An intelligent detection system for image semantic changes

Abstract

The present invention relates to the field of image detection, and more particularly to an intelligent detection system for semantic changes in images. The system uses a texture analysis module to determine the consistency parameters of each unit, a light and shadow analysis module to analyze the light and shadow interference characteristics of each unit, and a tendency analysis module to distinguish the tendency of each unit to cause irrelevant changes in remote sensing data. A change detection module adaptively filters remote sensing data, employing different analysis methods for filtered and unfiltered remote sensing data. This reduces the amount of data analysis required for remote sensing data, ensures the reliability of change detection, and improves change detection efficiency.

Description

Technical Field

[0001] The present invention relates to the field of image detection, and in particular to an intelligent detection system for image semantic changes. Background Art

[0002] Satellite remote sensing technology has developed over the years and now has the characteristics of multi-platform, multi-sensor, and multi-angle. The remote sensing images obtained have the characteristics of high spectral resolution, high radiation resolution, high spatial resolution, and high temporal resolution. These image data can be used to detect changes in surface targets and natural and non-natural phenomena, thereby better solving actual problems.

[0003] Chinese patent publication number CN110414343B discloses a method for detecting ships in on-orbit satellite remote sensing imagery. This method utilizes superpixels to segment high-resolution remote sensing images and achieve target pre-positioning while ensuring target integrity. Within the pre-positioning units, the algorithm utilizes local features and variable threshold nearest neighbor validation to achieve target detection. Within the performance constraints and real-time requirements of the onboard platform, the algorithm's operational efficiency meets the real-time detection requirements of the onboard platform and can accurately detect large ships in high-resolution remote sensing imagery.

[0004] However, the prior art still has the following problems:

[0005] The current change detection process involves massive amounts of remote sensing data, which requires a lot of computing power to analyze. In addition, some changes in remote sensing data are not substantial and may be caused by atmospheric interference or changes in illumination, resulting in low change detection efficiency and poor accuracy. Summary of the Invention

[0006] To this end, the present invention provides an intelligent detection system for image semantic changes to overcome the problems in the prior art that the analysis of massive remote sensing data requires a large amount of computing power, and some changes in remote sensing data may be caused by atmospheric interference or changes in lighting, resulting in low change detection efficiency and poor accuracy.

[0007] To solve the above problems, the present invention provides an intelligent detection system for image semantic changes, comprising:

[0008] Texture analysis module, which is used to extract the color histogram and contour features of the historical remote sensing data in each unit, and analyze the visual consistency parameters in each unit based on the difference in the color histogram and the distribution difference of the contour features;

[0009] A light and shadow analysis module is used to determine the light and shadow specific time domain segment based on the light and shadow characteristics of different historical time nodes in each unit, and analyze the light and shadow interference parameters of each unit according to the changes in historical remote sensing data within the light and shadow specific time domain segment;

[0010] a tendency analysis module, connected to the texture analysis module and the light and shadow analysis module, respectively, for determining remote sensing change excitation characteristics based on the visual consistency parameter and the light and shadow interference parameter, so as to distinguish the tendency of each unit to cause irrelevant changes in remote sensing data;

[0011] A change detection module is used to screen the remote sensing data based on the tendency of each unit to cause irrelevant changes in the remote sensing data, and analyze the screened remote sensing data, including:

[0012] The extraction unit corresponds to the dual-temporal remote sensing data, inputs the dual-temporal remote sensing data into a pre-trained change recognition model, and determines whether the dual-temporal remote sensing data has undergone substantial changes based on the change recognition model;

[0013] The time phase interval of the dual-time phase remote sensing data is predetermined.

[0014] Furthermore, the texture analysis module is used to determine the difference in color histograms and the distribution difference in contour features, wherein:

[0015] The texture analysis module determines the difference amount based on the Euclidean distance between the color histograms of the corresponding historical remote sensing data at each moment in the unit;

[0016] The texture analysis module determines the average distribution distance of each contour feature based on the historical remote sensing data corresponding to each moment in the unit, and determines the distribution difference of the contour features based on the average distribution distance.

