Change detection method and system based on multi-level feature fusion of subdivided grid images

Abstract

The present invention provides a change detection method and system based on multi-level feature fusion of subdivided grid images, comprising the following steps: S1, acquiring images of different phases of a to-be-detected area; S2, training a feature extraction model and a grid model for subdividing images according to different image sizes; S3, gridding the current image to generate a plurality of grid units with unique grid codes; S4, generating a 0 change matrix to extract a number of image features of different phases; the present invention can effectively fuse multiple levels and multiple models based on the change matrix, automatically adapt the model selection according to the slice size of different images, select a CNN model with a larger effective receptive field for large-size images, select a spectral difference analysis or a histogram difference algorithm for small-size images, map the extracted eigenvalues ​​to the positions of the change matrix through the grid code, and superimpose the eigenvalues ​​on the regional blocks of the change matrix to achieve effective fusion of multiple levels and multiple models.

Description

Technical Field

[0001] The present invention relates to the technical field of remote sensing change detection, and in particular to a change detection method and system based on multi-level feature fusion of subdivided grid images. Background Art

[0002] Remote sensing change detection technology is a technology that determines whether the objects in the same area of ​​multi-temporal remote sensing images have changed, and even detects how they have changed. Through manual and computer-assisted processing of multi-temporal remote sensing images covering the same area, the change information in the images can be accurately and quickly extracted to achieve dynamic monitoring of changing objects and analysis of surface change trends and their evolution laws. It plays an extremely important role in many fields such as urban expansion, land use change, forest vegetation change, ecological environment monitoring and disaster monitoring.

[0003] Among the existing image remote sensing change detection technologies, there are two main solutions:

[0004] One is to obtain the change information by comparing the spectral difference between the two images. This method is suitable for images with good geometric registration and small size.

[0005] The other is to use CNN convolutional neural network to extract image features. This method has a large effective receptive field and is suitable for images with larger sizes.

[0006] The above two solutions also have the following disadvantages:

[0007] First, taking the first image remote sensing change detection technology as an example, its accuracy in the process of image remote sensing change detection is based on the condition that the image geometric registration is intact. The change detection effect in a small range is good, but the scope of application is low. For images with larger sizes, it cannot be effectively detected;

[0008] 2. Taking the second image remote sensing change detection technology as an example, although the effective receptive field is large when extracting image features, many targets in remote sensing images are small targets with only dozens or even a few pixels. For change detection of small targets, the pooling layer of CNN will further reduce the amount of information, making the dimension too low to distinguish.

[0009] In summary, both methods have their own advantages and disadvantages. The current image remote sensing change detection technology has the problem of low change detection and recognition accuracy of a single model at different scales.

[0010] Therefore, a change detection method and system based on multi-level feature fusion of subdivided grid images was proposed. Summary of the invention

[0011] In view of this, the embodiments of the present invention hope to provide a change detection method and system based on multi-level feature fusion of subdivided grid images to solve or alleviate the technical problems existing in the prior art and at least provide a beneficial option.

[0012] The technical solution of the embodiment of the present invention is implemented as follows: a change detection method based on multi-level feature fusion of a subdivided grid image comprises the following steps:

[0013] S1, obtaining images of the area to be detected at different phases;

[0014] S2, training a feature extraction model according to different image sizes and a grid model for segmenting images;

[0015] S3, gridding the current image to generate multiple grid cells with unique grid codes;

[0016] S4, generating a zero-change matrix and extracting several image features at different phases;

[0017] S5, importing the image into the feature extraction model to determine the grid model required for the image of the current area to be detected;

[0018] S6. Map the extracted eigenvalues ​​to the positions of the change matrix through the grid code, and superimpose the eigenvalues ​​to the regional blocks of the change matrix to achieve effective fusion of multi-level and multi-models.

