A water surface small target detection method and system based on adaptive contrast quality enhancement

CN122156582APending Publication Date: 2026-06-05WUHAN UNIV OF TECH

Patent Information

Authority / Receiving Office: CN · China
Patent Type: Applications(China)
Current Assignee / Owner: WUHAN UNIV OF TECH
Filing Date: 2026-02-10
Publication Date: 2026-06-05

AI Technical Summary

Technical Problem

Existing technologies struggle to stably enhance the features of small targets and achieve accurate localization against complex water surface backgrounds, resulting in insufficient detection accuracy and recall for small targets on the water surface.

Method used

An adaptive contrast quality enhancement module and a quality localization attention module are introduced. Combined with a multi-scale detection structure, a dual enhancement strategy of geometry and spectral is used to process water surface images, enhance features and suppress background interference, and an improved YOLO model is used for multi-scale target detection.

Benefits of technology

It effectively enhances the characteristics of small targets on the water surface, suppresses background interference, improves detection accuracy and recall, and achieves robust detection of small targets on the water surface.

✦ Generated by Eureka AI based on patent content.

Smart Images

Figure CN122156582A_ABST

Patent Text Reader

Abstract

The application discloses a water surface small target detection method based on adaptive contrast quality enhancement, which comprises the following steps: performing scale normalization processing on a water surface image, and performing data expansion through a geometric and spectral double enhancement strategy to obtain a feature map; inputting the feature map into a main feature extraction network of an improved YOLO model for feature enhancement extraction, wherein the main feature extraction network comprises an adaptive contrast quality enhancement module for performing multi-scale convergence and contrast enhancement processing, and a quality positioning attention module for performing quality perception re-marking; sending the extracted features into a feature fusion and detection head branch component of the improved YOLO model which is constructed with four scales of P2, P3, P4 and P5, wherein the P2 scale is a high-resolution detection head newly added for extremely small target detection; performing post-processing on the prediction results output by the four scale detection heads, and outputting target detection results. The application can effectively improve the detection precision and recall rate of the water surface small target.

Need to check novelty before this filing date? Find Prior Art