CN-121561127-B - Method and system for creating multi-temporal remote sensing image database

CN121561127BCN 121561127 BCN121561127 BCN 121561127BCN-121561127-B

Abstract

The invention discloses a method and a system for creating a multi-temporal remote sensing image database, which relate to the technical field of remote sensing image data processing and comprise the following steps that S1, multi-temporal remote sensing image data are received as input; and S2, carrying out primary quality evaluation on each image, wherein the evaluation is used for constructing a multidimensional quality feature vector by extracting radiation statistical features, cloud index and spectral correlation with a reference image. The method and the system for creating the multi-temporal remote sensing image database solve the core technical problems that manual intervention is relied on, the processing efficiency is low and the radiation consistency is difficult to guarantee in the traditional multi-temporal remote sensing image database creation process, realize standardized and automatic management and control of the radiation quality of massive image data through an established complete quality guarantee system from primary evaluation to final judgment, reduce the labor cost and improve the database construction efficiency.

Inventors

HOU FUCHENG
SHEN YUAN
ZHANG HENG
BAI JIE

Assignees

山东产研卫星信息技术产业研究院有限公司

Dates

Publication Date: 20260508
Application Date: 20260123

Claims (10)

1. The method for creating the multi-temporal remote sensing image database is characterized by comprising the following steps of: s1, receiving multi-temporal remote sensing image data as input; S2, carrying out primary quality assessment on each image, wherein the assessment constructs a multidimensional quality feature vector by extracting radiation statistical features, cloud quantity indexes and spectral correlation with a reference image, and analyzing the vector by using a pre-trained lightweight classifier to output an initial consistency score; S3, intelligent shunting is executed based on the initial consistency score, the images are divided into high-confidence passing streams, to-be-processed streams and high-confidence failure streams, wherein the images of the high-confidence passing streams directly enter a secondary verification queue, the images of the to-be-processed streams flow into a strategy execution engine, and the images of the high-confidence failure streams are marked as unqualified and recorded; S4, selecting a radiation normalization algorithm to perform primary treatment on the image of the flow to be treated through a strategy execution engine, and generating first-generation treated data, wherein the selection is based on the specific composition of the multidimensional quality feature vector to adapt one algorithm from a plurality of radiation normalization algorithms; S5, inputting the image of the high confidence passing stream and the first generation processed data into a secondary verification queue together, and performing secondary verification to obtain a secondary verification result, wherein the secondary verification adopts a depth verification model to calculate a new round of consistency score, the depth verification model is different from a model of primary quality evaluation, and the depth verification model comprises local verification on a preset stable ground object sample area and is used for ensuring that the processed image and the reference image keep stable spectrum on a typical ground object; s6, performing secondary shunting based on a secondary verification result, namely approving the data passing through the secondary verification into a warehouse, and returning the data not passing through the secondary verification to a strategy execution engine and attaching a diagnosis report; S7, analyzing and processing failure reasons according to the diagnosis report, and switching or combining another radiation normalization algorithm to perform secondary processing to generate second-generation processed data, wherein in the secondary processing process, the strategy execution engine selects the radiation normalization algorithm with high calculation complexity or combines multiple radiation normalization algorithms to process according to failure reason analysis in the diagnosis report; And S8, submitting the processed data of the second generation to a final arbiter, carrying out final evaluation by using a physical spectrum model, outputting a final evaluation result, storing the passed data into a multi-time-phase remote sensing image database only if the final evaluation result is passed or not, marking the failed data as difficult data, and terminating the automatic flow.
2. The method for creating the multi-temporal remote sensing image database according to claim 1, wherein the extracting of the radiation statistical features of the images in the primary quality assessment comprises calculating the mean, variance and histogram distribution of the images, the extracting of cloud cover indexes of the images comprises identifying cloud cover areas and calculating cloud cover proportions by an image segmentation technique, the extracting of the spectral correlation of the images and the reference images comprises calculating correlation coefficients of the two images on the same band; The lightweight classifier is a support vector machine or a decision tree model, and the pre-training process uses historical multi-temporal remote sensing image data and corresponding radiation consistency labels to conduct supervised learning; In the intelligent diversion, the judgment condition of the high-confidence passing flow is that the initial consistency score is higher than a first threshold, the judgment condition of the flow to be processed is that the initial consistency score is between the first threshold and a second threshold, and the judgment condition of the high-confidence failing flow is that the initial consistency score is lower than the second threshold, wherein the first threshold and the second threshold are preset through historical data statistical analysis.
3. The method for creating the multi-temporal remote sensing image database according to claim 1, wherein the process of selecting the radiation normalization algorithm by the strategy execution engine comprises the steps of matching a predefined algorithm selection rule according to radiation statistical characteristics and cloud cover indexes in the multi-dimensional quality characteristic vector, wherein the rule maps the characteristic vector to a specific algorithm; And after the primary processing generates the first-generation processed data, recording the used algorithm and parameters thereof, wherein the strategy execution engine is also internally provided with an algorithm performance evaluation module which updates algorithm selection rules according to historical processing effects so as to optimize the subsequent processing efficiency.
4. The method for creating the multi-temporal remote sensing image database according to claim 1, wherein the depth verification model used in the secondary verification is a convolutional neural network model or a random forest model, and the training data comprises multi-temporal remote sensing images and spectrum stability labels thereof on a stable ground object sample area; The stable ground object sample area is a preselected ground surface coverage type stable area and comprises a water body, vegetation and bare land, local inspection is carried out by comparing the spectral reflectance values of the processed image and the reference image on the sample area, when the deviation exceeds a preset tolerance value, the image is judged to not pass the secondary inspection, in the secondary inspection, the data passing the secondary inspection are required to meet that a new round of consistency score is higher than a third threshold value, the local spectral deviation is lower than the preset tolerance value, and the diagnosis report attached to the data failing the secondary inspection comprises failure cause analysis, recommended processing algorithm and parameter adjustment suggestion.
