CN-122023122-A - Tide remote sensing time super-resolution method and system

CN122023122ACN 122023122 ACN122023122 ACN 122023122ACN-122023122-A

Abstract

The embodiment of the application provides a tidal remote sensing time super-resolution method and system, and relates to the technical field of remote sensing satellites. The method comprises the steps of firstly obtaining a satellite remote sensing image sequence, extracting a plurality of phase shooting images by using analog-to-digital operation, and then reordering and aligning the phase shooting images according to a phase time sequence to obtain a tide boundary data set. And identifying residual information by using the correction model, and correcting the tide boundary data set based on the residual information to obtain an hour-level tide dynamic observation sequence. The method can utilize non-whole day revisiting satellites to continuously observe and accumulate for many times, and achieve multi-phase complete sampling and phase reconstruction. And nonlinear characteristics and long-term dependence of tides are effectively captured through deep learning and modeling of irregular residual components. According to the method, the tidal dynamic observation sequence which accords with an astronomical period and contains real disturbance is generated through coupling phase reconstruction and deep learning, and the accuracy of tidal remote sensing observation data is improved.

Inventors

LIU ZIJING
CHENG JINXIANG
LIU SHENGQIANG
YAO HAIYUAN
ZHANG LIGUO
ZHENG CHAOHUI
ZHANG NING
GAO YUJIAN

Assignees

交通运输部规划研究院

Dates

Publication Date: 20260512
Application Date: 20251223

Claims (10)

