A high-resolution remote sensing monitoring system for smart cities

Abstract

A high-resolution remote sensing monitoring system for smart cities, relating to the field of remote sensing data technology, includes a monitoring center. The monitoring center is communicatively connected to a data acquisition module, a data processing module, a region matching module, a threshold determination module, and a land cover type extraction module. The data acquisition module acquires spectral and texture feature sets for different regions of the target city. The data processing module displays the feature curves of the spectral and texture feature sets on a three-dimensional coordinate system. The region matching module labels land cover types in different regions of the three-dimensional coordinate system and determines whether there are confused regions. The threshold determination module acquires a set of threshold values ​​for land cover types in all regions, as well as an additional set of threshold values ​​for confused regions. The land cover type extraction module acquires the land cover types of each region in the target city at the current time. This significantly improves the accuracy of land cover classification and target identification.

Description

Technical Field

[0001] The present invention relates to the field of remote sensing data technology, and in particular to a high-resolution remote sensing monitoring system for smart cities. Background Art

[0002] Prior art 2020103862829 "Urban green space high-resolution remote sensing monitoring method and system" This method includes a training sample set construction step, a multi-dimensional feature space construction step, a U-Net+ model construction step, and an image post-processing step. By constructing a multi-dimensional feature space and enhancing feature richness, a U-Net+ deep learning model is constructed at the same time, and then combined with the image post-processing method, the generalization and robustness of the monitoring method are improved, and the overfitting problem that is prone to occur due to limited training samples is solved, thereby achieving the goal of improving the accuracy and timeliness of high-resolution remote sensing monitoring of urban green spaces.

[0003] Prior art CN116343051B "A geological environment monitoring method and system based on remote sensing images" This method performs local analysis on the remote sensing image, and takes the local area containing only two grayscale value concentrated distribution areas as the area to be analyzed, and obtains the coefficient distribution structure ratio of the corresponding area to be analyzed according to the energy coefficient ratio of the two grayscale value concentrated distribution areas in the area to be analyzed, and then determines the corresponding high-frequency coefficient set and low-frequency coefficient set in the basic frequency domain coefficient, and further adjusts the high-frequency coefficient to obtain the adjusted high-frequency coefficient set, performs inverse transformation on the corresponding area of ​​the area to be analyzed in the transformed image to obtain a feature image, performs environmental monitoring based on the feature image, obtains accurate high-frequency coefficient distribution based on local analysis, and performs geological environment monitoring through feature images with obvious geological edge features, thereby improving the efficiency and reference of geological environment monitoring.

[0004] Environmental monitoring, resource exploration, regional multi-source fusion, regional building identification, regional road network identification, regional land use dynamic detection, key area illegal construction identification and other issues in urban management require a lot of manpower, material and financial resources to be handled using traditional manual methods, and the cycle is long, the effect is poor, and there is a lot of repetitive work. Therefore, more and more studies are trying to use artificial intelligence methods to deal with these problems. Using multi-temporal remote sensing images to detect changes on the earth's surface is one of the most important applications of remote sensing technology.

[0005] When processing high-resolution remote sensing data, it is necessary to classify objects and identify targets, such as building identification and road extraction. However, the urban environment is complex and changeable, and there are occlusions and similarities between objects. There are various interference sources in the urban environment, such as weather, clouds, and light. These interferences will affect the acquisition and processing of remote sensing data, reduce the accuracy of monitoring results, and pose challenges to object classification and target identification. In addition, the features of objects such as green spaces, water bodies, and roads are significant and vary greatly. They can be distinguished by thresholds based on data features. The difficulty lies in the problem that the feature differences between object types are not significant. For example, buildings and roads are highly similar in spectral and texture features and are difficult to distinguish. How to solve the complexity of building surfaces and the problem of insignificant feature differences is an issue that we urgently need to solve. Summary of the invention

[0006] In order to solve the above technical problems, the object of the present invention is to provide a high-resolution remote sensing monitoring system for smart cities, including a monitoring center, wherein the monitoring center is communicatively connected to a data acquisition module, a data processing module, a region matching module, a threshold determination module and a ground object type extraction module;

[0007] The data acquisition module is used to collect high-resolution remote sensing monitoring data of the target city and obtain spectral feature sets and texture feature sets of different areas of the target city;

[0008] The data processing module is used to construct a three-dimensional coordinate system and display characteristic curves of the spectral feature set and the texture feature set of different areas of the target city on the three-dimensional coordinate system;

[0009] The region matching module is used to mark the types of objects in different regions in the three-dimensional coordinate system and determine whether there is a confusing region in the three-dimensional coordinate system based on the characteristic curve;

[0010] The threshold determination module is used to obtain a determination threshold set of the ground object type in all regions and an additional determination threshold set of the confusion region according to the characteristic curve;

[0011] The ground feature type extraction module is used to construct a ground feature type extraction model, and obtain the ground feature type of each area of ​​the target city at the current moment according to the high-resolution remote sensing data of the target city at the current moment and the ground feature type extraction model.