5. The method for creating the multi-temporal remote sensing image database according to claim 1, wherein the combination of the plurality of radiation normalization algorithms comprises preprocessing by using a histogram matching algorithm, and adjusting by using a normalization algorithm based on pseudo-invariant features; the physical spectrum model used by the final arbiter is a model based on an atmospheric radiation transmission theory, the model simulates atmospheric conditions when an image is acquired, and calculates residual errors of theoretical spectrum values and actual values, when the final evaluation is carried out, the residual errors are lower than a preset threshold value, the passing of the model is judged, and otherwise, the passing of the model is judged.
6. The method for creating a multi-temporal remote sensing image database according to claim 1, wherein the evaluation result of the final resolver is used to optimize a lightweight classifier for primary quality evaluation and a deep verification model for secondary verification, the optimization is performed by machine learning, and the model parameters are updated using data passed by the final resolver as positive samples.
7. The method for creating the multi-temporal remote sensing image database according to claim 1, wherein the multi-temporal remote sensing image database is hierarchically structured and comprises an original data layer, a processed intermediate data layer and a finished product data layer, wherein the original data layer stores the input multi-temporal remote sensing image, the processed intermediate data layer stores primary quality evaluation results, first-generation processed data and second-generation processed data, the finished product data layer stores data passed by a final resolver, each layer of data is associated with metadata including acquisition time, sensor type, processing history and radiation consistency score, and the database further provides a query interface supporting retrieval based on a time range, a geographic area and a radiation consistency level.
8. A system for creating a multi-temporal remote sensing image database, for implementing a method for creating a multi-temporal remote sensing image database according to any one of claims 1 to 7, comprising: The data receiving module is used for receiving the multi-temporal remote sensing image data; the primary quality evaluation module is used for carrying out primary quality evaluation on each image, constructing a multi-dimensional quality feature vector by evaluating the radiation statistical feature, cloud quantity index and spectral correlation with a reference image of the extracted image, and analyzing the vector by using a pre-trained lightweight classifier to output an initial consistency score; The intelligent distribution module is used for executing intelligent distribution based on the initial consistency score, dividing the image into a high-confidence passing stream, a stream to be processed and a high-confidence failure stream, wherein the image of the high-confidence passing stream directly enters a secondary verification queue, the image of the stream to be processed flows into a strategy execution engine, and the image of the high-confidence failure stream is marked as unqualified and recorded; The strategy execution engine module is used for carrying out primary processing by selecting a radiation normalization algorithm through the strategy execution engine and generating first-generation processed data, wherein the selection is based on the specific composition of the multidimensional quality feature vector and is used for adapting an algorithm from a plurality of radiation normalization algorithms; The secondary verification module inputs the image of the high confidence passing stream and the first generation processed data into a secondary verification queue together for secondary verification to obtain a secondary verification result, the secondary verification calculates a new round of consistency score by adopting a depth verification model, and the depth verification model is different from a model of primary quality evaluation and comprises the steps of carrying out local verification on a preset stable ground object sample area so as to ensure that the processed image and the reference image keep spectrum stable on a typical ground object; the secondary shunting module is used for executing secondary shunting based on a secondary verification result, namely approving the data passing through the secondary verification to be put in storage, and returning the data not passing through the secondary verification to a strategy execution engine and attaching a diagnosis report; The secondary processing module is used for analyzing and processing failure reasons according to the diagnosis report on the data returned to the strategy execution engine, and switching or combining another radiation normalization algorithm to perform secondary processing to generate second-generation processed data; the final arbiter module submits the second generation processed data to a final arbiter, the final arbiter uses a physical spectrum model to carry out final evaluation, and outputs a final evaluation result, wherein the final evaluation result is only passed or not passed, and the failed data is marked as difficult data and the automatic flow is terminated; And the storage module is used for storing the data which pass through the final evaluation to a multi-time-phase remote sensing image database.
9. The system for creating the multi-temporal remote sensing image database according to claim 8, wherein the primary quality assessment module further comprises a feature extraction unit and a classifier unit, the feature extraction unit is configured to calculate radiation statistical features, cloud cover indexes and spectrum correlations in parallel, the classifier unit is periodically retrained with new data to maintain accuracy, the policy execution engine module comprises an algorithm library unit and a selection unit, the algorithm library unit stores a plurality of radiation normalization algorithms, the selection unit selects algorithms in real time according to multi-dimensional quality feature vectors, the secondary verification module comprises a depth model unit and a local verification unit, the depth model unit loads a pre-trained depth verification model, the local verification unit accesses a geographic information system to obtain coordinates of a stable ground object sample area, and the final arbiter module integrates physical spectrum simulation software which calculates theoretical spectrum values based on atmospheric parameters.
10. The system for creating the multi-temporal remote sensing image database according to claim 8, further comprising a monitoring and management module for displaying the state of the processing pipeline in real time, wherein the monitoring and management module comprises input and output data quantity, processing time and error logs of each module, and is used for providing a graphical user interface for manually adjusting a shunt threshold value, an algorithm selection rule and verification parameters, recording all operations through the log database by the system and supporting fault recovery and performance analysis, wherein the storage module adopts a distributed file system and supports multi-node parallel access so as to improve the processing efficiency under a large data quantity, and integrating the system with an external remote sensing data processing platform through an application programming interface to realize data sharing and collaborative processing.

