MUMBAI, India, Aug. 31 -- Intellectual Property India has published a patent application (202641101775 A) filed by Madhankumar C; Mrs. D. Kousari Kumari - Vignan'S Institute Of Management And Technology For Women; Mrs. K. Helini - Vignan'S Institute Of Management And; Dr. G. Apparao Naidu - Vignan'S Institute Of Management And Technology For Women; Ms. Ch. Prathyusha - Vignan'S Institute Of Management; and Dr. D. Shanthi - Vignan'S Institute Of Management And on August 23, 2026, for Self-Supervised Soil Health Index From Multi-Spectral Drone Imagery Without Ground-Truth Labels.
Inventors include Mrs. D. Kousari Kumari - Vignan'S Institute Of; Mrs. K. Helini - Vignan'S Institute Of Management And; Dr. G. Apparao Naidu - Vignan'S Institute Of Management; Ms. Ch. Prathyusha - Vignan'S Institute Of Management And Technology For Women; and Dr. D. Shanthi - Vignan'S Institute Of Management And Technology For Women.
The application for the patent was published on August 28, 2026, under issue no. 35/2026.
Abstract: Self-Supervised Soil Health Index from Multi-Spectral Drone Imagery Without Ground-Truth Labels ABSTRACT The present invention relates to a Self-Supervised Soil Health Index Generation System from Multi-Spectral Drone Imagery Without Ground-Truth Labels, designed to estimate and monitor soil health conditions using aerial multi-spectral images without requiring extensive manually labelled soil datasets or conventional ground-truth measurements for model training. Conventional soil health assessment generally requires physical soil sampling, laboratory analysis, field measurements, and manually labelled datasets. These processes can be time consuming, expensive, spatially limited, and difficult to scale across large agricultural areas. Further, soil properties may vary significantly across different geographic locations, crop stages, moisture conditions, and environmental conditions. The present invention addresses these limitations by providing a self-supervised learning framework capable of extracting meaningful soil characteristics directly from multi-spectral drone imagery. The proposed system comprises a drone-based multi-spectral imaging module, image acquisition module, geo-referencing module, radiometric preprocessing module, spectral feature extraction module, self- supervised representation learning module, soil-zone segmentation module, adaptive soil health estimation module, Soil Health Index generation module, temporal monitoring module, and visualization interface. The drone based imaging module captures images using multiple spectral bands including visible and near-infrared bands and optionally red-edge, thermal, or other suitable spectral bands. The radiometric preprocessing module performs image correction, normalization, noise reduction, image alignment, and geometric correction. The spectral feature extraction module derives spectral indices and learned representations associated with vegetation condition, soil moisture, organic matter indicators, surface characteristics, nutrient-related patterns, and other soil-related features. The self-supervised representation learning module generates training signals directly from the acquired imagery rather than depending on manually labelled ground-truth soil health classes. The module may employ techniques including image reconstruction, masked spectral-band prediction, multi-view consistency learning, contrastive learning, temporal consistency learning, spectral transformation prediction, or combinations thereof. The resulting learned representation captures spatial, spectral, and contextual characteristics associated with different soil conditions. The soil-zone segmentation module divides the agricultural field into spatially meaningful zones based on learned spectral and spatial representations. The adaptive soil health estimation module analyzes each zone using multiple learned and derived features and generates component-level indicators corresponding to soil condition. These indicators are combined to generate a normalized Soil Health Index (SHI) representing the relative soil health condition of different field regions. The system further incorporates temporal imagery acquired during different dates or crop cycles to identify changes in soil condition. Temporal consistency enables the system to distinguish persistent soil characteristics from temporary variations caused by illumination, moisture, crop cover, or environmental conditions. The generated Soil Health Index may be displayed as a spatial soil-health map containing different health levels or continuous index values. The system may identify potentially degraded, moderate, and healthy soil zones and provide decision-support information for precision agriculture, irrigation management, fertilizer planning, soil remediation, crop planning, and sustainable farm management. The invention eliminates the requirement for large-scale manually labelled ground-truth datasets during the primary representation-learning stage, thereby enabling scalable soil-health assessment over large agricultural areas. The proposed system provides a non-destructive, spatially distributed, adaptive, and cost-effective approach for monitoring soil health using multi-spectral drone imagery and self-supervised learning. The invention is applicable to precision agriculture, smart farming, soil degradation monitoring, agricultural resource management, crop planning, irrigation optimization, nutrient management, environmental monitoring, and sustainable agricultural practices.
Disclaimer: Curated by HT Syndication.
