Small target detection method, device and medium based on improved YOLOv5s network

By improving the YOLOv5s network and adopting the CMBFF module, Adaptive-MSDiFE and BiFormer mechanism, the feature representation and generalization ability are enhanced, solving the accuracy and robustness problems of small object detection in complex scenes, and achieving efficient small object detection.

CN120997731BActive Publication Date: 2026-06-09NANJING HUAKANG INTELLIGENT TECH CO LTD +1
View PDF 2 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
NANJING HUAKANG INTELLIGENT TECH CO LTD
Filing Date
2025-08-06
Publication Date
2026-06-09

AI Technical Summary

Technical Problem

Existing technologies suffer from low detection accuracy, poor real-time performance, and weak robustness in small target detection, especially in complex scenarios where it is difficult to effectively capture the key features of small targets and improve detection accuracy.

Method used

The CMBFF composite multi-branch feature fusion module replaces the CSP1_1 module of the YOLOv5s backbone network. Combined with the Adaptive-MSDiFE adaptive multi-scale dilated deep feature enhancement mechanism and the BiFormer bidirectional attention mechanism, the GCRCSA-AFPN feature fusion module is used to replace the FPN feature fusion module, thereby enhancing feature representation and generalization ability.

Benefits of technology

It improves the accuracy and precision of small target detection, enables the model to understand local and global context, enhances detection performance in complex backgrounds, and reduces the number of model parameters.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN120997731B_ABST
    Figure CN120997731B_ABST
Patent Text Reader

Abstract

The application provides a small target detection method, device and medium based on an improved YOLOv5s network, the small target detection method is optimized by taking a YOLOv5s network model as a detection model, a CMBFF composite multi-branch feature fusion module is used to replace a first CSP1_1 module in a backbone network to extract backbone features, features of different scales are fused to improve the expression ability of the model, and multi-scale information in an image can be effectively captured; in the neck network, an Improved BRA based on an Adaptive-MSDiFE adaptive multi-scale expansion depth feature enhancement mechanism and a dynamic scaling attention factor, and a GCRCSA-AFPN fused feature map are used, key region features are strengthened, a Top-k routing strategy is introduced to reduce redundant calculation amount, a self-attention enhancement mechanism is used to enhance the ability of identifying occluded targets by dynamically weighting features based on a gating mechanism, and residual connection is used to retain original details, and the small target detection precision and robustness in a complex scene are significantly optimized.
Need to check novelty before this filing date? Find Prior Art