[0012] Furthermore, the data acquisition module collects high-resolution remote sensing monitoring data of the target city and obtains spectral feature sets and texture feature sets of different areas of the target city, and the process includes:

[0013] The high-resolution remote sensing data of the target city collected by the high-resolution remote sensing satellite is obtained, and the spectral reflectance data of each band in the relevant band combination of surface reflectance of different areas of the target city is obtained according to the high-resolution remote sensing data, and a number of spectral indexes of different areas of the target city are obtained according to the spectral reflectance data bands, and a spectral feature set of different areas of the target city is constructed according to the spectral index data, and a grayscale co-occurrence matrix of the high-resolution remote sensing data is constructed at the same time, and a number of texture eigenvalues ​​in the grayscale co-occurrence matrix are extracted, and a texture feature set of different areas of the target city is constructed according to the several texture eigenvalues ​​in the grayscale co-occurrence matrix.

[0014] Furthermore, the spectral reflectance data of each band in the band combination related to the surface reflectance of the target city mainly includes red, green and blue bands, which are usually used to obtain the real color information and morphological characteristics of the surface, where:

[0015] Infrared band: including near-infrared band and far-infrared band, used to monitor vegetation health, land cover changes, etc.;

[0016] Short-wave infrared band: including short-wave infrared 1 band and short-wave infrared 2 band, used to detect vegetation moisture content, soil characteristics, etc.;

[0017] Microwave band: used to penetrate clouds and vegetation to obtain information such as surface topography and soil moisture;

[0018] Several spectral indices obtained by band calculation include but are not limited to: Normalized Difference Vegetation Index NDVI, Normalized Difference Water Index NDWI, where:

[0019]

[0020] In the formula, NIR represents the reflectance value of the near-infrared band, Red represents the reflectance value of the red band, and the NDVI value range is between -1 and 1, where negative values ​​indicate non-vegetation coverage (such as water bodies or bare soil), values ​​close to 0 indicate sparse vegetation coverage, and values ​​close to 1 indicate dense vegetation coverage;

[0021]

[0022] In this formula, NIR represents the reflectance value of the near-infrared band, and Green represents the reflectance value of the green band;

[0023] Furthermore, the process of constructing the texture feature set of the target city includes:

[0024] Using the acquired high-resolution remote sensing data, the grayscale co-occurrence matrix around each pixel is calculated by sliding the window. The specific details include quantizing the grayscale level of the remote sensing image into discrete grayscale levels, dividing the grayscale value range into several levels, for example, 8-bit images can be divided into 16, 32, and 64 levels, and defining the parameters required for the grayscale co-occurrence matrix, including distance (d) and direction (θ), which are determined according to actual application requirements. Then, based on the calculated grayscale co-occurrence matrix, a series of texture feature values ​​are extracted to construct a texture feature set. The texture features in the texture feature set include but are not limited to:

[0025] Contrast: describes the statistical characteristics of the contrast between pixels of different gray levels in an image;

[0026] Correlation: describes the degree of correlation between pixels of different gray levels in an image;

[0027] Energy: describes the uniformity of the grayscale distribution of pixels in an image;

[0028] Homogeneity: describes the similarity of the gray levels of adjacent pixels in an image;

[0029] Entropy: describes the degree of uncertainty of image texture.

[0030] Furthermore, the data processing module constructs a three-dimensional coordinate system and displays the characteristic curves of the spectral feature set and the texture feature set of different areas of the target city on the three-dimensional coordinate system, and the process includes:

[0031] A three-dimensional coordinate system is established, a city map of the target city is obtained, and the city map of the target city is mapped to the xy-axis plane in the three-dimensional coordinate system, and the xy-axis mapping coordinates of the spectral feature sets and texture feature sets of different areas of the target city on the GIS surface plan of the target city are obtained, and according to the xy-axis mapping coordinates, a number of spectral indices in the spectral feature set and a number of texture feature values ​​in the texture feature set are input into the z-axis corresponding to the xy-axis mapping coordinates, and then the z-axis values ​​corresponding to the several spectral indices in the spectral feature set on different xy-axis mapping coordinates are plotted into a curve to obtain the characteristic curves of several spectral indices of the target city. Similarly, the characteristic curves of several texture feature values ​​of the target city are obtained.

[0032] Furthermore, the region matching module labels the types of objects in different regions in the three-dimensional coordinate system, obtains the discrimination coefficients corresponding to a plurality of spectral indices and a plurality of texture feature values ​​between different types of objects according to the characteristic curve, and determines whether there is a confusing region in the three-dimensional coordinate system according to the discrimination coefficient. The process includes:

[0033] Annotate the xy-axis plane in the three-dimensional coordinate system with the type of the object, obtain the xy-axis mapping coordinate sets corresponding to the different object types in the target city on the xy-axis plane, the object types include but are not limited to park green space, construction land, buildings, water bodies, bare land, roads, etc., select any one of the characteristic curves of the spectral index from the characteristic curves of the spectral index, obtain the z-axis value corresponding to the xy-axis mapping coordinate sets of the characteristic curves of the different object types, compare the z-axis values ​​corresponding to the xy-axis mapping coordinate sets of the different object types in pairs, obtain the z-axis mean square error value corresponding to each object type and other different object types, and normalize the z-axis mean square error values ​​corresponding to the spectral indexes between the different object types to convert them into discrimination coefficients. The threshold range of the discrimination coefficient is [0, 1]. The larger the difference in the z-axis values, the larger the discrimination coefficient. By analogy, obtain the discrimination coefficients corresponding to the several spectral indexes between the different object types. Similarly, obtain the discrimination coefficients corresponding to the several texture feature values ​​between the different object types.