[0019] Further preferably: the feature extraction model includes a CNN feature extraction model and a change difference extraction model;

[0020] When extracting the feature extraction model, the size of each level of mesh is customized;

[0021] The CNN feature extraction model is used to extract features from large-size images, such as images with a level less than or equal to 20 encoded using GeoSOT gridding;

[0022] The variation difference extraction model is used to extract variation differences from small-size images, such as images encoded using GeoSOT grid subdivision. When the variation difference extraction model extracts images, it can use a variety of algorithm fusions for extraction, such as using any of spectral differences or histograms to extract variation differences.

[0023] Further preferably, when the image of the area to be detected is segmented by a grid model, the following steps are included:

[0024] S31, gridding the current image through the coding unit to generate a plurality of grid units with unique grid codes;

[0025] S32, initializing the task stack: obtaining a grid set corresponding to the facet of the minimum bounding rectangle of the image;

[0026] S33, calculating the size of the change matrix and generating a 0 change matrix;

[0027] S34. Specify the end level.

[0028] Further preferably, when the image features are extracted, the following steps are included:

[0029] S41, extracting grid codes one by one from the task stack, and obtaining the area image corresponding to the grid code;

[0030] S42, calculating a number of image features at two different time phases;

[0031] S43, multi-model hybrid calculation of image feature values.

[0032] Further preferably, after the feature extraction is completed, a change matrix updating step is further included, wherein the change matrix updating includes the following steps:

[0033] S51, after the mixed calculation of the image eigenvalues ​​is completed, the grid code is mapped to the area in the change matrix, and the eigenvalues ​​are weighted and summed;

[0034] S52, judging whether the end level has been reached:

[0035] If yes: go to step S53;

[0036] If not: go to step S54;

[0037] S53, the sub-level grid code is put into the task stack, the current grid code is taken out of the task stack, and the process returns to step S41;

[0038] S54, the current grid code outputs the task stack;

[0039] S55. Determine whether the task stack is empty:

[0040] If not: go to step S41;

[0041] If yes: Go to the next step.

[0042] Further preferably: after the update of the change matrix is ​​completed, threshold segmentation is also included, and the threshold segmentation includes segmenting the image according to the change matrix.

[0043] Further preferably, when the feature extraction model performs feature extraction, the following formula is used:

[0044]

[0045]

[0046] f 3 (g) = P(g)

[0047] Where: g is the grid code, g 1 , g 2 are two images of the grid code at different times, f 1 There are at least two or more CNN convolutional neural network models to judge g 1 , g 2 The confidence of the category, f 2 It is to combine at least two traditional image processing algorithms to calculate g 1 , g 2 , w is the feature coefficient weight, similarity is the similarity feature value, and confidence is the confidence of the neural network in judging the grid type; f 3 is the probability that the grid code obtained from other data sources may change;

[0048] When the image is of large size, for example, using the GeoSOT grid encoding level, when the current grid level is between 0 and 20, the convolutional neural network f is used. 1 To perform feature extraction;

[0049] When the image is of small size, for example, when the level of the GeoSOT grid is coded, and the current level of the grid is between 20 and 31, the traditional visual algorithm f 2 To perform feature extraction;

[0050] Prior knowledge can be used at any level 3 Assist in correcting the recognition results.

[0051] The change detection system based on multi-level feature fusion of subdivided grid images includes:

[0052] An acquisition module, which is used to acquire images of the area to be detected at different phases;

[0053] An encoding module, the encoding module is used to perform grid division on the current image and generate a plurality of grid units with unique grid codes;

[0054] A recognition module, the recognition module is used to recognize the size information of the current image, and select different feature extraction models to perform feature extraction according to the image size information;

[0055] An extraction module, the extraction module is used to extract a number of image features at different time phases;

[0056] The detection module is used to map the extracted eigenvalues ​​to the positions of the change matrix through the grid code, and to superimpose the eigenvalues ​​into the regional blocks of the change matrix to achieve effective fusion of multi-level and multi-models.

[0057] A computer device comprises a processor and a memory coupled to the processor, wherein the memory stores program instructions, and when the program instructions are executed by the processor, the processor executes the steps of the change detection method based on multi-level feature fusion of subdivided grid images as described above.

