MUMBAI, India, July 7 -- Intellectual Property India has published a patent application (202641079055 A) filed by Sr University on June 25, 2026, for A Multi-Scale Feature Fusion Network With Recursive Attention For Robust Small Object Detection.
Inventors include Suneetha Mudumba; and Dr. Pramod Kumar Poladi.
The application for the patent was published on July 03, 2026, under issue no. 27/2026.
Abstract: A MULTI-SCALE FEATURE FUSION NETWORK WITH RECURSIVE ATTENTION FOR ROBUST SMALL OBJECT DETECTION A Multi-Scale Feature Fusion Network with Recursive Attention (RA-MFFN) for robust small object detection is disclosed. The network comprises a feature extraction backbone, a multi-scale feature fusion architecture, a Recursive Attention Module (RAM), and one or more detection heads. The RAM iteratively refines multi-scale feature representations through recursive processing steps. At each iteration, spatial attention maps and channel attention maps are generated to identify salient object regions and important feature channels. The attention maps are utilized to re-weight feature representations, enhancing weak small-object features while suppressing background noise and irrelevant contextual information. Refined feature maps are recursively fed back into the RAM for progressive enhancement before being provided to detection heads for object classification and localization. The invention improves detection accuracy, robustness, and feature representation of small objects in applications including autonomous driving, aerial surveillance, medical imaging, and industrial inspection while maintaining computational efficiency through recursive processing of previously extracted feature maps.
Disclaimer: Curated by HT Syndication.
