MUMBAI, India, Nov. 2 -- Intellectual Property India has published a patent application (202621112489 A) filed by Sneha Ramchnadra Burnase; Poonam Niraj Sonar; and Dr. Udhav Bhosle on September 19, 2026, for A Method For Hyperspectral Image Classification Using Contrastive Multi-Head Spectral-Spatial Feature Learning.
Inventors include Sneha Ramchnadra Burnase; Poonam Niraj Sonar; and Dr. Udhav Bhosle.
The application for the patent was published on October 30, 2026, under issue no. 44/2026.
Abstract: A METHOD FOR HYPERSPECTRAL IMAGE CLASSIFICATION USING CONTRASTIVE MULTI-HEAD SPECTRAL-SPATIAL FEATURE LEARNING The present invention relates to a computer-implemented method for hyperspectral image classification using contrastive multi-head spectral-spatial feature learning. The method comprises receiving hyperspectral image data comprising a plurality of spectral bands and spatial information, preprocessing the data to reduce noise and redundant spectral information, and generating spectral-spatial patches. Augmented views of the patches are generated and processed using a contrastive learning module. Spectral-spatial feature representations are extracted using a plurality of learning heads configured to learn complementary spectral and spatial characteristics. The feature representations generated by the learning heads are fused to obtain a fused feature representation. A classification module utilizes the fused feature representation to classify at least one pixel or region of the hyperspectral image. The method utilizes labelled and unlabelled hyperspectral image data and is applicable to precision agriculture, land-cover mapping, environmental monitoring, forest and vegetation analysis, and water-resource management.
Disclaimer: Curated by HT Syndication.
