Shallow sea water depth remote sensing extraction method based on physical constraint deep learning

By employing a physical constraint-based deep learning method, combined with multi-scale input and spectral reconstruction branches, the accuracy and reliability issues of satellite remote sensing depth in deep water areas were resolved, enabling high-precision underwater terrain extraction and cross-regional applications.

CN121937901APending Publication Date: 2026-04-28NANJING UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
NANJING UNIV
Filing Date
2026-01-22
Publication Date
2026-04-28

AI Technical Summary

Technical Problem

Existing satellite remote sensing depth sensing technologies suffer from reduced accuracy, physical unreliability, and weak generalization ability in deep water areas (20-30 meters), especially in complex water environments such as coral reefs and clear nearshore waters.

Method used

We employ a physical constraint-based deep learning approach, extracting multi-level feature maps through multi-scale input tensors and hierarchical transformers. We combine water depth prediction and spectral reconstruction branches, use a composite loss function for end-to-end training, and introduce physical constraints to ensure that the model conforms to the laws of marine optics.

Benefits of technology

It achieves high-precision and stable underwater terrain extraction across the entire water depth range, possesses excellent cross-regional generalization ability and interpretability, solves the problem of sharp drop in accuracy caused by signal attenuation in deep water areas, and improves the reliability and generalization ability of the model.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN121937901A_ABST
    Figure CN121937901A_ABST
Patent Text Reader

Abstract

The invention relates to the field of shallow sea underwater terrain inversion, and discloses a shallow sea water depth remote sensing extraction method based on physical constraint deep learning, and the method comprises the steps: constructing a unified multi-scale input tensor; inputting the constructed multi-scale input tensor into a backbone network of a hierarchical converter, and extracting a multi-level feature map; inputting the extracted multi-level feature map into a water depth prediction branch and a spectrum reconstruction branch in parallel, and outputting a corresponding prediction result to construct a physical constraint condition of underwater terrain extraction; and constructing an underwater terrain extraction network by combining a multi-scale input tensor, a multi-level feature map, a corresponding prediction result and a physical constraint condition, and performing end-to-end training and optimization on the underwater terrain extraction network by using a composite loss function. According to the method, a spectrum reconstruction branch based on a physical model is introduced as a strong constraint, so that the problem that the precision is suddenly reduced due to signal attenuation of a traditional data driving model in a deep water area of 20-30 meters is solved.
Need to check novelty before this filing date? Find Prior Art