[0058] A storage medium stores program instructions capable of implementing the change detection method based on multi-level feature fusion of subdivided grid images as described above.

[0059] The embodiment of the present invention has the following advantages due to the adoption of the above technical solution:

[0060] The present invention can effectively fuse multiple layers and multiple models based on the change matrix, and automatically selects models according to the slice sizes of different images. A CNN model with a larger effective receptive field is selected for large-size images, and a spectral difference analysis or histogram difference algorithm is selected for small-size images. The extracted eigenvalues ​​are mapped to the positions of the change matrix through grid codes, and the eigenvalues ​​are superimposed on the regional blocks of the change matrix to achieve effective fusion of multiple layers and multiple models.

[0061] The present invention can correct the change detection result through prior knowledge, and some prior knowledge can be obtained by obtaining an external data source through a unique grid code index, thereby achieving the purpose of assisting in correcting the change result.

[0062] The above summary is for illustrative purposes only and is not intended to be limiting in any way. In addition to the illustrative aspects, embodiments and features described above, further aspects, embodiments and features of the present invention will be readily apparent by reference to the accompanying drawings and the following detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or technical descriptions will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0064] Figure 1 is a flow chart of the method of the present invention;

[0065] Figure 2 It is the overall framework diagram of the grid change detection of the present invention;

[0066] Figure 3 It is a phase 1 image diagram of images of different phases in the same longitude and latitude range according to the second embodiment of the present invention;

[0067] Figure 4 The phase 2 image is an image of a different phase in the same longitude and latitude range according to the second embodiment of the present invention;

[0068] Figure 5 This is a schematic diagram of geographic information of the smallest outer bounding rectangular patch of an image according to the second embodiment of the present invention;

[0069] Figure 6 Embodiment 2 of the present invention Figure 5 Schematic diagram of image affine transformation matrix update;

[0070] Figure 7 Schematic diagram of capturing two-phase images using four-corner coordinates in the second embodiment of the present invention Figure 1 ;

[0071] Figure 8 Schematic diagram of capturing two-phase images using four-corner coordinates in the second embodiment of the present invention Figure 2 ;

[0072] Fig. 9 A schematic diagram of adding eigenvalues ​​to a specified position of a change matrix according to a second embodiment of the present invention;

[0073] Fig.10 This is a schematic diagram of an area where a 17-level grid code of the second sub-level is located according to an embodiment of the present invention;

[0074] Fig.11 This is a visual diagram of the change matrix of the second embodiment of the present invention. DETAILED DESCRIPTION

[0075] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0076] The term "plurality" in this application means at least two, such as two, three, etc., unless otherwise specifically defined. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not limited to the listed steps or units, but optionally includes steps or units that are not listed, or optionally includes other steps or units inherent to these processes, methods, products or devices.

[0077] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0078] Embodiment 1

[0079] like Figure 1-2 As shown, the embodiment of the present invention provides a change detection method based on multi-level feature fusion of subdivided grid images, comprising the following steps:

[0080] S1, obtaining images of the area to be detected at different phases;

[0081] S2, training a feature extraction model according to different image sizes and a grid model for segmenting images;

[0082] The feature extraction model includes a CNN feature extraction model and a variation difference extraction model;

[0083] When extracting the feature extraction model, the size of each level of mesh is customized;

[0084] The CNN feature extraction model is used to extract features from large-size images, such as images with a level less than or equal to 20 encoded using GeoSOT gridding;

[0085] The variation difference extraction model is used to extract variation differences from small-size images, such as images encoded using GeoSOT grid subdivision. When the variation difference extraction model extracts images, it can use a variety of algorithms to fuse and extract, such as using any of spectral differences or histograms to extract variation differences;

[0086] When the feature extraction model performs feature extraction, the following formula is used:

[0087]

[0088]

[0089] f 3 (g) = P(g)