[0017] Furthermore, the texture analysis module is used to calculate visual consistency parameters, including:

[0018] for determining a ratio of the difference amount to a predetermined difference amount threshold as a color consistency component;

[0019] for determining the ratio of the distribution difference to the distribution difference threshold as the distribution consistency component;

[0020] It is used to weight the sum of the color consistency component and the distribution consistency component to obtain the visual consistency parameter.

[0021] Furthermore, the light and shadow analysis module determines the light and shadow specific time domain segments based on the light and shadow characteristics of different historical time nodes in each unit, including:

[0022] Used to obtain the light and shadow characteristics of the unit at different historical time nodes, including solar altitude angle, solar azimuth angle and atmospheric transmittance;

[0023] Used to identify the mutation time period based on the changes in light and shadow characteristics in the time domain dimension;

[0024] It is used to merge consecutive mutation time periods to form a continuous time period, which is defined as the light-shadow specific time domain segment.

[0025] Furthermore, the light and shadow analysis module analyzes the light and shadow interference parameters of each unit based on the changes in historical remote sensing data within the light and shadow specific time domain segment, including:

[0026] To obtain the historical remote sensing data of the unit at the start and end of the light and shadow specific time domain period, and determine the amount of chromaticity change and contour area change;

[0027] It is used to determine the light-shadow interference parameters according to the chromaticity change and the contour area change.

[0028] Furthermore, the tendency analysis module is used to determine the remote sensing change excitation characteristics based on the visual consistency parameters and the light and shadow interference parameters, including:

[0029] It is used to perform weighted summation of visual consistency parameters and light-shadow interference parameters to obtain remote sensing change excitation features.

[0030] Furthermore, the trend analysis module is used to distinguish the tendency of each unit to cause irrelevant changes in remote sensing data, wherein:

[0031] If the remote sensing change excitation feature corresponding to the unit is greater than or equal to a predetermined remote sensing change excitation feature threshold, then the tendency of the unit to cause irrelevant changes in the remote sensing data is determined to be a first tendency;

[0032] If the remote sensing change excitation feature corresponding to the unit is less than a predetermined remote sensing change excitation feature threshold, the tendency of the unit to cause irrelevant changes in the remote sensing data is determined to be the second tendency.

[0033] Furthermore, the change detection module screens the remote sensing data based on the tendency of each unit to cause irrelevant changes in the remote sensing data, wherein:

[0034] Filter out the remote sensing data corresponding to the unit with the first tendency.

[0035] Furthermore, the change detection module is also used to analyze the unfiltered remote sensing data, including:

[0036] The difference features between the unfiltered remote sensing data and the basic remote sensing data include the difference in color histogram and the distribution difference in contour features;

[0037] If the difference condition is met, it is determined that there is a substantial change in the unit corresponding to the remote sensing data;

[0038] The difference condition includes that the difference amount is greater than a predetermined sample difference amount threshold and the distribution difference is greater than a predetermined sample distribution difference threshold.

[0039] Furthermore, the change detection module is used to call historical remote sensing data based on the remote sensing data and the time interval, and combine the remote sensing data with the historical remote sensing data to form dual-time remote sensing data.

[0040] Compared with the prior art, the beneficial effect of the present invention lies in that the present invention determines the consistency parameters of each unit through a texture analysis module, analyzes the light and shadow interference characteristics of each unit through a light and shadow analysis module, distinguishes the tendency of each unit to cause irrelevant changes in remote sensing data through a tendency analysis module, and adaptively filters remote sensing data through a change detection module. Different analysis methods are used for filtered remote sensing data and non-filtered remote sensing data, thereby reducing the amount of data analysis on remote sensing data, ensuring the reliability of change detection, and improving the efficiency of change detection.

