MUMBAI, India, Aug. 17 -- Intellectual Property India has published a patent application (202641096347 A) filed by Narsimha Reddy Engineering College Nrcm on August 10, 2026, for Track Defect Detection System Using An Improved Yolov5s Network With Attention-Enhanced Small-Defect Detection And Geolocation.
Inventors include Darla Varshitha; Dara Indhu; Takkellapati Sujana; Aechan Laxmi Vaishnavi; Gariganti Shiva Kumar; and Malla Sowmya.
The application for the patent was published on August 14, 2026, under issue no. 33/2026.
Abstract: A track defect detection system based on an improved YOLOv5s network is disclosed. An image acquisition module (110) captures images of a rail track surface, which a rail region-of-interest extraction module (112) processes to isolate the rail head region. An improved YOLOv5s backbone network (116), incorporating a lightweight coordinate attention module (118), extracts multi-resolution feature maps that a multi-scale feature fusion neck (120), including an additional shallow feature map fusion path, fuses for input to a small-object detection head (122) operating alongside pre-existing detection heads across at least four spatial resolutions. A bounding box regression module (124) employs a combined focal loss and efficient intersection-over-union loss function, while a defect classification module (125) assigns each detected defect to a predefined category. A defect geolocation tagging module (128) correlates each defect's pixel position with positioning data to generate a track-chainage location, enabling improved detection of small, early-stage, and irregularly shaped rail defects with automated, precise physical localization.
Disclaimer: Curated by HT Syndication.
