Network, Random forest, Support vector machine. Experiments were conducted on a curated dataset comprising 10360 labeled samples - 7910 for training and 2450 for testing - organized through a meta data file and balanced carefully.Data augmentation including horizontal flipping,rotation,zoom and brightness variation was applied during training to simulate the real - world variability of field photographs.The accuracy of 93 - 96% with F1 score of 0.93. The performance gap between the pretrained models and the from scratch baseline clearly demonstrates the value of transfer learning in agricultural image classification,where domain specific datasets are rarely enough to train models and deep networks from zero."

Disclaimer: Curated by HT Syndication.