[0041] In particular, the texture analysis module of the present invention determines the visual consistency parameter. In actual situations, the geographical features in each unit are different, and their tendency to be affected by the atmosphere and light and shadow characteristics is different. Therefore, the tendency of the corresponding remote sensing image to produce irrelevant changes is also different. Based on this, the present invention considers the difference in the color histogram of the historical remote sensing data corresponding to the identification unit and the distribution difference of the contour features. In actual situations, the geographical features of some units are simple, such as plains and inland seas. In their historical remote sensing data, the visual color is single and the variability is small. Therefore, the color histogram is used to reflect the variability of the visual color of the unit corresponding to the historical remote sensing data, and the unit The contour features of the corresponding remote sensing data reflect the entity contours or light and shadow contours in the unit to a certain extent. The distribution changes of the contour features can reflect the changes in the light and shadow contours of the entity after light and shadow exposure to a certain extent, and then reflect the tendency of irrelevant changes in the remote sensing data. If the geographical features of the unit are simple, the light and shadow exposure has little effect on its remote sensing data, and the corresponding distribution differences of its contour features are small. Based on this, the visual consistency parameters are calculated to provide a basis for the subsequent determination of remote sensing change excitation features, facilitate the adaptive analysis of remote sensing data, and thus reduce the amount of data analysis of remote sensing data, ensure the reliability of change detection, and improve the efficiency of change detection.

[0042] In particular, the light and shadow characteristics of different historical time nodes in each unit will change rapidly in part of the time. By analyzing the remote sensing data before and after the rapid change, it can be better reflected that the remote sensing data in the unit is affected by the light and shadow changes. The present invention determines the light and shadow specific time domain segment, analyzes the changes in historical remote sensing data, and then determines the light and shadow interference parameters, which can reflect the tendency of the remote sensing data in the unit to produce irrelevant changes due to light and shadow illumination. The data is highly representative, providing data support for the subsequent calculation of remote sensing change excitation characteristics, facilitating the adaptive analysis of remote sensing data, thereby reducing the amount of data analysis of remote sensing data, ensuring the reliability of change detection, and improving the efficiency of change detection.

[0043] In particular, the present invention considers and calculates remote sensing change excitation characteristics from two dimensions, and considers the tendency of the unit to produce irrelevant changes in the corresponding remote sensing data. For the units with the first tendency, the tendency of the corresponding remote sensing data to produce irrelevant changes is relatively high, and a deep analysis method that consumes more computing power is adopted to ensure that the change characteristics are captured and avoid omissions. For the units with the second tendency, the tendency of the corresponding remote sensing data to produce irrelevant changes is relatively low, and the changes they produce are mostly substantial changes. Therefore, the changes in their more basic image features can be directly used to identify whether they have produced substantial changes, thereby reducing the amount of data analysis of the remote sensing data, ensuring the reliability of change detection, and improving the efficiency of change detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 This is a schematic diagram of the structure of an intelligent detection system for image semantic changes according to an embodiment of the invention;

[0045] Figure 2 A logic block diagram for distinguishing the tendency of each unit to cause irrelevant changes in remote sensing data according to an embodiment of the present invention;

[0046] Figure 3 A logic block diagram of an embodiment of the invention for filtering remote sensing data based on the tendency of each unit to cause irrelevant changes in the remote sensing data;

[0047] Figure 4 This is a logic block diagram for analyzing filtered remote sensing data and unfiltered remote sensing data according to an embodiment of the present invention. DETAILED DESCRIPTION

[0048] In order to make the objects and advantages of the present invention more clearly understood, the present invention is further described below in conjunction with embodiments; it should be understood that the specific embodiments described herein are merely used to explain the present invention and are not intended to limit the present invention.

[0049] The preferred embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood by those skilled in the art that these embodiments are only used to explain the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0050] It should be noted that, in the description of the present invention, terms such as "up", "down", "left", "right", "inside", and "outside" indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings. This is only for the convenience of description and does not indicate or imply that the device or element must have a specific orientation, be constructed and operated in a specific orientation. Therefore, it should not be understood as a limitation on the present invention.

