A method for detecting small targets based on perception sampling attention of a transformer

By employing a Transformer-based perceptual sampling attention-based small target detection method, and utilizing the Target-Focused MetaFormer Block and feature pyramid network, the problems of insufficient local feature representation and background noise interference in small target detection are solved, achieving high-precision detection of small targets in complex backgrounds.

CN122289651APending Publication Date: 2026-06-26SHENYANG UNIVERSITY OF TECHNOLOGY
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
SHENYANG UNIVERSITY OF TECHNOLOGY
Filing Date
2026-03-27
Publication Date
2026-06-26

AI Technical Summary

Technical Problem

Existing target detection methods struggle to effectively detect small targets, especially those smaller than 16×16 pixels, in complex backgrounds. They suffer from insufficient local feature representation, are susceptible to background noise interference, and have poor multi-scale feature fusion performance, resulting in inadequate detection accuracy and robustness.

Method used

We adopt a Transformer-based perceptual sampling attention-based small target detection method. By enhancing the feature map through the Target-Focused MetaFormer Block and combining it with a local spatial enhancement network and a feature pyramid network, we can achieve cross-scale feature interaction and target query optimization, thereby improving the robustness and accuracy of small target detection.

Benefits of technology

It significantly improves the detection accuracy and robustness of small targets in complex backgrounds, effectively suppresses background noise interference, enhances local feature representation, adapts to changes in target scale, and improves the accuracy and stability of small target detection.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122289651A_ABST
    Figure CN122289651A_ABST
Patent Text Reader

Abstract

This invention provides a Transformer-based perceptual sampling attention method for small target detection, belonging to the field of small target detection technology. It utilizes deep learning technology to achieve small target detection in UAV images. Specifically, the method includes: using a public dataset; expanding the dataset using data augmentation methods and dividing it into training, testing, and validation sets; constructing a Transformer-based perceptual sampling attention small target detection model, introducing a Target-Focused MetaFormer Block to suppress complex background noise and highlight target-related representations; training the constructed model using the training and validation sets; and using the trained model to detect small targets in UAV images. This invention can accurately capture local details of small targets, enhance feature discernibility in complex backgrounds, and improve multi-scale contextual representation capabilities, achieving precise localization of small targets in UAV images.
Need to check novelty before this filing date? Find Prior Art