An unmanned aerial vehicle vision target detection method and system based on a density-guided feature pyramid

By adaptively adjusting the multi-scale feature fusion using a density-guided feature pyramid structure, the problems of weakened small target features and background interference in UAV detection are solved, achieving high-precision and robust UAV target detection.

CN122157052APending Publication Date: 2026-06-05QINGYANXIN (NINGBO) COMMUNICATION TECHNOLOGY CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
QINGYANXIN (NINGBO) COMMUNICATION TECHNOLOGY CO LTD
Filing Date
2026-03-04
Publication Date
2026-06-05

AI Technical Summary

Technical Problem

Existing deep learning detection algorithms suffer from problems such as weakened features of small targets, fixed fusion weights, and interference from complex backgrounds in drone detection, which cannot effectively improve robustness and accuracy.

Method used

A UAV visual target detection method based on density-guided feature pyramid is adopted. The method adaptively adjusts multi-scale feature fusion through a density-aware mechanism, including an image input module, a backbone feature extraction module, a density estimation module, and a density-guided fusion module, to optimize the feature extraction and fusion process.

Benefits of technology

It significantly improves the performance of small target detection, enhances the robustness and accuracy of UAV detection, improves the detail fidelity of small target areas, suppresses background noise, and improves the overall detection accuracy.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122157052A_ABST
    Figure CN122157052A_ABST
Patent Text Reader

Abstract

The application discloses a kind of unmanned aerial vehicle vision target detection method and system based on density guide feature pyramid, method includes obtaining real-time image and pre-processing, generates standardization tensor input;From the image after pre-processing, extract multi-level spatial and semantic features;Through density distribution modeling mechanism, provide spatial adjustment signal for feature pyramid, use density map to guide multi-scale fusion;Mechanism including predicting small target density, encoding into dynamic weight and self-adaptive fusion feature layer is constructed, and spatial and semantic of multi-scale feature fusion are double self-adaptive optimization;Multi-scale branch prediction and density enhancement confidence fusion are carried out, the multi-scale feature map output by density guide feature pyramid is converted into final detection result;Model is iteratively trained, while optimizing inference process.The application introduces small target density map prediction branch, and self-adaptive adjustment is carried out to multi-scale feature fusion process, which significantly improves the detection performance of unmanned aerial vehicle.
Need to check novelty before this filing date? Find Prior Art