[0051] Furthermore, it should be noted that, in the description of the present invention, unless otherwise expressly specified or limited, the terms "mounted," "connected," and "connected" should be understood in a broad sense. For example, they may refer to fixed connections, detachable connections, or integral connections; mechanical connections or electrical connections; direct connections or indirect connections through an intermediate medium; and internal communication between two components. Those skilled in the art will understand the specific meanings of the above terms in the present invention based on specific circumstances.

[0052] See also Figure 1 As shown, Figure 1 This is a schematic diagram of the structure of an intelligent detection system for image semantic changes according to an embodiment of the invention. The intelligent detection system for image semantic changes according to an embodiment of the invention comprises:

[0053] Texture analysis module, which is used to extract the color histogram and contour features of the historical remote sensing data in each unit, and analyze the visual consistency parameters in each unit based on the difference in the color histogram and the distribution difference of the contour features;

[0054] A light and shadow analysis module is used to determine the light and shadow specific time domain segment based on the light and shadow characteristics of different historical time nodes in each unit, and analyze the light and shadow interference parameters of each unit according to the changes in historical remote sensing data within the light and shadow specific time domain segment;

[0055] A tendency analysis module, which is connected to the texture analysis module and the light and shadow analysis module respectively, is used to determine the remote sensing change excitation characteristics based on the visual consistency parameter and the light and shadow interference parameter, so as to distinguish the tendency of each unit to cause irrelevant changes in the remote sensing data;

[0056] A change detection module is used to screen the remote sensing data based on the tendency of each unit to cause irrelevant changes in the remote sensing data, and analyze the screened remote sensing data, including:

[0057] The extraction unit corresponds to the dual-temporal remote sensing data, inputs the dual-temporal remote sensing data into a pre-trained change recognition model, and determines whether the dual-temporal remote sensing data has undergone substantial changes based on the change recognition model;

[0058] The time phase interval of the dual-time phase remote sensing data is predetermined.

[0059] Specifically, there is no limitation on the specific structures of the texture analysis module, light and shadow analysis module, tendency analysis module and change detection module, and they can all be composed of logic components or a combination of logic components, and the logic components include field programmable processors, computers or microprocessors in computers.

[0060] Specifically, in implementation, a unit is a geospatial area whose extent is equal to the image coverage of the remote sensing image.

[0061] Specifically, in implementation, remote sensing data refers to remote sensing images acquired through acquisition equipment, which will not be elaborated here.

[0062] Specifically, there is no limitation on the form of the change recognition model. Those skilled in the art can choose an open source image processing model in the existing technology that can identify substantial changes in remote sensing data, or they can train a change recognition model to implement the corresponding function by themselves. For example, they can manually mark a number of remote sensing data with substantial changes in advance as training samples to train the corresponding change recognition model. There is no limitation on the model architecture of the change recognition model. For example, a neural network architecture can be used, which will not be elaborated here.

[0063] Specifically, the texture analysis module is used to determine the difference in color histograms and the distribution difference in contour features, where:

[0064] The texture analysis module determines the difference amount based on the Euclidean distance between the color histograms of the corresponding historical remote sensing data at each moment in the unit;

[0065] It can be understood that there are multiple color histograms, and the Euclidean distances between the color histograms need to be calculated respectively, and the mean of the Euclidean distances is determined as the difference amount;

[0066] The texture analysis module determines the average distribution distance of each contour feature based on the historical remote sensing data corresponding to each moment in the unit, and determines the distribution difference of the contour features based on the average distribution distance.

[0067] In the implementation, the average distribution distance is the mean of the distances between the center of each contour and the center of the nearest contour;

[0068] The variance of the average distribution distance corresponding to the historical remote sensing data at each moment is determined as the distribution difference.

[0069] In the implementation, a color histogram of the RGB color system is used, and the color histogram includes sub-color histograms of three color channels;

[0070] Therefore, when calculating the Euclidean distance between two color histograms, it is necessary to calculate the sub-Euclidean distances between the sub-color histograms of the corresponding color channels respectively, and use the mean of the sub-Euclidean distances as the Euclidean distance between the two color histograms.

