Remote sensing image change detection method and system based on information entropy

Abstract

The invention provides a remote sensing image change detection method and system based on information entropy, and belongs to the technical field of remote sensing data processing. The method comprises the steps that remote sensing image data and auxiliary data of a target area are acquired and preprocessed, and a plurality of change sub-areas of the target area are extracted based on the preprocessed remote sensing image data and auxiliary data, extracting the characteristic change representing the natural ground feature and the characteristic change representing the artificial ground feature corresponding to each change sub-region; and performing conditional entropy calculation and geographic detector analysis, and outputting a detection result and a driving force analysis result. Two types of feature sets are introduced, natural ground feature feature changes and artificial ground feature feature changes are reflected respectively, the two types of features are matched and verified mutually, interference of abnormal information on change detection is relieved, the changes can be rapidly detected through combination of conditional entropy and a geographic detector, meanwhile, corresponding driving factor explanation is given, and the detection accuracy is improved. And scientific reference is provided for decision-making of related departments.

Description

Technical Field

[0001] The present application relates to the technical field of remote sensing data processing, and in particular to a remote sensing image change detection method and system based on information entropy. Background Art

[0002] With the acceleration of global urbanization, urban renewal has become a crucial path for optimizing spatial structure, improving the living environment, and promoting sustainable development. Urban renewal involves dynamic changes in multiple dimensions, including building demolition and reconstruction, infrastructure upgrades, and green space system optimization. Traditional manual surveying or statistical methods struggle to achieve efficient and accurate monitoring. For example, in the renovation of older residential communities, it is necessary to monitor the dynamics of illegal building demolition, road widening, and the distribution of public facilities in real time. In the upgrading of industrial parks, it is necessary to quickly identify land use conversions (such as industrial land to commercial land) and changes in development intensity.

[0003] Remote sensing image change detection utilizes multi-source remote sensing images and related geospatial data covering the same surface area over different time periods. Combining the characteristics of the corresponding features with remote sensing imaging mechanisms, it employs image and graphics processing theory and mathematical modeling to identify and analyze changes in features within the area, including changes in their location, extent, nature, and state. Remote sensing image change detection technology, with its advantages of wide-area coverage, dynamic monitoring, and objective quantification, has become an indispensable technical support for urban renewal.

[0004] Based on the detection method, remote sensing image change detection can be divided into direct change detection and post-classification change detection. Direct change detection involves directly comparing remote sensing images at the pixel level at two or more time points and then determining the areas where changes have occurred. Post-classification change detection, on the other hand, involves first classifying the remote sensing images at each time point and then comparing the classification results to identify areas of change.

[0005] In essence, both direct change detection and post-classification change detection methods discover changed areas by classifying and interpreting images or processed image data. They are greatly affected by image processing methods and have poor anti-interference capabilities against abnormal information, such as the "different objects with the same spectrum" and "same objects with different spectra" phenomena. It is also difficult to give explainable reasons for the changed areas.

[0006] Therefore, it is necessary to provide an improved technical solution to the above-mentioned deficiencies in the prior art. Summary of the Invention

[0007] The purpose of this application is to provide a remote sensing image change detection method and system based on information entropy to solve or alleviate the problems existing in the above-mentioned prior art.

[0008] In order to achieve the above objectives, this application provides the following technical solutions:

[0009] This application provides a remote sensing image change detection method based on information entropy, including:

[0010] remote sensing image data and auxiliary data of a target area are acquired and preprocessed to obtain preprocessed remote sensing image data and auxiliary data; the remote sensing image data includes remote sensing images acquired at a first moment and a second moment, respectively, the first moment and the second moment being different observation times of the remote sensing image; and the acquisition time of the auxiliary data corresponds to the acquisition time of the remote sensing image data;

[0011] Based on the preprocessed remote sensing image data and auxiliary data, a plurality of changed sub-regions of the target area are extracted respectively, and a first feature set and a second feature set corresponding to each changed sub-region are extracted, wherein the first feature set is used to characterize the feature changes of natural objects, and the second feature set is used to characterize the feature changes of artificial objects;

[0012] Based on the first feature set and the second feature set of each change sub-region, respectively calculating a first conditional entropy set and a second conditional entropy set of each change sub-region;

[0013] Taking the spectral angle of each pixel in each change sub-region as the dependent variable and each feature in the first conditional entropy set and the second conditional entropy set as the explanatory factor, performing geographic detector analysis on the dependent variable and the explanatory factor, and calculating the q value corresponding to each feature;

[0014] Sort and score the q values, and output the change detection and driving force analysis results.

[0015] Preferably, the auxiliary data includes: nighttime light remote sensing data, road network data and POI data.

[0016] Preferably, the first feature set includes: differences in the Normalized Difference Vegetation Index, Enhanced Vegetation Index, Soil Adjusted Vegetation Index, Normalized Difference Water Index, and Photochemical Vegetation Index between two moments;

[0017] The second feature set includes: differences in the normalized building index, the impervious surface index, and the building land index between the two moments, and differences in various auxiliary data between the two moments.

[0018] Preferably, the remote sensing image data and auxiliary data of the target area are preprocessed, including:

[0019] Correction processing is performed on the remote sensing image data, and resampling and nearest neighbor assignment processing are performed on the auxiliary data in sequence to obtain pre-processed remote sensing image data and auxiliary data.

