Double-normal-form combined remote sensing image end-to-end fine-grained target detection method

By employing a dual-paradigm joint end-to-end remote sensing image detection method, which combines a sparse directional proposal network and a query-aware refinement head with dual auxiliary supervision heads, the problem of high computational overhead and insufficient sparse supervision in fine-grained target detection of remote sensing images is solved, and high-precision fine-grained target detection is achieved.

CN122157019APending Publication Date: 2026-06-05BEIHANG UNIV
0 Cites 0 Cited by

Patent Information

Application Number
CN202610296920.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-03-12
Publication Date
2026-06-05

AI Technical Summary

Technical Problem

In existing methods for fine-grained target detection in remote sensing images, two-stage detectors rely on non-maximum suppression, resulting in high computational overhead. Query-based detectors suffer from insufficient sparse supervision, and the contradiction between global deduplication and fine-grained recognition remains unresolved, affecting detection accuracy.

Method used

A dual-paradigm joint approach is adopted, which generates non-repeating region proposals through a sparse directional proposal network. Combined with a query-aware refinement head and dual auxiliary supervision heads, it realizes one-to-one and one-to-many matching strategies. A two-stage consensus scoring mechanism is introduced to provide dense supervision signals and reduce interference from multiple tasks.

Benefits of technology

It eliminates the need for nonmaximum suppression post-processing, improving the stability and inference efficiency of detection results, enhancing the accuracy of fine-grained classification, and improving positioning accuracy and fine-grained recognition capabilities.

✦ Generated by Eureka AI based on patent content.
Patent Text Reader

Abstract

The invention discloses a double-normal-form combined remote sensing image end-to-end fine-grained target detection method, and relates to the technical field of computer vision and pattern recognition, and the method comprises the steps: extracting a multi-scale feature map through a feature extraction network; based on the sparse orientation proposal network, a one-to-one sparse matching strategy is adopted to generate non-repeated sparse orientation region proposals and proposal features and coarse-grained target confidence corresponding to the non-repeated sparse orientation region proposals; based on the query perception refining head, adopting a one-to-many dense matching strategy to obtain a fine-grained classification probability and a refined bounding box; in the training stage, double auxiliary supervision heads are introduced to provide additional dense supervision signals; in the reasoning stage, a two-stage consensus scoring mechanism is adopted, and finally a fine-grained target detection result is obtained. According to the method, NMS-dependence-free end-to-end remote sensing image fine-grained target detection is realized, and the positioning precision and the fine-grained classification performance are effectively improved through sparse and dense double-normal-form cooperation.
Need to check novelty before this filing date? Find Prior Art