[0071] Specifically, the texture analysis module is used to calculate visual consistency parameters, including:

[0072] for determining a ratio of the difference amount to a predetermined difference amount threshold as a color consistency component;

[0073] for determining the ratio of the distribution difference to the distribution difference threshold as the distribution consistency component;

[0074] It is used to weight the sum of the color consistency component and the distribution consistency component to obtain the visual consistency parameter.

[0075] Specifically, in implementation, the weight of the color consistency component is 0.55, and the weight of the distribution consistency component is 0.45.

[0076] Specifically, the predetermined difference threshold and the distribution difference threshold are predetermined, wherein the historical remote sensing data in the corresponding units of the desert, ocean and Gobi are determined in advance by technical personnel in this field, the difference and distribution difference corresponding to each unit are determined, the mean difference and the mean distribution difference are solved, the difference threshold is set to the product of the mean difference and the error offset coefficient, and the distribution difference threshold is set to the product of the mean distribution difference and the error offset coefficient. The error offset coefficient is selected within the interval [1.15, 1.25].

[0077] The texture analysis module of the present invention determines the visual consistency parameters. In actual situations, the geographical features in each unit are different, and their tendencies to be affected by the atmosphere and light and shadow characteristics are different. Therefore, the corresponding remote sensing images also have different tendencies to produce irrelevant changes. Based on this, the present invention considers the difference in the color histogram of the historical remote sensing data corresponding to the identification unit and the distribution difference of the contour features. In actual situations, the geographical features of some units are simple, such as plains and inland seas. In their historical remote sensing data, the visual color is single and the variability is small. Therefore, the color histogram is used to reflect the variability of the visual color of the unit corresponding to the historical remote sensing data, and the unit is The contour features of remote sensing data reflect the entity contours or light and shadow contours in the unit to a certain extent. The distribution changes of contour features can reflect the changes in the light and shadow contours of the entity after light and shadow exposure to a certain extent, and then reflect the tendency of irrelevant changes in remote sensing data. If the geographical features of the unit are simple, the light and shadow exposure has little effect on its remote sensing data, and the corresponding distribution differences of its contour features are small. Based on this, the visual consistency parameters are calculated to provide a basis for the subsequent determination of remote sensing change excitation features, facilitate the adaptive analysis of remote sensing data, and thus reduce the amount of data analysis of remote sensing data, ensure the reliability of change detection, and improve the efficiency of change detection.

[0078] Specifically, the light and shadow analysis module determines the light and shadow specific time domain segments based on the light and shadow characteristics of different historical time nodes in each unit, including:

[0079] Used to obtain the light and shadow characteristics of the unit at different historical time nodes, including solar altitude angle, solar azimuth angle and atmospheric transmittance;

[0080] Used to identify the mutation time period based on the changes in light and shadow characteristics in the time domain dimension;

[0081] It is used to merge consecutive mutation time periods to form a continuous time period, which is defined as the light-shadow specific time domain segment.

[0082] In the implementation, the changing speed of the solar altitude angle, the changing speed of the solar azimuth angle and the changing speed of the atmospheric transmittance in each time period are determined respectively, and the time period is selected in the interval [5 minutes, 10 minutes];

[0083] If the solar altitude angle change rate within the time period is greater than a predetermined altitude angle change rate threshold, or the solar azimuth angle change rate is greater than a predetermined azimuth angle change rate threshold, or the atmospheric transmittance change rate is greater than a predetermined transmittance change rate threshold, then the time period is determined to be a mutation time period.

[0084] The altitude angle change rate threshold is selected within 1.3 to 1.5 times the historical average of the solar altitude angle change rate;

[0085] The azimuth angle change rate threshold is selected within 1.3 to 1.5 times the historical average of the solar azimuth angle change rate;

[0086] The transmittance change rate threshold is selected within 1.1 to 1.3 times the historical average atmospheric transmittance change rate.

