An adaptive matching method for deep and shallow features

By coordinating the contributions of deep and shallow features through adaptive weighting operations, the problem of limited performance in small target detection in deep neural networks is solved, achieving more efficient detection results.

CN116681981BActive Publication Date: 2026-06-26JIANGSU UNIV OF SCI & TECH
View PDF 2 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
JIANGSU UNIV OF SCI & TECH
Filing Date
2023-03-17
Publication Date
2026-06-26

AI Technical Summary

Technical Problem

In small target detection, the rigid splicing of deep and shallow features in existing deep neural networks cannot fully coordinate their contributions, thus limiting the detection performance.

Method used

An adaptive weighted mask generation submodule is adopted to coordinate the contribution of deep and shallow features through adaptive weighting operations, thereby achieving feature matching.

Benefits of technology

It significantly improves the accuracy of small target detection, reduces false detections and missed detections, and enhances detection performance.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN116681981B_ABST
    Figure CN116681981B_ABST
Patent Text Reader

Abstract

The application discloses a kind of self-adapting matching methods for deep, shallow feature, as follows: data preprocessing, data set is converted into the format that network model can operate;Model selection, select appropriate single-stage target detection detection network;Model editing, by adding deep, shallow feature matching module in model FPN structure, use difference mask to weight deep, shallow feature, coordinate the contribution of different level features to terminal prediction model;Model training, use binary cross-entropy loss and IOU loss to optimize the model;Model evaluation, use the trained model to predict the class and position of test sample.The application has higher detection target precision, lower miss detection rate, and can effectively solve the problem of insufficient detection precision of single-stage target detection model for small targets.
Need to check novelty before this filing date? Find Prior Art

Citation Information

Patent Citations

  • Target detection method based on deep feature and shallow feature fusion

    CN113887649A

  • Fire detection model training method and detection method based on improved DETR

    CN114639007A