[0090] Where: g is the grid code, g 1 , g 2 are two images of the grid code at different times, f 1 There are at least two or more CNN convolutional neural network models to judge g 1 , g2 The confidence of the category, f 2 It is to combine at least two traditional image processing algorithms to calculate g 1 , g 2 , w is the feature coefficient weight, similarity is the similarity feature value, and confidence is the confidence of the neural network in judging the grid type; f 3 is the probability that the grid code obtained from other data sources may change;

[0091] When the image is of large size, for example, using the GeoSOT grid encoding level, when the current grid level is between 0 and 20, the convolutional neural network f is used. 1 To perform feature extraction;

[0092] When the image is of small size, for example, when the level of the GeoSOT grid is coded, and the current level of the grid is between 20 and 31, the traditional visual algorithm f 2 To perform feature extraction;

[0093] Prior knowledge can be used at any level 3 Assist in correcting the recognition results;

[0094] S3, gridding the current image to generate multiple grid cells with unique grid codes;

[0095] When the image of the area to be detected is segmented through the grid model, the following steps are included:

[0096] S31, gridding the current image through the coding unit to generate a plurality of grid units with unique grid codes;

[0097] S32, initializing the task stack: obtaining a grid set corresponding to the facet of the minimum bounding rectangle of the image;

[0098] S33, calculating the size of the change matrix and generating a 0 change matrix;

[0099] S34, specifying the end level;

[0100] S4, generating a zero-change matrix and extracting several image features at different phases;

[0101] When extracting features from image features, the following steps are included:

[0102] S41, extracting grid codes one by one from the task stack, and obtaining the area image corresponding to the grid code;

[0103] S42, calculating a number of image features at two different time phases;

[0104] S43, multi-model hybrid calculation of image feature values;

[0105] S5, importing the image into the feature extraction model to determine the grid model required for the image of the current area to be detected;

[0106] After the feature extraction is completed, the step of updating the change matrix is ​​also included, wherein the change matrix updating includes the following steps:

[0107] S51, after the mixed calculation of the image eigenvalues ​​is completed, the grid code is mapped to the area in the change matrix, and the eigenvalues ​​are weighted and summed;

[0108] S52, judging whether the end level has been reached:

[0109] If yes: go to step S53;

[0110] If not: go to step S54;

[0111] S53, the sub-level grid code is put into the task stack, the current grid code is taken out of the task stack, and the process returns to step S41;

[0112] S54, the current grid code outputs the task stack;

[0113] S55. Determine whether the task stack is empty:

[0114] If not: go to step S41;

[0115] If yes: go to the next step;

[0116] After the change matrix is ​​updated, it also includes threshold segmentation, which includes segmenting the image according to the change matrix;

[0117] S6. Map the extracted eigenvalues ​​to the positions of the change matrix through the grid code, and superimpose the eigenvalues ​​to the regional blocks of the change matrix to achieve effective fusion of multi-level and multi-models.

[0118] A change detection system based on multi-level feature fusion of subdivided grid images, comprising:

[0119] An acquisition module, which is used to acquire images of the area to be detected at different phases;

[0120] An encoding module, the encoding module is used to perform grid division on the current image and generate a plurality of grid units with unique grid codes;

[0121] A recognition module, the recognition module is used to recognize the size information of the current image, and select different feature extraction models to perform feature extraction according to the image size information;

[0122] An extraction module, the extraction module is used to extract a number of image features at different time phases;

[0123] The detection module is used to map the extracted eigenvalues ​​to the positions of the change matrix through the grid code, and to superimpose the eigenvalues ​​into the regional blocks of the change matrix to achieve effective fusion of multi-level and multi-models.

[0124] A computer device comprises a processor and a memory coupled to the processor, wherein the memory stores program instructions, and when the program instructions are executed by the processor, the processor executes the steps of the change detection method based on multi-level feature fusion of subdivided grid images as described above.

[0125] The processor is called CPU. A processor is an integrated circuit chip that has signal processing capabilities.