[0020] Preferably, based on the pre-processed remote sensing image data and auxiliary data, a plurality of changed sub-regions of the target region are extracted respectively, including:

[0021] For all pixels in the remote sensing image data and the auxiliary data, calculating the change difference between the first moment and the second moment for each feature in the first feature set and the second feature set to obtain a first change difference for each feature;

[0022] Based on the first change difference, a principal component analysis method is used to construct a difference image between two moments, and a local minimum error probability method is used to determine a threshold, thereby extracting a change pattern from the difference image;

[0023] Performing morphological partitioning on the change pattern according to connectivity to obtain a plurality of initial sub-regions;

[0024] The bounding rectangle of each initial sub-region is extracted as the final changed sub-region.

[0025] Preferably, based on the first feature set and the second feature set of each change sub-region, respectively calculating a first conditional entropy set and a second conditional entropy set of each change sub-region includes:

[0026] For each change sub-region, calculating the change difference between the first moment and the second moment for each feature in the first feature set and the second feature set to obtain a second change difference for each feature;

[0027] Calculating the conditional entropy of each feature based on the second change difference of each feature;

[0028] The conditional entropies of the features are sorted, and the top N bits with the smallest conditional entropies in the first feature set and the second feature set are extracted respectively, and combined into a first conditional entropy set and a second conditional entropy set corresponding to the first feature set and the second feature, respectively.

[0029] Preferably, after respectively calculating the first conditional entropy set and the second conditional entropy set of each change sub-region based on the first feature set and the second feature set of each change sub-region, the method further includes:

[0030] classifying the second change differences of the respective features to obtain change difference categories;

[0031] Correspondingly, the spectral angle of each pixel in each change sub-region is used as the dependent variable, and the corresponding features in the first conditional entropy set and the second conditional entropy set are used as explanatory factors to perform geographic detector analysis, and calculate the q value corresponding to each feature, specifically:

[0032] Taking the spectral angle of each pixel in each change sub-area as the dependent variable, the change difference category of each feature corresponding to the first conditional entropy set and the second conditional entropy set is input into the geographic detector for factor detection to obtain the q value corresponding to each feature.

[0033] Preferably, the q values are sorted and scored, and the change detection and driving force analysis results are output, including:

[0034] Sort the q values and use the sorted numbers as the corresponding feature values;

[0035] Summarizing the feature values in the first feature set and the second feature set to obtain the overall ranking of the first feature set and the second feature set;

[0036] Based on the total rankings of the first feature set and the second feature set of each change sub-region, all change sub-regions are judged one by one to determine the driving force of each change sub-region.

[0037] Preferably, based on the total ranking of each of the first feature set and the second feature set of each change sub-region, all change sub-regions are judged one by one to determine the driving force of each change sub-region, including:

[0038] If the total ranking of the first feature set is greater than the total ranking of the second feature set, the change in the sub-region is considered to be artificial influence;

[0039] If the total ranking of the first feature set is less than the total ranking of the second feature set, the change in the sub-region is considered to be a natural influence;

[0040] If the total ranking of the first feature set is equal to the total ranking of the second feature set, the change in the sub-region is considered to be a common influence.

[0041] This embodiment provides a remote sensing image change detection system based on information entropy, including:

[0042] a data acquisition and preprocessing unit configured to acquire remote sensing image data and auxiliary data of a target area and perform preprocessing to obtain preprocessed remote sensing image data and auxiliary data; the remote sensing image data includes remote sensing images covering the target area at a first moment and a second moment, the first moment and the second moment being different observation times of the remote sensing image; and the auxiliary data is collected at a time corresponding to the remote sensing image data;

[0043] an extraction unit configured to extract, based on the preprocessed remote sensing image data and the auxiliary data, a plurality of changed sub-regions of the target area, and extract a first feature set and a second feature set corresponding to each changed sub-region, wherein the first feature set is used to characterize the characteristics of natural objects, and the second feature set is used to characterize the characteristics of artificial objects;

[0044] A calculation unit configured to calculate a first conditional entropy set and a second conditional entropy set corresponding to each change sub-region based on the first feature set and the second feature set;

[0045] a geographic detection unit configured to perform geographic detector analysis using the spectral angle of each pixel in each change sub-region as a dependent variable and corresponding features in the first conditional entropy set and the second conditional entropy set as explanatory factors, and calculate a q value corresponding to each feature;

[0046] The output unit is configured to sort and score the q values and output the change detection and driving force analysis results.

[0047] The technical solution of the embodiment of the present application has the following beneficial effects:

[0048] This method extracts change subregions and their first and second feature sets from remote sensing imagery and auxiliary data. This method not only considers the natural characteristics of the features themselves but also incorporates environmental factors such as human activities. This method comprehensively assesses change pixels from a multidisciplinary perspective, reducing change detection bias caused by processing methods and improving the ability to resist interference from anomalous information. Furthermore, conditional entropy filters out temporal mismatches in individual change subregions, further improving the stability and accuracy of change detection. Furthermore, conditional entropy analyzes change from a qualitative, global perspective, while geographic detectors analyze change from a quantitative, local perspective. This combination considers both global and local influencing factors, improving the rationality and accuracy of attribution. From an information processing perspective, conditional entropy and geographic detectors filter the primary factors influencing change, reducing data dimensionality and information redundancy. In summary, this method can rapidly detect areas where surface land use types have changed and provide explanations of the driving factors, providing a scientific basis for urban planning and decision-making within relevant departments. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 A flowchart of a remote sensing image change detection method based on information entropy according to some embodiments of the present application is provided.

[0050] Figure 2 A flowchart of a remote sensing image change detection method based on information entropy according to some embodiments of the present application is provided.

[0051] Figure 3 This is a schematic diagram of an electronic device provided according to some embodiments of the present application.

[0052] Description of reference numerals:

[0053] 200 - electronic device, 201 - processor, 202 - communication interface, 203 - memory, 204 - communication bus. DETAILED DESCRIPTION

[0054] The embodiments of the present application are described below with reference to the accompanying drawings.