1. A tidal remote sensing temporal super resolution method, the method comprising: The method comprises the steps of acquiring a satellite remote sensing image sequence, wherein the satellite remote sensing image sequence comprises a plurality of remote sensing images obtained by remotely sensing an observation tidal region by an observation satellite, and the revisit period of the observation satellite is larger than a daily observation period and is not equal to an integer multiple of the daily observation period; extracting a plurality of phase shooting images from the satellite remote sensing image sequence by using modulus operation, wherein the plurality of phase shooting images are images obtained by remote sensing observation of the observation tidal region at different times of a plurality of daily observation periods; Generating a tidal boundary dataset from a plurality of said phase-captured images, said tidal boundary dataset being an hour-level discrete sequence obtained by reordering and aligning said phase-captured images at a plurality of times within a same day-level observation period; Identifying residual information in the tidal boundary dataset by using a correction model, wherein the correction model is a Informer-architecture-based deep learning network model; Correcting the tidal boundary dataset based on the residual information to obtain an hour level tidal dynamic observation sequence.
2. The method of claim 1, wherein acquiring a satellite remote sensing image sequence comprises: Acquiring a revisit period of the observation satellite; Extracting a phase shift feature based on the revisit period, the phase shift feature including a plurality of observation moments; Constructing a time super-resolution sampling mechanism in the daily observation period according to the phase shift characteristic; and acquiring the satellite remote sensing image sequence according to the time super-resolution sampling mechanism.
3. The method of claim 2, wherein constructing a temporal super-resolution sampling scheme within the daily observation period from the phase shift characteristic comprises: acquiring initial transit time of the observation satellite in the observation tide area; Calculating revisit time according to the initial border passing time and the revisit period, wherein the revisit time is obtained by calculating the product of border passing sequence and revisit period and taking the module of the daily observation period; And setting remote sensing observation time of the observation satellite according to the revisit time to generate the time super-resolution sampling mechanism.
4. The method of claim 1, wherein generating a tidal boundary dataset from a plurality of the phase captured images comprises: reading a shooting time stamp of the phase shooting image; sorting the plurality of phase captured images according to the capture time stamps to generate a time sequence; extracting waterway boundaries in the phase shooting images; According to the time sequence, carrying out phase alignment on waterway boundaries in the plurality of phase shooting images so as to synthesize a tide motion track of a full-day observation period; An hour-level discrete sequence is constructed in accordance with the tidal motion profile to generate the tidal boundary dataset.
5. The method of claim 1, wherein identifying residual information in the tidal boundary dataset using a correction model comprises: Acquiring auxiliary data, wherein the auxiliary data comprises meteorological data and historical observation data, and the historical observation data comprises a tide level observation sequence which is different from the satellite remote sensing image sequence source; generating joint input data according to the tide boundary data set and the auxiliary data, wherein the joint input data comprises a joint input sequence formed by splicing the tide boundary data set and the tide level observation sequence; Inputting the joint input data into the correction model to extract dominant periodic information from the joint input sequence through the correction model, wherein the dominant periodic information comprises a periodic driving item and a disturbance residual item; And generating the residual information according to the dominant period information.
6. The method of claim 5, wherein inputting the joint input data into the correction model to extract dominant periodic information from the joint input sequence by the correction model comprises: performing encoding on the joint input sequence using an encoder of the correction model to obtain an output tensor of a last layer of the encoder; Performing fast fourier transform on the output tensor along a time dimension to obtain a frequency domain power spectrum, wherein the frequency domain power spectrum comprises spectrum powers of a plurality of channels; Identifying periodic frequency components from the frequency domain power spectrum, wherein the periodic frequency components are obtained by averaging or principal component analysis of spectrum powers of a plurality of channels in the frequency domain power spectrum; dominant period information is determined from the period frequency components, the dominant period information including different types of tidal periods and atypical periods caused by climate disturbances.
7. The method of claim 6, wherein smoothing the tidal boundary dataset based on the residual portion to obtain an hour-level tidal dynamic observation sequence comprises: determining a to-be-corrected item from the tide boundary dataset according to the residual information, wherein the to-be-corrected item comprises an abnormal value and a missing value; constructing a periodic function based on the periodic frequency component, and acquiring a predicted value output by the correction model; Calculating a correction value according to the periodic function and the predicted value; correcting the item to be corrected in the tidal boundary dataset using the correction values to obtain the hour level tidal dynamic observation sequence.
8. The method of claim 1, wherein smoothing the tidal boundary dataset based on the residual portion to obtain an hour-level tidal dynamic observation sequence comprises: determining outliers and missing values from the tidal boundary dataset according to the residual information; Acquiring a correction reference sequence, wherein the correction reference sequence is an image sequence obtained by performing remote sensing observation on the observation tidal region by a satellite different from the observation satellite; Inputting the tidal boundary dataset and the corrective reference sequence into the correction model such that the correction model calculates a predicted value by sharing a periodic structure with an offset modeling; Acquiring a preset correction intensity coefficient; Residual correction is carried out according to the correction intensity coefficient, the abnormal value and the predicted value so as to generate a normal value; Adding said predicted value and said normal value to said tidal boundary dataset according to the location of said missing value and said outlier in said tidal boundary dataset to obtain said hour level tidal dynamic observation sequence.
9. The method of claim 8, wherein inputting the tidal boundary dataset and the corrective reference sequence into the correction model to cause the correction model to calculate a predicted value by sharing a periodic structure with offset modeling comprises: Extracting irregular characteristics through the correction model, wherein the irregular characteristics comprise long-range related characteristics and local pulsation characteristics; modeling irregular residual errors according to the irregular characteristics, and acquiring theoretical astronomical tide and remote sensing water judging line errors; reconstructing according to the theoretical astronomical tide, the irregular residual error and the remote sensing waterline judging error to obtain an hour-level boundary tide level sequence; predicting a residual value of an future daily observation period based on the hour level boundary tide level sequence; and constructing a correction function according to the residual value and the theoretical astronomical tide, and calculating the predicted value by using the correction function.
10. A tidal remote sensing temporal super resolution system, the system comprising: The system comprises an image acquisition module, a satellite remote sensing image sequence, a storage module and a display module, wherein the satellite remote sensing image sequence comprises a plurality of remote sensing images obtained by remotely observing an observation tidal region by an observation satellite, and the revisit period of the observation satellite is larger than the daily observation period and is not equal to the integral multiple of the daily observation period; The image extraction module is used for extracting a plurality of phase shooting images in the satellite remote sensing image sequence by using modulus operation, wherein the plurality of phase shooting images are images obtained by remote sensing observation of the observation tidal region at different times of a plurality of daily observation periods; A sequence generation module, configured to generate a tidal boundary data set according to a plurality of the phase-captured images, where the tidal boundary data set is an hour-level discrete sequence obtained by reordering and aligning the phase-captured images at a plurality of times in a same daily-level observation period; the residual analysis module is used for identifying residual information in the tide boundary data set by using a correction model, wherein the correction model is a depth learning network model based on Informer architecture; And the correction module is used for correcting the tide boundary data set based on the residual information so as to obtain an hour-level tide dynamic observation sequence.