[0126] The processor may also be a general purpose processor, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic device, a discrete gate or transistor logic device, a discrete hardware component. A general purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0127] A storage medium storing program instructions capable of implementing the change detection method based on multi-level feature fusion of a subdivided grid image as described above. The storage medium of an embodiment of the present application stores program instructions capable of implementing all of the above methods, wherein the program instructions can be stored in the above storage medium in the form of a software product, including a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) or a processor to execute all or part of the steps of the methods described in each embodiment of the present application.

[0128] The aforementioned storage media include: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program code, or computer devices such as computers, servers, mobile phones, tablets, etc. The server can be an independent server, or it can be a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDN), and big data and artificial intelligence platforms.

[0129] Embodiment 2

[0130] like Figure 1-11As shown, the present invention also provides an embodiment of performing image remote sensing detection according to the method of embodiment 1, comprising the following steps:

[0131] Step 1: Parameter preparation

[0132] like Figure 3-4 As shown, two images with the same longitude and latitude range but different phases are prepared.

[0133] Initialize the task stack: The task stack is a linear list that only allows insertion or deletion at one end, and it has a last-in-first-out structural characteristic. First, obtain the grid set of the facets corresponding to the minimum bounding rectangle of the two-phase image, that is, find a grid code set of a level that just divides the image into four parts, and use the grid set to initialize the insertion task stack. In the example, the grid code of the facet corresponding to the minimum bounding rectangle of the given image is ['417394478127513600-16', '417394486717448192-16', '417394491012415488-16', '417394499602350080-16'], and its corresponding geographic location information is as follows Figure 5 The level of the grid code corresponding to the face of the smallest bounding rectangle of the image is taken as the starting level. start , the user specifies the level as the end level level lend , where level start <level end .

[0134] Generate a change matrix: The change matrix is ​​used to describe the changes of two images at different times. The size of the generated change matrix is ​​the same as the coverage of the grid code of the facet corresponding to the smallest enclosing rectangle, and the initial state of the change matrix is ​​a matrix of all 0s. To calculate the size of the change matrix, the following steps are required:

[0135] Calculate the affine transformation matrix of the change matrix. First, calculate the latitude and longitude of the upper left corner of the image's smallest enclosing rectangular patch grid set, such as Figure 6 , update the image affine transformation matrix GeoTransform(113.04222222222222,9.9999999999993514e-06,0.0,23.6833333333333334,0.0,-9.9999999999992915e-06). The meaning of each bit parameter of the affine transformation matrix is ​​as follows:

[0136] GeoTransform[0]: the longitude coordinate of the upper left corner of the image;

[0137] GeoTransform[1]: image resolution in the east-west direction;

[0138] GeoTransform[2]: rotation angle. If the north of the image is facing upward, the value is 0.

[0139] GeoTransform[3]: the latitude coordinate of the upper left corner of the image;

[0140] GeoTransform[4]: rotation angle. If the north of the image is facing upward, the value is 0.

[0141] GeoTransform[5]: image north-south resolution;

[0142] Calculate the size of the change matrix. Based on the affine transformation matrix and the longitude and latitude of the lower right corner, the formula for calculating the size of the change matrix is ​​as follows.

[0143]

[0144]

[0145] Among them, img weight is the pixel width of the change matrix, img height is the pixel height of the change matrix, x r is the longitude of the lower right corner, y b is the latitude of the lower right corner.

[0146] Step 2: Feature extraction

[0147] Take out the top grid code 417394478127513600-16 from the task stack one by one, and calculate the latitude and longitude coordinates of the four corners of the top grid code (lower left corner longitude: 113.042222222222222, lower left corner latitude: 23.66666666666666668, upper right corner longitude: 113.05, upper right corner latitude: 23.6755555555555555).

[0148] According to the latitude and longitude coordinates of the grid code, two images of different phases are intercepted (such as Figure 7-8 ) change characteristics. Note that the grid code is at level 16 at this time. Images with less than 20 levels can be regarded as large-size images, and CNN convolutional neural network can be used to extract features for large-size images. On the contrary, images with more than 20 levels are considered small-size images, and spectral feature extraction or histogram feature extraction can be used.