[0055] The embodiments of the present application can be applied to Figure 3 Among the electronic devices shown, the electronic device can be, but is not limited to, mobile terminals such as mobile phones, tablet computers, handheld computers, personal digital assistants (PDAs), smart home devices such as smart TVs and smart cameras, wearable devices such as smart bracelets, smart watches, and smart glasses, or other computer devices such as desktops, laptops, notebook computers, ultra-mobile personal computers (UMPCs), netbooks, and smart screens.

[0056] like Figure 3 As shown, the electronic device 200 may include one or more of the following components: a processor 201, a memory 203, a communication interface 202, and a communication bus 204. The memory 203 may be connected to the processor 201 via the bus 204. The bus can transmit data between the processor 201 and the memory 203. The bus can be divided into an address bus, a data bus, a control bus, and the like.

[0057] The processor 201 may include one or more processing cores. The processor 201 may utilize various interfaces and lines to connect various components within the entire electronic device 200. By running or executing instructions, programs, code sets, or instruction sets stored in the memory 203, and calling data stored in the memory 203, the processor 201 performs various functions of the electronic device 200 and processes data. For example, the processor 201 may include an application processor (AP), a modem processor, a CPU, a graphics processing unit (GPU), an image signal processor (ISP), a controller, a video codec, a digital signal processor (DSP), a field-programmable gate array (FPGA), a programmable logic array (PLA), and / or a neural network processing unit (NPU). Among them, the CPU mainly processes the operating system, user interface, and application programs; the GPU is responsible for rendering and drawing the content to be displayed; the NPU is used to implement artificial intelligence (AI) functions; and the modem is used to handle wireless communications. Different processing units can be independent devices or integrated into one or more processors. For example, the multiple processing units shown above are all integrated into a SoC, or the AP is a separate semiconductor chip and the other processing units are integrated into a SoC. This application is not limited to this.

[0058] The memory 203 may include random access memory (RAM), read-only memory (ROM), or non-transitory computer-readable storage medium. The memory 203 may be used to store instructions, programs, codes, code sets, or instruction sets. The memory 203 may include a program storage area and a data storage area. The program storage area may store instructions for implementing an operating system and instructions for at least one function, such as a remote sensing image change detection method based on information entropy. The data storage area may store data collected or generated based on the use of the electronic device 200, such as remote sensing image data from multiple sources and auxiliary data, conditional entropy values, and analysis results of a geographic detector.

[0059] In addition, those skilled in the art will appreciate that the structure of the electronic device 200 shown in the above figures does not limit the electronic device 200. The electronic device may include more or fewer components than shown, or may combine certain components, or arrange the components differently. For example, the electronic device 200 may also include a microphone, a speaker, a radio frequency circuit, a sensor, an audio circuit, a power supply, a Bluetooth module, and other components, which will not be described in detail here.

[0060] This embodiment provides a remote sensing image change detection method based on information entropy. Figure 1 As shown, the method includes steps S101 to S105, which are specifically as follows:

[0061] Step S101: Acquire remote sensing image data and auxiliary data of a target area and perform preprocessing to obtain preprocessed remote sensing image data and auxiliary data.

[0062] In this embodiment, the target area can be an area of any geographical range, for example, it can be an area of a national range, a provincial range, a municipal range or a county range. In addition, the target area can also be specified by a latitude and longitude value range, or drawn by software. This embodiment does not limit the size and specific location of the target area.

[0063] In this embodiment, the remote sensing image data is time series data. Specifically, the remote sensing image data includes remote sensing images collected at a first moment and a second moment, respectively, and the first moment and the second moment are different observation times of the remote sensing images.

[0064] When acquiring remote sensing image data of a target area, remote sensing images (respectively denoted as RS1 and RS2) at a first moment (denoted as T1) and a second moment (denoted as T2) may be acquired according to a vector range of the target area.

[0065] Furthermore, the remote sensing images may be Landsat satellite images at time T1 and time T2. At the same time, image data provided by other remote sensing platforms may be obtained according to research needs, which is not limited in this embodiment.

[0066] In this embodiment, auxiliary data is data used to enhance, interpret or supplement remote sensing image information. Its acquisition time corresponds to the acquisition time of the remote sensing image data. Time synchronization can significantly enhance the correlation between the two, while effectively avoiding errors caused by time differences and improving the accuracy of detection results.

[0067] Furthermore, the auxiliary data includes nighttime light remote sensing data, road network data, and POI data. POI data can be obtained from Open Street Map (OSM) data. Data corresponding to times T1 and T2 can be filtered out, downloaded, and stored in local storage. Road network data is used to describe the structure and attributes of a road network. In this embodiment, road network data can be downloaded from the OSM website or obtained from other sources, such as government agencies, commercial data providers, and academic research, though this embodiment does not limit this.

[0068] In this embodiment, night light remote sensing data is a special type of remote sensing data, which is mainly used to monitor the intensity and distribution of human activities on the surface. For example, it can be data from specific satellite sensors (such as DMSP-OLS, NPP-VIIRS, and China Luojia-1).

[0069] Furthermore, the remote sensing image data and auxiliary data of the target area are preprocessed, including: correcting the remote sensing image data, and resampling and nearest neighbor assignment processing the auxiliary data in sequence to obtain preprocessed remote sensing image data and auxiliary data.

[0070] Specifically, the remote sensing images at time T1 and time T2 are preprocessed with radiation correction, geometric correction, and atmospheric correction to obtain preprocessed remote sensing images, which are denoted as RS_T1 and RS_T2 respectively.