[0087] Specifically, the solar altitude angle, solar azimuth angle and atmospheric transmittance can be obtained based on the geographic remote sensing information of the area where the unit is located. For example, it can be obtained by directly calling the authorized remote sensing information database, or by collecting remote sensing information of the corresponding area in real time. This will not be repeated here.

[0088] The light and shadow characteristics of different historical time nodes in each unit will change rapidly in part of the time. By analyzing the remote sensing data before and after the rapid change, it can be better reflected that the remote sensing data in the unit is affected by the light and shadow changes. The present invention determines the light and shadow specific time domain segment, analyzes the changes in historical remote sensing data, and then determines the light and shadow interference parameters. It can reflect the tendency of the remote sensing data in the unit to produce irrelevant changes due to light and shadow illumination. The data is highly representative and provides data support for the subsequent calculation of remote sensing change excitation characteristics, which facilitates the adaptive analysis of remote sensing data, thereby reducing the amount of data analysis of remote sensing data, ensuring the reliability of change detection, and improving the efficiency of change detection.

[0089] Specifically, the light and shadow analysis module analyzes the light and shadow interference parameters of each unit based on the changes in historical remote sensing data within the light and shadow specific time domain segment, including:

[0090] To obtain the historical remote sensing data of the unit at the start and end of the light and shadow specific time domain period, and determine the amount of chromaticity change and the amount of contour area change;

[0091] It is used to determine the light-shadow interference parameters according to the chromaticity change and the contour area change.

[0092] In implementation, the ratio of the chromaticity change amount to the standard chromaticity change amount threshold is calculated to obtain the chromaticity light and shadow interference component;

[0093] Calculate the ratio of the contour area change to the standard contour area change threshold to obtain the contour area light and shadow interference component;

[0094] The chromaticity light and shadow interference component and the contour area light and shadow interference component are weighted and summed to obtain the light and shadow interference parameter;

[0095] In implementation, the weight of the chromaticity light and shadow interference component is 0.35, and the weight of the contour area light and shadow interference component is 0.65.

[0096] The standard chromaticity change threshold and the standard contour area change threshold are predetermined, wherein historical remote sensing data of several units in the light and shadow specific time domain segment are obtained in advance, the chromaticity change and the contour area change are recorded, the mean chromaticity change and the mean contour area change for each unit are solved, the standard chromaticity change threshold is set as the product of the mean chromaticity change and the error coefficient, and the standard contour area change threshold is set as the product of the mean contour area change and the error coefficient, and the error coefficient is selected within the interval [0.78, 0.84].

[0097] Specifically, the tendency analysis module is used to determine the remote sensing change excitation characteristics based on the visual consistency parameter and the light and shadow interference parameter, including:

[0098] The visual consistency parameter and the light-shadow interference parameter are weighted and summed to obtain the remote sensing change excitation feature.

[0099] In implementation, the weights of the visual consistency parameter and the light and shadow interference parameter are both 0.5.

[0100] See also Figure 2 As shown, Figure 2 This is a logic block diagram for distinguishing the tendency of each unit to cause irrelevant changes in remote sensing data according to an embodiment of the invention; the tendency analysis module is used to distinguish the tendency of each unit to cause irrelevant changes in remote sensing data, wherein:

[0101] If the remote sensing change excitation feature corresponding to the unit is greater than or equal to a predetermined remote sensing change excitation feature threshold, then the tendency of the unit causing irrelevant changes in the remote sensing data is determined to be a first tendency;

[0102] If the remote sensing change excitation feature corresponding to the unit is less than a predetermined remote sensing change excitation feature threshold, the tendency of the unit to cause irrelevant changes in the remote sensing data is determined to be the second tendency.

[0103] In implementation, the remote sensing change excitation feature threshold is selected in the interval [1.23, 1.36].

[0104] Specifically, see Figure 3 As shown, Figure 3 This is a logic block diagram of an embodiment of the invention for filtering remote sensing data based on the tendency of each unit to cause irrelevant changes in the remote sensing data. The change detection module filters the remote sensing data based on the tendency of each unit to cause irrelevant changes in the remote sensing data, wherein:

[0105] Filter out the remote sensing data corresponding to the unit with the first tendency.