Description

Tide remote sensing time super-resolution method and system Technical Field The application relates to the technical field of remote sensing satellites, in particular to a tidal remote sensing time super-resolution method and system. Background Remote sensing observation refers to an area monitoring technology for acquiring surface images or surface characteristic data of a specific area through aircrafts such as satellites. Taking the remote sensing satellite technology as an example, the remote sensing satellite can utilize a satellite sensor to observe the earth from space so as to acquire large-range and high-resolution monitoring data, and the remote sensing satellite becomes an indispensable means for monitoring the change of the earth surface environment. One technical indicator of remote sensing observation is the time resolution of satellite data. But is limited by satellite orbit dynamics and research cost, the revisiting period of the observation satellite is longer, namely, the interval time of the satellite flying to the same area is longer, so that the time resolution of satellite data obtained by the observation satellite is on the daily level or multiple daily levels. The theoretical revisitation period, such as WorldView-3, is 1.7 days. For many critical scientific research fields, the long revisit period can make it difficult for most satellites to capture the fast dynamic processes that occur on the earth's surface. For example, in pollution source diffusion tracking in environmental monitoring, disaster rapid assessment in emergency response, and tidal dynamic observation in marine science, observation data with time resolution reaching an hour scale are required. In order to improve the time resolution, mathematical interpolation methods such as linear interpolation may be used to estimate the observed data of the missing period. However, the tide is a highly complex nonlinear system, not only has astronomical period, but also is influenced by multiple factors such as 'day inequality', meteorological disturbance, topography and the like, and linear interpolation and other methods completely ignore the nonlinear characteristics, so that the accuracy of an estimated result is seriously insufficient, the fluctuation process of the tide cannot be truly reflected, and the authenticity and accuracy of data are reduced. Disclosure of Invention In view of the above, the embodiment of the application provides a tidal remote sensing time super-resolution method and a tidal remote sensing time super-resolution system, which are used for solving the problem of insufficient accuracy of remote sensing observation data. According to a first aspect of the present application there is provided a tidal remote sensing temporal super resolution method, the method comprising: The method comprises the steps of acquiring a satellite remote sensing image sequence, wherein the satellite remote sensing image sequence comprises a plurality of remote sensing images obtained by remotely sensing an observation tidal region by an observation satellite, and the revisit period of the observation satellite is larger than a daily observation period and is not equal to an integer multiple of the daily observation period; extracting a plurality of phase shooting images from the satellite remote sensing image sequence by using modulus operation, wherein the plurality of phase shooting images are images obtained by remote sensing observation of the observation tidal region at different times of a plurality of daily observation periods; Generating a tidal boundary dataset from a plurality of said phase-captured images, said tidal boundary dataset being an hour-level discrete sequence obtained by reordering and aligning said phase-captured images at a plurality of times within a same day-level observation period; Identifying residual information in the tidal boundary dataset by using a correction model, wherein the correction model is a Informer-architecture-based deep learning network model; Correcting the tidal boundary dataset based on the residual information to obtain an hour level tidal dynamic observation sequence. In some embodiments, acquiring a sequence of satellite remote sensing images includes: Acquiring a revisit period of the observation satellite; Extracting a phase shift feature based on the revisit period, the phase shift feature including a plurality of observation moments; Constructing a time super-resolution sampling mechanism in the daily observation period according to the phase shift characteristic; and acquiring the satellite remote sensing image sequence according to the time super-resolution sampling mechanism. In some embodiments, constructing a temporal super-resolution sampling mechanism within the daily observation period from the phase shift characteristic comprises: acquiring initial transit time of the observation satellite in the observation tide area; Calculating revisit time according to th