A multi-modal feature fusion water area target detection method

By employing a multimodal feature fusion method, utilizing quaternion optical flow stabilization and mirror mask separation, combined with the optimal transmission algorithm and conformal saliency ratio determination, the problem of high false detection rate in target detection in aquatic environments is solved, and the stability and robustness of detection are improved. This method is applicable to scenarios such as unmanned surface vessels, port monitoring, and underwater robots.

CN121392252BActive Publication Date: 2026-06-09LIAOCHENG UNIV
View PDF 2 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
LIAOCHENG UNIV
Filing Date
2025-11-14
Publication Date
2026-06-09

AI Technical Summary

Technical Problem

Existing single-modal image detection models exhibit significant performance degradation in aquatic environments due to lighting interference, reflection disturbances, and dynamic water backgrounds, resulting in high false detection rates and difficulty in achieving stable target detection.

Method used

By employing a multimodal feature fusion method, a stable image frame is generated using quaternion optical flow stabilization. Mirror perturbations are separated and a mirror mask and motion cue map are generated. Features are aligned and fused using the optimal transmission algorithm, and closed-loop verification is performed through a conformal saliency ratio judgment mechanism to achieve self-verification and dynamic correction of target detection.

Benefits of technology

It significantly reduces the false positive and false negative rates, enhances the robustness and continuous frame consistency of the model in complex aquatic environments, and is suitable for scenarios such as unmanned surface vessels, port monitoring, and underwater robots.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN121392252B_ABST
    Figure CN121392252B_ABST
Patent Text Reader

Abstract

This invention belongs to the field of target detection technology, specifically relating to a multimodal feature fusion method for water target detection. The method includes the following steps in sequence: a preprocessing step, including: based on two adjacent RGB image frames as input, separating mirror perturbations to generate a mirror mask and a motion cue map; an alignment and fusion step, including: using a stable frame as the primary modality and the mirror mask and motion cue map as secondary modalities, extracting primary and secondary modal features; and inputting the fused features into a detection head to obtain initial candidate boxes and category scores; and a loop closure determination step, including: based on the conformal saliency ratio calculated for the initial candidate boxes and calibrated within the same frame, outputting the final detection result for that frame. This invention can automatically identify and eliminate unstable candidates caused by reflection. It also enhances the robustness and consistency of the model in complex scenes such as strong reflections, water wave disturbances, and changes in viewing angle.
Need to check novelty before this filing date? Find Prior Art