MUMBAI, India, Nov. 2 -- Intellectual Property India has published a patent application (202641125861 A) filed by Srm Trp Engineering College on October 22, 2026, for Aquavision: An Adaptive, Detection-Guided Lightweight System For Joint Underwater Image Enhancement And Object Detection.
Inventors include Rooban Raj S; Pragadhish P; Vijayarajan S; and Mirdula S.
The application for the patent was published on October 30, 2026, under issue no. 44/2026.
Abstract: This invention describes AquaVision, a lightweight neural network system for jointly enhancing underwater images and detecting objects within them. A severity- estimation module first analyzes an input underwater image and produces a score indicating the degree of degradation caused by underwater light scattering and absorption. This score drives an adaptive gating mechanism that controls the strength of a physics-based enhancement process, which decomposes the image into clean scene, background light, and transmission components to produce a corrected output. The enhanced image is then passed to a lightweight object detection module to identify and localize objects. During training, a detection-guided feedback loss, derived from the object detector's accuracy, is used alongside a self-supervised reconstruction loss to tune the enhancement module so that its output directly benefits detection performance rather than only improving visual appearance. At inference, the system performs a single efficient forward pass without the feedback computation, preserving real-time performance. The invention combines condition- adaptive processing and detection-guided training within one lightweight, unified architecture, addressing a gap not covered by existing underwater enhancement and detection systems, each of which typically achieves only one of these properties individually.
Disclaimer: Curated by HT Syndication.