[0071] For auxiliary data in vector format (such as POI data and road network data), the following rasterization processing is performed according to the resolution of night light remote sensing data (for example, 1km):

[0072] (a) POI data is rasterized, and the value of each grid is the number of POI points covered by the grid;

[0073] (b) The road network data is rasterized, and the value of each grid is the road network density within the grid range.

[0074] The auxiliary data are then resampled and nearest neighbor assigned to obtain preprocessed remote sensing image data and auxiliary data. Specifically, the processed nighttime light remote sensing data, POI grid (gridded POI), and road network density grid data (gridded road network) are resampled to obtain resampled auxiliary data. The resampled grids are assigned values using nearest neighbor assignment to keep them consistent with the processed Landsat data resolution at time T1 and T2. The processed auxiliary data are recorded as: RS_L (road network), RS_P (POI), and RS_R (nighttime light).

[0075] Among them, the specific operation of the nearest neighbor assignment processing is as follows: any blank pixel in the resampled auxiliary data is used as the target position, and the nearest known data point is found at the target position, and the value of the known data point is directly assigned to the pixel at the target position. The advantage of doing this is that it can avoid introducing new values, prevent data from deviating from the original value, preserve the authenticity of the data, and provide a reliable data basis for subsequent steps.

[0076] Step S102: Based on the preprocessed remote sensing image data and auxiliary data, multiple changed sub-regions of the target area are extracted respectively, and the first feature set and the second feature set corresponding to each changed sub-region are extracted. The first feature set is used to characterize the feature changes of natural objects, and the second feature set is used to characterize the feature changes of artificial objects.

[0077] It should be noted that the preprocessed remote sensing imagery and auxiliary data are raster data with consistent resolution. By comprehensively analyzing these data and employing change detection algorithms (such as interpolation, post-classification comparison, and deep learning models), we can extract multiple sub-regions of change within the target area. These sub-regions can represent changes in specific landform types, such as newly built urban areas, destroyed roads, and expanded farmland. By integrating these two types of data (remote sensing imagery and auxiliary data) within the same spatiotemporal framework, the accuracy of regional change detection can be improved.

[0078] Among them, the first feature set includes: the difference between the normalized difference vegetation index (NDVI), enhanced vegetation index (EVI), soil adjusted vegetation index (SAVI), normalized difference moisture index (NDMI), and photochemical vegetation index (PRI) between two moments. The above indices are used to characterize the characteristic changes of natural objects; the second feature set includes: the difference between the normalized building index (NDBI), impervious surface index (UI), and built-up area index (IBI) between two moments, as well as the differences of various auxiliary data between two moments.

[0079] By introducing these feature sets, the spatial change information of natural and artificial objects can be fully captured, significantly improving the accuracy, reliability and applicability of remote sensing image change detection. For example, the spectral characteristics of buildings and bare land are relatively similar, but there are generally POI points, road networks or night light remote sensing values near buildings, while bare land generally does not have these auxiliary elements. Therefore, combined with auxiliary data such as road networks, POI points or night light remote sensing data, objects with similar spectral characteristics can be effectively distinguished, and abnormal information caused by the "different objects with the same spectrum" and "same objects with different spectra" phenomena can be filtered to a certain extent, reducing change detection errors. From another perspective, the first feature set and the second feature set selected in this embodiment, especially the addition of auxiliary data such as road network data, night light remote sensing data and POI data, make the changes in each feature have a mutual confirmation effect to a certain extent, thereby improving the accuracy of change detection.

[0080] The calculation methods of the above-mentioned indexes can be implemented with reference to the existing technology, and will not be described in detail here.

[0081] Preferably, based on the pre-processed remote sensing image data and auxiliary data, multiple changed sub-regions of the target area are extracted respectively, including:

[0082] Step S112: for all pixels in the remote sensing image data and the auxiliary data, calculate the change difference between the first moment and the second moment for each feature in the first feature set and the second feature set to obtain a first change difference for each feature.

[0083] Specifically, based on the preprocessed remote sensing images RS_T1 and RS_T2, the normalized difference vegetation index (NDVI), enhanced vegetation index (EVI), soil adjusted vegetation index (SAVI), normalized difference moisture index (NDMI), photochemical vegetation index difference (PRI), normalized building index (NDBI), impervious surface index (UI), and built-up land index (IBI) at time T1 and time T2 were extracted respectively, and the difference of each index between the two times was calculated to obtain the change difference of each index.

[0084] Calculate the change difference of each auxiliary data (i.e. RS_L, RS_P, RS_R) between the two moments, that is, subtract the pixel value in each auxiliary data at time T2 from that at time T1 to obtain the calculation result, and then combine the calculation result with the change difference of the above-mentioned indexes to obtain the first change difference of each feature.

[0085] By calculating the change differences of the first feature set and the second feature set of all pixels in the remote sensing image data and auxiliary data, we can comprehensively capture the various feature information of the objects changing over time, enhance the sensitivity to complex changes, provide high-quality input data for subsequent principal component analysis and difference image construction, and ensure the accuracy and reliability of subsequent steps.

[0086] Step S122: Based on the first change difference, a principal component analysis (PCA) method is used to construct a difference image between the two moments, and a local minimum error probability method is used to determine a threshold, and then a change pattern is extracted from the difference image.

[0087] Specifically, the first change difference is subjected to principal component analysis (PCA) transformation to obtain multiple principal components, and then the top K principal components are selected for change vector analysis to obtain the difference image at two moments, ensuring that the difference image has sufficient information and less noise.