[0149]

[0150]

[0151] Where g is the grid code, g 1 , g 2 are two images of the grid code at different times, f1 There are at least two or more CNN convolutional neural network models to judge g 1 , g 2 The confidence of the category, f 2 It is to combine at least two traditional image processing algorithms to calculate g 1 , g 2 , w is the feature coefficient weight, similarity is the similarity feature value, and confidence is the confidence of the neural network in judging the grid type; f 3 is the probability that the grid code obtained from other data sources may change. When the current grid level is between 0 and 20, the convolutional neural network f is used. 1 To extract features; when the current grid level is between 20 and 31, use the traditional visual algorithm f 2 to extract features. Prior knowledge can be used at any level 3 Assist in correcting the recognition results.

[0152] Calculate the position of the current grid code, 417394478127513600-16, in the change matrix. Since the current level is 16, use method f 1 Perform feature extraction (if the level is greater than 20, the feature value extraction method uses f 2 ), and then use the averaging strategy to calculate multiple f 1 Features, that is, f 1 Multiple sets of eigenvalues ​​f 11 ,f 12 ,……,f 1n Weighted average (if the level is greater than 20, then f 2 ),like Fig. 9 At the location of the lower left area in the middle, add the eigenvalue H(x) to this position in the change matrix.

[0153]

[0154]

[0155] Add subgrid code update task stack: After the value of grid code 417394478127513600-16 in the change matrix is ​​updated, if the current grid code level has not reached the specified end level, calculate the sub-level grid code of grid code 417394478127513600-16 ["417394478127513600-17","417394479201255424-17","417394480274997248-17","417394481348739072-17"], the subgrid code is as follows Fig.10As shown in the lower left corner, the sub-grid code is added to the top of the task stack, and the grid code 417394478127513600-16 is popped out of the stack. At this time, the top grid of the task stack is the sub-grid code 417394478127513600-17 of the 17th level. The top grid code is taken out, and the operation of step 2 is repeated. The sub-grid code 417394478127513600-17 is feature extracted, and the sub-grid code is calculated after feature extraction until the level specified by the user is reached.

[0156] Step 3: Threshold segmentation

[0157] Segmentation image: Change matrix such as Fig.11 As shown, the darker the area of ​​the image, the greater the possibility of change. The purpose of image segmentation can be achieved by setting a threshold or using mathematical morphological methods.

[0158] Multi-level grid change detection method based on grid division: Different algorithm models can be used for feature extraction according to images of different sizes. For large-size images with less than 20 levels, CNN can be used for feature extraction. Otherwise, spectral differences or histograms can be used to extract change differences.

[0159] Feature fusion method based on change matrix: Through the one-to-one mapping between each grid code and the change matrix, the change features of the grid images are extracted one by one and then superimposed into the change matrix. By superimposing the change feature values ​​of different models at different levels, the fusion of multiple levels and multiple models is achieved.

[0160] In the several embodiments provided in the present application, it should be understood that the disclosed terminals, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of units is only a logical function division, and there may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0161] In addition, each functional unit in each embodiment of the present application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The above integrated unit can be implemented in the form of hardware or in the form of software functional units. The above is only an implementation method of the present application, and does not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made by using the description and drawings of this application, or directly or indirectly used in other related technical fields, is also included in the patent protection scope of the present application.