Description

Method and system for creating multi-temporal remote sensing image database Technical Field The invention relates to the technical field of remote sensing image data processing, in particular to a method and a system for creating a multi-temporal remote sensing image database. Background With the rapid development of earth observation technology, the demands of various industries on multi-temporal remote sensing image data are increasing, and the construction of large-scale and high-quality multi-temporal remote sensing image databases has become a necessary trend. However, in the existing database creation method, the technical emphasis is generally focused on the basic functions of efficient storage, rapid indexing, query management and the like of the images, but a key link affecting the final value of the data is ignored, namely, how to automatically guarantee the core quality of the radiation consistency between the warehouse-in images. Current practice often relies on either heavy manual preprocessing prior to library construction or the default use of corrected data sources, which not only makes the overall process inefficient, severely relies on expert personal experience, but also results in inherent radiation differences at the bottom of the constructed database. After the noise of the real change of the non-ground object is brought into the database, subsequent high-order applications such as change detection, time sequence analysis and the like can be seriously misled, and the problem is often found after the database is put into use, so that huge reworking cost and analysis deviation are caused. Therefore, the current technology system lacks an embedded and intelligent process to automatically identify and process the problem of inconsistent radiation, which has become a major obstacle to the automatic construction of a high-quality multi-temporal remote sensing image database. The invention aims to solve the practical problem that the conventional method cannot intelligently ensure the radiation consistency of the warehouse-in images in the process of automatically creating the multi-time-phase remote sensing image database. Disclosure of Invention The invention aims to provide a method and a system for creating a multi-temporal remote sensing image database, which are used for solving the problems in the background technology. In order to solve the technical problems, the invention provides a method for creating a multi-temporal remote sensing image database, which comprises the following steps: s1, receiving multi-temporal remote sensing image data as input; S2, carrying out primary quality assessment on each image, wherein the assessment constructs a multidimensional quality feature vector by extracting radiation statistical features, cloud quantity indexes and spectral correlation with a reference image, and analyzing the vector by using a pre-trained lightweight classifier to output an initial consistency score; S3, intelligent shunting is executed based on the initial consistency score, the images are divided into high-confidence passing streams, to-be-processed streams and high-confidence failure streams, wherein the images of the high-confidence passing streams directly enter a secondary verification queue, the images of the to-be-processed streams flow into a strategy execution engine, and the images of the high-confidence failure streams are marked as unqualified and recorded; S4, selecting a radiation normalization algorithm to perform primary processing on the image of the flow to be processed through a strategy execution engine, generating first-generation processed data, and adapting an algorithm from a plurality of radiation normalization algorithms based on specific composition of the multidimensional quality feature vector; S5, inputting the image of the high confidence passing stream and the first generation processed data into a secondary verification queue together, and performing secondary verification to obtain a secondary verification result, wherein the secondary verification adopts a depth verification model to calculate a new round of consistency score, the depth verification model is different from a model of primary quality evaluation, and the depth verification model comprises local verification on a preset stable ground object sample area and is used for ensuring that the processed image and the reference image keep stable spectrum on a typical ground object; s6, performing secondary shunting based on a secondary verification result, namely approving the data passing through the secondary verification into a warehouse, and returning the data not passing through the secondary verification to a strategy execution engine and attaching a diagnosis report; S7, analyzing and processing failure reasons according to the diagnosis report on the data returned to the strategy execution engine, and switching or combining another radiation normalization algorithm to perform se