[0088] After obtaining the difference image, the local minimum error probability method is used to determine the threshold: First, multiple moving windows of size n×n are determined according to the size of the remote sensing image. For the i-th window, the initial threshold T0 is set. The value range of T0 is between the minimum and maximum values of the difference image pixels. The value p of each pixel in window i is compared with the size of T0. When p>T0, it is marked as changed, otherwise it is marked as changed; the mean, variance and the percentage Q and (1-Q) of the changed and unchanged pixels in the window are calculated, and the threshold is estimated according to the following formula:

[0089]

[0090] Where, T i To estimate the threshold, μ1 and μ2 are the means of the changed pixels and unchanged pixels in the window, respectively. σ1 and σ2 are the variances of the changed pixels and the unchanged pixels in the window, respectively.

[0091] When ΔT=T i When T0 is less than a certain preset minimum threshold, the search stops and the local threshold of the final window is (T i + T0 / 2), otherwise, T0=T0+1, and repeat the steps of comparing the value p of each pixel in the window i with the size of T0 and the steps thereafter.

[0092] The pixels inside each window are binary segmented using the corresponding local threshold to obtain the change spots.

[0093] This embodiment adopts the local minimum error probability method to determine the threshold, which can dynamically adjust the threshold according to the statistical characteristics of the difference image, thereby improving the accuracy of change detection in complex scenes.

[0094] Step S132: Perform morphological partitioning on the change patch according to connectivity to obtain multiple initial sub-regions.

[0095] In this embodiment, morphological partitioning is performed based on connectivity, which can aggregate discrete change patches into initial sub-regions with spatial connectivity, making the spatial distribution of the change regions clearer and facilitating subsequent analysis.

[0096] Step S142: extract the bounding rectangle of each initial sub-region as the final changed sub-region. This arrangement simplifies the expression of the changed region, can express complex changed regions in a concise form, and improves the efficiency of subsequent analysis.

[0097] By combining key technical links such as feature change difference calculation, principal component analysis (PCA), threshold determination, morphological partitioning and bounding rectangle extraction, efficient and accurate change detection is achieved.

[0098] After the final changed sub-region is determined, the first feature set and the second feature set corresponding to each changed sub-region are extracted. The specific calculation content is described in the above embodiment and will not be repeated here.

[0099] Step S103 : Based on the first feature set and the second feature set of each change sub-region, respectively calculate a first conditional entropy set and a second conditional entropy set of each change sub-region.

[0100] It should be noted that conditional entropy is an important concept in information theory, used to measure the uncertainty (i.e., the amount of information) of a random variable under certain known conditions. The first conditional entropy set is calculated based on the first feature set (natural feature characteristics) and reflects the uncertainty of natural feature changes. The second conditional entropy set is calculated based on the second feature set (artificial feature characteristics) and reflects the uncertainty of artificial feature changes.

[0101] Furthermore, based on the first feature set and the second feature set of each change sub-region, a first conditional entropy set and a second conditional entropy set of each change sub-region are calculated respectively, including:

[0102] Step S113: For each change sub-region, calculate the change difference between the first moment and the second moment for each feature in the first feature set and the second feature set to obtain a second change difference for each feature.

[0103] It should be noted that the second change difference is a feature change with the change sub-region as the carrier, whereas the first change difference is a feature change with each pixel as the carrier. This is the difference between the two.

[0104] Step S123: Calculate the conditional entropy of each feature based on the second change difference of each feature.

[0105] Step S133: sort the conditional entropies of the features, extract the conditional entropies ranked in the top N positions in the first feature set and the second feature set, and combine them into the first conditional entropy set and the second conditional entropy set corresponding to the first feature set and the second feature, respectively.

[0106] For example, the conditional entropy can be calculated for the change difference of each feature in each change sub-region; then the N features with the smallest conditional entropy in each type of feature (artificial, natural) are screened out, for example, the top 3 features with the smallest conditional entropy in each type of feature (artificial, natural) are screened out to obtain the first conditional entropy set and the second conditional entropy set corresponding to artificial and natural (i.e., the first feature set and the second feature set).

[0107] By calculating conditional entropy, we can assess the distribution of information within each feature's changing subregion. Using conditional entropy sorting and filtering, we can filter out time mismatches. For example, when there are significant temporal differences, conditional entropy can filter out seasonal variations, further improving the accuracy of change detection. Furthermore, the conditional entropy of each feature is ranked from smallest to largest. The magnitude of the conditional entropy reflects the complexity and importance of feature change. Higher conditional entropy values indicate greater uncertainty in the feature information within that subregion, while lower values indicate more concentrated feature information. This helps us understand the drivers of feature change from a qualitative and global perspective. Combining this with geographic detectors for attribution analysis can enhance the interpretability of the causes of change.

[0108] Step S104: Taking the spectral angle of each pixel in each change sub-region as the dependent variable and each feature in the first conditional entropy set and the second conditional entropy set as the explanatory factor, perform geographic detector analysis on the dependent variable and the explanatory factor, and calculate the q value corresponding to each feature.

[0109] The purpose of step S104 is to perform geographic detector analysis. Since the input independent variable (explanatory factor) of the geographic detector is categorical data, and the value of the second change difference is a continuous variable, the continuous variable needs to be graded. In some embodiments, the method also includes: classifying the second change difference of each feature to obtain a change difference category.

[0110] Specifically, the second variation differences of each feature may be classified using methods such as natural discontinuity method, equal-interval classification, and equal-frequency classification. This embodiment does not limit the specific classification method.

