Intellectual Property India Publishes Patent Application for Vision Transformer Based Multi-Task Framework With Kolmogorov-Arnold Networks For Plant Disease Classification And Severity Estimation

MUMBAI, India, Aug. 24 -- Intellectual Property India has published a patent application (202621081506 A) filed by Symbiosis International Deemed University on July 01, 2026, for Vision Transformer Based Multi-Task Framework With Kolmogorov-Arnold Networks For Plant Disease Classification And Severity Estimation.

Inventors include Nishit Bohra M; Avantika R. Patil; Omkar H. Thorve; Dr. Aniket K. Shahade; Dr. Priyanka V. Deshmukh; Dr. Shruti Patil; and Dr. Ketan Kotecha.

The application for the patent was published on August 21, 2026, under issue no. 34/2026.

Abstract: ABSTRACT VISION TRANSFORMER BASED MULTI-TASK FRAMEWORK WITH KOLMOGOROV-ARNOLD NETWORKS FOR PLANT DISEASE CLASSIFICATION AND SEVERITY ESTIMATION The present invention discloses a multi-task framework (100), RoViT-KAN, employing a DeiT-Tiny Vision Transformer backbone (122) with a disease classification head (130), a severity estimation head (140) using a Kolmogorov-Arnold Network module (142) with learnable B-spline transformations for interpretable continuous severity scores, a heteroscedastic uncertainty prediction head (150), a four-stage curriculum learning module (160), and a multi-task loss module (170). The KAN module provides interpretable severity mappings while the curriculum learning ensures stable multi-task optimization. The framework achieves 99.70 percent classification accuracy with calibrated uncertainty estimates for trustworthy plant disease monitoring. [

Disclaimer: Curated by HT Syndication.