Claims

1. A change detection method based on multi-level feature fusion of subdivided grid images, characterized in that: The following steps are involved: S1, obtaining images of the area to be detected at different phases; S2, training different feature extraction models and grid models for segmenting images according to the size of the image, The feature extraction model includes a CNN feature extraction model and a variation difference extraction model; When extracting the feature extraction model, the size of each level of mesh is customized; The CNN feature extraction model is used to extract features from large-size images; The variation difference extraction model is used to extract variation differences from small-size images; When the feature extraction model performs feature extraction, the following formula is used: f3(g)=P(g) Among them: g is the grid code, g1 and g2 are two images of the grid code at different phases, f1 is the confidence of at least two CNN convolutional neural network models in judging the categories of g1 and g2, f2 is the feature similarity of g1 and g2 calculated by combining at least two traditional image processing algorithms, w is the feature coefficient weight, similarity is the similarity feature value, and confidence is the confidence of the neural network in judging the grid type; f3 is the probability that the grid code may change obtained from other data sources; When the image is of large size, the convolutional neural network f1 is used for feature extraction; When the image is of small size, the traditional visual algorithm f2 is used for feature extraction; When assisting in correcting the recognition result, the prior knowledge f3 is used to assist in correcting the recognition result; S3, gridding the current image to generate multiple grid cells with unique grid codes; S4, generating a zero-change matrix and extracting several image features at different phases; S5, importing the image into the feature extraction model to determine the grid model required for the image of the current area to be detected; S6. Map the extracted eigenvalues ​​to the positions of the change matrix through the grid code, and superimpose the eigenvalues ​​to the regional blocks of the change matrix to achieve effective fusion of multi-level and multi-models.

2. The change detection method based on multi-level feature fusion of subdivided grid images according to claim 1 is characterized in that: The feature extraction model includes a CNN feature extraction model and a change difference extraction model; When extracting the feature extraction model, the size of each level of mesh is customized; The CNN feature extraction model is used to extract features from large-size images; The variation difference extraction model is used to extract variation differences from a small-size image. When the variation difference extraction model extracts the image, it uses any one of spectral difference and histogram to extract variation differences.

3. The change detection method based on multi-level feature fusion of subdivided grid images according to claim 1 is characterized in that: When the image of the area to be detected is segmented by a grid model, the following steps are included: S31, gridding the current image through the coding unit to generate a plurality of grid units with unique grid codes; S32, initializing the task stack: obtaining a grid set corresponding to the facet of the minimum bounding rectangle of the image; S33, calculating the size of the change matrix and generating a 0 change matrix; S34. Specify the end level.

4. The change detection method based on multi-level feature fusion of subdivided grid images according to claim 1 is characterized in that: When the image features are extracted, the following steps are included: S41, extracting grid codes one by one from the task stack, and obtaining the area image corresponding to the grid code; S42, calculating a number of image features at two different time phases; S43, multi-model hybrid calculation of image feature values.

5. The change detection method based on multi-level feature fusion of subdivided grid images according to claim 4 is characterized in that: After the feature extraction is completed, the step of updating the change matrix is ​​also included, wherein the change matrix updating includes the following steps: S51, after the mixed calculation of the image eigenvalues ​​is completed, the grid code is mapped to the area in the change matrix, and the eigenvalues ​​are weighted and summed; S52, judging whether the end level has been reached: If yes: go to step S53; If not: go to step S54; S53, the sub-level grid code is put into the task stack, the current grid code is taken out of the task stack, and the process returns to step S41; S54, the current grid code outputs the task stack; S55. Determine whether the task stack is empty: If not: go to step S41; If yes: Go to the next step.

6. The change detection method based on multi-level feature fusion of subdivided grid images according to claim 5 is characterized in that: After the change matrix is ​​updated, threshold segmentation is also included, and the threshold segmentation includes segmenting the image according to the change matrix.

7. A system according to any one of claims 1 to 6, characterized in that: include: An acquisition module, which is used to acquire images of the area to be detected at different phases; An encoding module, the encoding module is used to perform grid division on the current image and generate a plurality of grid units with unique grid codes; A recognition module, the recognition module is used to recognize the size information of the current image, and select different feature extraction models to perform feature extraction according to the image size information; An extraction module, the extraction module is used to extract a number of image features at different time phases; The detection module is used to map the extracted eigenvalues ​​to the positions of the change matrix through the grid code, and to superimpose the eigenvalues ​​into the regional blocks of the change matrix to achieve effective fusion of multi-level and multi-models.

Images