[0106] In implementation, the remote sensing data corresponding to all the screened units with the first tendency are summarized into a first data set, and the remote sensing data corresponding to all the unscreened units are summarized into a second data set.

[0107] Specifically, see Figure 4 As shown, Figure 4 This is a logic block diagram for analyzing filtered remote sensing data and unfiltered remote sensing data according to an embodiment of the present invention.

[0108] Specifically, the change detection module is also used to analyze the unfiltered remote sensing data, including:

[0109] The difference features between the unfiltered remote sensing data and the basic remote sensing data include the difference in color histogram and the distribution difference in contour features;

[0110] If the difference condition is met, it is determined that there is a substantial change in the unit corresponding to the remote sensing data;

[0111] The difference condition includes that the difference amount is greater than a predetermined sample difference amount threshold and the distribution difference is greater than a predetermined sample distribution difference threshold.

[0112] Specifically, the basic remote sensing data is pre-collected remote sensing data within a unit for reference, the purpose of which is to consider changes in subsequent remote sensing images based on the basic remote sensing data.

[0113] Specifically, the sample difference threshold and the sample distribution difference threshold are predetermined. A number of remote sensing data with substantial changes in units are calibrated in advance by technicians in this field to determine the difference characteristics of each remote sensing data relative to the basic remote sensing data, and then the difference value mean of the color histogram and the distribution difference mean of the contour feature are solved. The sample difference threshold is set as the product of the difference value mean of the color histogram and the adjustment coefficient, and the sample distribution difference threshold is set as the product of the distribution difference mean of the contour feature and the adjustment coefficient. The adjustment coefficient is selected within the interval [0.65, 0.85].

[0114] Specifically, the change detection module is used to call historical remote sensing data based on the remote sensing data and the time interval, and combine the remote sensing data with the historical remote sensing data to form dual-time remote sensing data.

[0115] The time interval is a time period determined based on actual needs. Those skilled in the art may set the time interval based on the changes in remote sensing data relative to historical remote sensing data at a certain time, which will not be elaborated here.

[0116] The present invention considers and calculates remote sensing change excitation characteristics from two dimensions. The consideration is the tendency of the unit to produce irrelevant changes in the corresponding remote sensing data. For the unit with the first tendency, the tendency of the corresponding remote sensing data to produce irrelevant changes is relatively high. A deep analysis method that consumes more computing power is adopted to ensure that the change characteristics are captured and avoid omissions. For the unit with the second tendency, the tendency of the corresponding remote sensing data to produce irrelevant changes is relatively low, and the changes it produces are mostly substantial changes. Therefore, the changes in its more basic image features can be directly used to identify whether it has produced substantial changes, thereby reducing the amount of data analysis of the remote sensing data, ensuring the reliability of change detection, and improving the efficiency of change detection.

[0117] Thus far, the technical solutions of the present invention have been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art may make equivalent changes or substitutions to the relevant technical features, and the technical solutions after such changes or substitutions will fall within the scope of protection of the present invention.