[0111] The natural breaks method is preferably based on the principle of minimum variance, dividing data into groups with minimal internal variance and maximum inter-class variance, thereby maximizing the discrimination between categories. Therefore, using the natural breaks method for classification enables the q-value calculated by the geodetector to more clearly reflect the explanatory power of the variable on the dependent variable, avoiding information loss caused by the classification method. The natural breaks method is also commonly used in GIS spatial analysis, and its classification results can be directly used for spatial visualization, making spatial distribution characteristics clearer.

[0112] On the basis of continuous variable classification, the spectral angle of each pixel in each change sub-region is used as the dependent variable, and the corresponding features in the first conditional entropy set and the second conditional entropy set are used as explanatory factors to perform geographic detector analysis, and the q value corresponding to each feature is calculated. Specifically, the spectral angle of each pixel in each change sub-region is used as the dependent variable, and the change difference category of each feature corresponding to the first conditional entropy set and the second conditional entropy set is input into the geographic detector for factor detection to obtain the q value corresponding to each feature.

[0113] In this embodiment, the dependent variable of the geographic detector analysis is the spectral angle of each pixel in each change sub-region. The spectral angle mapper (SAM) is used to measure the similarity between the spectral characteristics of the same pixel in two temporal images. The specific calculation method is as follows: first, the values of each pixel in the remote sensing image (or auxiliary data) at the first moment and the second moment in each band are expressed as an m-dimensional spectral vector, where m is the number of bands. Then, the angle between the two spectral vectors is calculated to obtain the spectral angle θ. The above steps are repeated for all pixels in the change sub-region to calculate the spectral angle θ of each pixel in the two temporal images, thereby generating a spectral angle image, in which the value of each pixel represents the spectral angle of the pixel in the two temporal images.

[0114] Then, taking the spectral angle of each pixel in each change sub-area as the dependent variable (i.e., as the Y value of the geographic detector), the change difference category of each corresponding feature in the first conditional entropy set and the second conditional entropy set is input into the geographic detector for factor detection, that is, the change difference category of all features in the first conditional entropy set (such as the top 3 with the smallest conditional entropy in the first feature set) and the second conditional entropy set (such as the top 3 with the smallest conditional entropy in the second feature set) (i.e., the data classified by the natural break method) is used as the explanatory factor (independent variable X), and the geographic detector analysis is performed to obtain the q value corresponding to each feature.

[0115] Among them, in the factor detection of the geographic detector, the q value calculation formula is as follows:

[0116]

[0117] Where L is the total number of change difference categories of a certain feature, h is the sequence number of the change difference category, N h , σ h ′ are the total number of samples and variance of category h respectively.

[0118] It should be noted that the range of q value is [0,1], and the larger the q value is, the stronger the explanatory power of the factor is.

[0119] Step S105 sorts and scores the q values, and outputs the change detection and driving force analysis results.

[0120] Since the q value reflects the size of the factor's explanatory power, the main driving factors can be analyzed by sorting the q values and performing scoring processing, which can further simplify the comparison of the explanatory power of each feature and simultaneously output the change detection and driving force analysis results.

[0121] In some embodiments, q values are sorted and scored, and change detection and driving force analysis results are output, including:

[0122] Step S115: Sort the q values and use the sorted serial numbers as corresponding feature values.

[0123] First, sort and obtain the sorting sequence number of each feature from small to large. Then, check the sorting sequence number of each feature. For example, if the q-value sorting sequence number of the feature NDVI is 1, then assign 1 to the feature NDVI and save it as one of its attributes. Similarly, if the q-value sorting sequence number of the POI change difference is 2, then assign 2 to the feature POI change difference and save it as its attribute.

[0124] By sorting the q values and using the sorting numbers as feature values, the originally complex q values can be converted into numerical values that are easy to understand and compare, making the quantitative results more intuitive and providing a clear order for subsequent analysis.

[0125] Step S125 : Summarize the feature values in the first feature set and the second feature set to obtain the overall rankings of the first feature set and the second feature set.

[0126] It should be noted that, since the factors in factor detection are explained by the features in the first conditional entropy set and the second conditional entropy set, only the features in the first conditional entropy set and the second conditional entropy set have q values. Correspondingly, after scoring, only the features in the first conditional entropy set and the second conditional entropy set have scores, and other features are not scored and will not affect the overall ranking. That is to say, the total ranking only reflects the summary of feature rankings after conditional entropy screening. For example, the top three conditional entropy features included in the first conditional entropy set are NDVI, SAVI, and PRI, and their q-value sorting numbers are 1, 5, and 6, respectively. The top three conditional entropy features included in the second conditional entropy set are IBI, POI change difference, and road network change difference, and their q-value sorting numbers are 2, 3, and 4, respectively. The total ranking of the first feature set is 1+5+6=12, and the total ranking of the second feature set is 2+3+4=9. Therefore, the total ranking of the second feature set is 9, which is lower than the total ranking of the first feature set, which is 12. The changes in this sub-region are more affected by human activities, that is, the driving force of the change in this sub-region is human activities.

[0127] In this embodiment, the aggregation method of ranking numbers can fully reflect the overall contribution of the two feature sets in change detection, which is conducive to quickly determining which feature set has a greater impact on change detection.

[0128] Step S135: Based on the overall ranking of the first and second feature sets for each change sub-region, each change sub-region is evaluated individually to determine the driving force of each change sub-region. This evaluation of each change sub-region based on the overall ranking can accurately identify the primary driving force of each change sub-region, enhancing the interpretability of the change detection results.

[0129] Furthermore, in step S135, based on the total rankings of the first feature set and the second feature set of each changed sub-region, all changed sub-regions are judged one by one to determine the driving force of each changed sub-region, specifically: if the total ranking of the first feature set is greater than the total ranking of the second feature set, the change in the sub-region is considered to be artificially influenced; if the total ranking of the first feature set is less than the total ranking of the second feature set, the change in the sub-region is considered to be naturally influenced; if the total ranking of the first feature set is equal to the total ranking of the second feature set, the change in the sub-region is considered to be jointly influenced.