Claims

1. An intelligent detection system for image semantic changes, characterized in that: include: Texture analysis module, which is used to extract the color histogram and contour features of the historical remote sensing data in each unit, and analyze the visual consistency parameters in each unit based on the difference in the color histogram and the distribution difference of the contour features; A light and shadow analysis module is used to determine the light and shadow specific time domain segment based on the light and shadow characteristics of different historical time nodes in each unit, and analyze the light and shadow interference parameters of each unit according to the changes in historical remote sensing data within the light and shadow specific time domain segment; a tendency analysis module, connected to the texture analysis module and the light and shadow analysis module, respectively, for determining remote sensing change excitation characteristics based on the visual consistency parameter and the light and shadow interference parameter, so as to distinguish the tendency of each unit to cause irrelevant changes in remote sensing data; The change detection module is used to filter the remote sensing data based on the tendency of each unit to cause irrelevant changes in the remote sensing data, and analyze the filtered remote sensing data. include, The extraction unit corresponds to the dual-temporal remote sensing data, inputs the dual-temporal remote sensing data into a pre-trained change recognition model, and determines whether the dual-temporal remote sensing data has undergone substantial changes based on the change recognition model; Wherein, the temporal interval of the dual-temporal remote sensing data is predetermined; The texture analysis module is used to calculate visual consistency parameters, including: for determining a ratio of the difference amount to a predetermined difference amount threshold as a color consistency component; for determining a ratio of the distribution difference to a distribution difference threshold as a distribution consistency component; to perform weighted summation of the color consistency component and the distribution consistency component to obtain the visual consistency parameter; The light and shadow analysis module analyzes the light and shadow interference parameters of each unit based on the changes in historical remote sensing data within the light and shadow specific time domain segment, including: To obtain the historical remote sensing data of the unit at the start and end of the light and shadow specific time domain period, and determine the amount of chromaticity change and the amount of contour area change; Used to determine light and shadow interference parameters based on chromaticity change and contour area change; The tendency analysis module is used to determine the remote sensing change excitation characteristics based on the visual consistency parameter and the light and shadow interference parameter, including: The visual consistency parameter and the light-shadow interference parameter are weighted and summed to obtain the remote sensing change excitation feature.

2. The intelligent detection system for image semantic changes according to claim 1, characterized in that: The texture analysis module is used to determine the difference in color histogram and the distribution difference of contour features, wherein: The texture analysis module determines the difference based on the Euclidean distance between the color histograms of the historical remote sensing data corresponding to each moment in the unit; The texture analysis module determines the average distribution distance of each contour feature based on the historical remote sensing data corresponding to each moment in the unit, and determines the distribution difference of the contour features based on the average distribution distance.

3. The intelligent detection system for image semantic changes according to claim 1, characterized in that: The light and shadow analysis module determines the light and shadow specific time domain segment based on the light and shadow characteristics of different historical time nodes in each unit, including: Used to obtain the light and shadow characteristics of the unit at different historical time nodes, including solar altitude angle, solar azimuth angle and atmospheric transmittance; Used to identify the mutation time period based on the changes in light and shadow characteristics in the time domain dimension; It is used to merge consecutive mutation time periods to form a continuous time period, which is defined as the light-shadow specific time domain segment.

4. The intelligent detection system for image semantic changes according to claim 1, characterized in that: The tendency analysis module is used to distinguish the tendency of each unit to cause irrelevant changes in remote sensing data, wherein: If the remote sensing change excitation feature corresponding to the unit is greater than or equal to a predetermined remote sensing change excitation feature threshold, then the tendency of the unit causing irrelevant changes in the remote sensing data is determined to be a first tendency; If the remote sensing change excitation feature corresponding to the unit is less than a predetermined remote sensing change excitation feature threshold, the tendency of the unit to cause irrelevant changes in the remote sensing data is determined to be the second tendency.

5. The intelligent detection system for image semantic changes according to claim 1, characterized in that: The change detection module screens the remote sensing data based on the tendency of each unit to cause irrelevant changes in the remote sensing data, wherein: Filter out the remote sensing data corresponding to the unit with the first tendency.

6. The intelligent detection system for image semantic changes according to claim 5, characterized in that: The change detection module is also used to analyze the unfiltered remote sensing data, including: The difference features between the unfiltered remote sensing data and the basic remote sensing data include the difference in color histogram and the distribution difference in contour features; If the difference condition is met, it is determined that there is a substantial change in the unit corresponding to the remote sensing data; The difference condition includes that the difference amount is greater than a predetermined sample difference amount threshold and the distribution difference is greater than a predetermined sample distribution difference threshold.

7. The intelligent detection system for image semantic changes according to claim 1, characterized in that: The change detection module is used to call historical remote sensing data based on remote sensing data and time intervals, and combine the remote sensing data with the historical remote sensing data to form dual-time remote sensing data.

Images