[0130] That is to say, sorting and scoring the obtained q values can be performed as follows:

[0131] a. If the sum of the q-value rankings of the artificial features is smaller, the change in the sub-region is considered to be artificially affected;

[0132] b. If the sum of the q-value rankings of the artificial features is greater, the change in the sub-region is considered to be a natural influence;

[0133] c. If the sum of the q-value rankings of the artificial features is the same as the q-value rankings of the natural features, the changes in the sub-region are considered to be jointly affected.

[0134] This embodiment classifies and determines the driving forces of the change sub-regions by comparing the overall rankings of the first feature set and the second feature set, which can significantly improve the scientificity, interpretability, and practicality of the change detection results. Specifically, the first feature set (natural feature) and the second feature set (artificial feature) represent two different types of change drivers. By setting clear rules (the overall ranking relationship), the driving forces of the change sub-regions are divided into three categories: "natural influence", "artificial influence", and "common influence". This classification method is simple and intuitive, easy to understand and operate. In addition, the driving force classification results can deeply reveal the reasons behind the changes, enhancing the scientificity and interpretability of the change detection results.

[0135] Furthermore, the change detection and driving force analysis results are output, including: mapping, generating a change area map, and then visually interpreting the type of change (i.e., from which land use type to which land use type), to form the final change area map and change driving force attribution analysis report.

[0136] Refer to the following Figure 2 The method provided in this embodiment is further described.

[0137] like Figure 2 As shown, the method provided in this embodiment first obtains remote sensing images at time T1 and time T2, and corresponding auxiliary data, and then inputs the remote sensing images and auxiliary data into the preprocessing module for preprocessing to obtain preprocessed remote sensing image data and auxiliary data; then uses the feature extraction module to extract features from the preprocessed remote sensing image data and auxiliary data to obtain a first feature set representing the feature changes of natural objects and a second feature set representing the feature changes of artificial objects; then uses the conditional entropy module to calculate the conditional entropy of each feature change, and on this basis, uses the geographic detector module to carry out geographic detector analysis, and finally uses the change detection and attribution module to perform change detection and attribution, and outputs the change detection and driving force analysis results.

[0138] In summary, in view of the current state of technology in which decision makers in the process of urban renewal are more concerned about the driving forces: that is, in the process of urban renewal, relevant government departments are not only concerned with the precise changes in surface elements, but are more concerned with the reasons behind the changes in surface types, such as expansion caused by urban construction or illegal construction, fallow land or abandonment of cultivated land, etc. The method provided in this embodiment combines geographic detectors and conditional entropy to accurately detect changes in land use types, and at the same time gives the driving forces of the changes, providing a scientific reference for urban planning and related departments.

[0139] The method provided in this embodiment utilizes a combination of conditional entropy and geographic detectors to simultaneously perform change attribution while detecting changes. Conditional entropy performs attribution analysis from a qualitative, global perspective, while geographic detectors perform attribution analysis from a quantitative, local perspective. Combining the two takes into account both global and local influencing factors and reduces information redundancy.

[0140] Based on the same inventive concept, this embodiment provides a remote sensing image change detection system based on information entropy, the system comprising:

[0141] a data acquisition and preprocessing unit configured to acquire remote sensing image data and auxiliary data of a target area and perform preprocessing to obtain preprocessed remote sensing image data and auxiliary data; the remote sensing image data includes remote sensing images covering the target area at a first moment and a second moment, the first moment and the second moment being different observation times of the remote sensing image; and the auxiliary data is collected at a time corresponding to the remote sensing image data;

[0142] an extraction unit configured to extract, based on the preprocessed remote sensing image data and the auxiliary data, a plurality of changed sub-regions of the target area, and extract a first feature set and a second feature set corresponding to each changed sub-region, wherein the first feature set is used to characterize the characteristics of natural objects, and the second feature set is used to characterize the characteristics of artificial objects;

[0143] A calculation unit configured to calculate a first conditional entropy set and a second conditional entropy set corresponding to each change sub-region based on the first feature set and the second feature set;

[0144] a geographic detection unit configured to perform geographic detector analysis using the spectral angle of each pixel in each change sub-region as a dependent variable and corresponding features in the first conditional entropy set and the second conditional entropy set as explanatory factors, and calculate a q value corresponding to each feature;

[0145] The output unit is configured to sort and score the q values and output the change detection and driving force analysis results.

[0146] The present embodiment provides a remote sensing image change detection system based on information entropy, which can implement the steps and processes of the remote sensing image change detection method based on information entropy provided in any of the above embodiments and achieve the same technical effects, and will not be described in detail here.

[0147] The foregoing description is merely a preferred embodiment of the present application and is not intended to limit the present application. Various modifications and variations are readily apparent to those skilled in the art. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present application shall be included within the scope of protection of the present application.

Claims

1. A remote sensing image change detection method based on information entropy, characterized in that: include: remote sensing image data and auxiliary data of a target area are acquired and preprocessed to obtain preprocessed remote sensing image data and auxiliary data; the remote sensing image data includes remote sensing images acquired at a first moment and a second moment, respectively, the first moment and the second moment being different observation times of the remote sensing image; and the acquisition time of the auxiliary data corresponds to the acquisition time of the remote sensing image data; Based on the preprocessed remote sensing image data and auxiliary data, a plurality of changed sub-regions of the target area are extracted respectively, and a first feature set and a second feature set corresponding to each changed sub-region are extracted, wherein the first feature set is used to characterize the feature changes of natural objects, and the second feature set is used to characterize the feature changes of artificial objects; Based on the first feature set and the second feature set of each change sub-region, respectively calculating a first conditional entropy set and a second conditional entropy set of each change sub-region; Taking the spectral angle of each pixel in each change sub-region as the dependent variable and each feature in the first conditional entropy set and the second conditional entropy set as the explanatory factor, performing geographic detector analysis on the dependent variable and the explanatory factor, and calculating the q value corresponding to each feature; Sort and score the q values, and output the change detection and driving force analysis results.

2. The method according to claim 1, characterized in that The auxiliary data includes: night light remote sensing data, road network data and POI data.

3. The method according to claim 1, characterized in that The first feature set includes: the difference between the normalized difference vegetation index, enhanced vegetation index, soil adjusted vegetation index, normalized difference moisture index, and photochemical vegetation index between two moments; The second feature set includes: differences in the normalized building index, the impervious surface index, and the building land index between the two moments, and differences in various auxiliary data between the two moments.

4. The method according to claim 1, wherein Preprocess the remote sensing image data and auxiliary data of the target area, including: Correction processing is performed on the remote sensing image data, and resampling and nearest neighbor assignment processing are performed on the auxiliary data in sequence to obtain pre-processed remote sensing image data and auxiliary data.

5. The method according to claim 1, wherein Based on the pre-processed remote sensing image data and auxiliary data, multiple changed sub-regions of the target region are extracted respectively, including: For all pixels in the remote sensing image data and the auxiliary data, calculating the change difference between the first moment and the second moment for each feature in the first feature set and the second feature set to obtain a first change difference for each feature; Based on the first change difference, a principal component analysis method is used to construct a difference image between two moments, and a local minimum error probability method is used to determine a threshold, thereby extracting a change pattern from the difference image; Performing morphological partitioning on the change pattern according to connectivity to obtain a plurality of initial sub-regions; The bounding rectangle of each initial sub-region is extracted as the final changed sub-region.

6. The method according to claim 1, characterized in that Based on the first feature set and the second feature set of each change sub-region, respectively calculating a first conditional entropy set and a second conditional entropy set of each change sub-region, including: For each change sub-region, calculating the change difference between the first moment and the second moment for each feature in the first feature set and the second feature set to obtain a second change difference for each feature; Calculating the conditional entropy of each feature based on the second change difference of each feature; The conditional entropies of the features are sorted, and the top N bits with the smallest conditional entropies in the first feature set and the second feature set are extracted respectively, and combined into a first conditional entropy set and a second conditional entropy set corresponding to the first feature set and the second feature, respectively.

7. The method according to claim 6, characterized in that After respectively calculating a first conditional entropy set and a second conditional entropy set for each change sub-region based on the first feature set and the second feature set for each change sub-region, the method further includes: classifying the second change differences of the respective features to obtain change difference categories; Correspondingly, the spectral angle of each pixel in each change sub-region is used as the dependent variable, and the corresponding features in the first conditional entropy set and the second conditional entropy set are used as explanatory factors to perform geographic detector analysis, and calculate the q value corresponding to each feature, specifically: Taking the spectral angle of each pixel in each change sub-area as the dependent variable, the change difference category of each feature corresponding to the first conditional entropy set and the second conditional entropy set is input into the geographic detector for factor detection to obtain the q value corresponding to each feature.

8. The method according to claim 1, characterized in that Sort and score the q values, and output the change detection and driving force analysis results, including: Sort the q values and use the sorted numbers as the corresponding feature values; Summarizing the feature values in the first feature set and the second feature set to obtain the overall ranking of the first feature set and the second feature set; Based on the total rankings of the first feature set and the second feature set of each change sub-region, all change sub-regions are judged one by one to determine the driving force of each change sub-region.

9. The method according to claim 8, characterized in that Based on the total rankings of the first feature set and the second feature set of each change sub-region, all change sub-regions are judged one by one to determine the driving force of each change sub-region, including: If the total ranking of the first feature set is greater than the total ranking of the second feature set, the change in the sub-region is considered to be artificial influence; If the total ranking of the first feature set is less than the total ranking of the second feature set, the change in the sub-region is considered to be a natural influence; If the total ranking of the first feature set is equal to the total ranking of the second feature set, the change in the sub-region is considered to be a common influence.

10. A remote sensing image change detection system based on information entropy, characterized in that: include: a data acquisition and preprocessing unit configured to acquire remote sensing image data and auxiliary data of a target area and perform preprocessing to obtain preprocessed remote sensing image data and auxiliary data; the remote sensing image data includes remote sensing images covering the target area at a first moment and a second moment, the first moment and the second moment being different observation times of the remote sensing image; and the auxiliary data is collected at a time corresponding to the remote sensing image data; an extraction unit configured to extract, based on the preprocessed remote sensing image data and the auxiliary data, a plurality of changed sub-regions of the target area, and extract a first feature set and a second feature set corresponding to each changed sub-region, wherein the first feature set is used to characterize the characteristics of natural objects, and the second feature set is used to characterize the characteristics of artificial objects; A calculation unit configured to calculate a first conditional entropy set and a second conditional entropy set corresponding to each change sub-region based on the first feature set and the second feature set; a geographic detection unit configured to perform geographic detector analysis using the spectral angle of each pixel in each change sub-region as a dependent variable and corresponding features in the first conditional entropy set and the second conditional entropy set as explanatory factors, and calculate a q value corresponding to each feature; The output unit is configured to sort and score the q values and output the change detection and driving force analysis results.

Images