[0034] A discrimination coefficient threshold is preset, and the discrimination coefficients of a number of spectral indices between different types of land objects and the discrimination coefficients of a number of texture eigenvalues ​​between different types of land objects are compared with the discrimination coefficient threshold, and the spectral indices or texture eigenvalues ​​whose discrimination coefficients are less than the discrimination coefficient threshold are judged and screened out, and a set of xy-axis mapping coordinate sets of different types of land objects corresponding to the spectral indices or texture eigenvalues ​​are obtained, and the areas corresponding to the set of xy-axis mapping coordinate sets of different types of land objects in the three-dimensional coordinate system are marked as a set of confusing areas.

[0035] Furthermore, the threshold determination module obtains a determination threshold set of the ground object type in all regions according to the characteristic curve of the spectral feature set and the characteristic curve of the texture feature set in the three-dimensional coordinate system, and the process includes:

[0036] The xy-axis mapping coordinate sets of the same land feature type in all regions in the three-dimensional coordinate system are divided into a group, and the z-axis value ranges of the characteristic curves of several spectral indices and the z-axis value ranges of the characteristic curves of several texture eigenvalues ​​corresponding to the xy-axis mapping coordinate sets of any group of land feature types in the three-dimensional coordinate system are obtained. The z-axis value ranges of the characteristic curves of the several spectral indices and the z-axis value ranges of the characteristic curves of the several texture eigenvalues ​​are used as the judgment threshold set of the group of land feature types, and so on, to obtain the judgment threshold sets of different groups of land feature types in the three-dimensional coordinate system.

[0037] Furthermore, the threshold determination module obtains an additional determination threshold set for the confusion region according to the discrimination coefficients of the spectral indexes and the texture feature values ​​in the confusion region, and the process includes:

[0038] Determine whether there is a spectral index or texture feature value with a discrimination coefficient greater than or equal to a discrimination coefficient threshold in a set of confusion regions; if there is a spectral index or texture feature value with a discrimination coefficient greater than or equal to the discrimination coefficient threshold in the confusion region, then use the z-axis value range corresponding to the spectral index or texture feature value with a discrimination coefficient greater than or equal to the discrimination coefficient threshold in the confusion region in the three-dimensional coordinate system as an additional determination threshold set for the confusion region;

[0039] If there is no spectral index or texture feature value whose discriminant coefficient is greater than or equal to the discriminant coefficient threshold in the confusion region, feature addition judgment processing is performed on the confusion region.

[0040] Furthermore, the confused area is subjected to feature addition judgment processing, and the process includes:

[0041] According to the xy-axis mapping coordinate sets corresponding to different types of land objects in the target city on the xy-axis plane, the three-dimensional coordinate system is segmented into regions, and the regions corresponding to each segmented type of land object are used as regional nodes. Shape features of each regional node are extracted according to high-resolution remote sensing data, and the geometric shape features of each regional node are extracted. The geometric shape features of the regional nodes in the confused area are compared with the geometric shape features of other regional nodes for similarity, and the shape similarity coefficients of the regional nodes in the confused area and other regional nodes are obtained;

[0042] Obtain the two-dimensional coordinates of the geometric center and the regional coverage area of ​​each regional node, obtain the Euclidean distance between the regional node of the confusion area and other regional nodes according to the two-dimensional coordinates of the geometric center of each regional node, build an influence index model, input the Euclidean distance of other regional nodes and the regional coverage area of ​​other regional nodes into the influence index model, and obtain the influence coefficient of other regional nodes on the regional node of the confusion area;

[0043] According to the shape similarity coefficient of the regional node of the confused area and other regional nodes and the influence coefficient of other regional nodes on the regional nodes of the confused area, the importance coefficient of other regional nodes is obtained, the regional node with the largest importance coefficient among other regional nodes is screened out, and the judgment threshold set of the regional nodes is used as the additional judgment threshold set of the confused area.

[0044] Furthermore, the process of extracting shape features of each regional node based on high-resolution remote sensing data is as follows:

[0045] Perform edge detection on the remote sensing image to identify the boundary features of the target. The edge detection algorithm used in the present invention includes but is not limited to Sobel, Canny, etc. The external shape information of the target can be obtained by detecting the boundary;

[0046] Fit the detected boundaries and use simple geometric shapes such as rectangles and circles to approximate the actual shape. For example, the minimum bounding box can be used to fit a rectangular shape, or the least squares method can be used to fit a straight line, circle, etc.

[0047] Extract the contour of the target, obtain the shape features of the target, and use the contour extraction algorithm, such as the findContours function in OpenCV, to extract contour information on the boundary of the target;

[0048] For regular convex shapes such as rectangles, the convex hull detection algorithm is used to enclose the target in the smallest convex polygon. For water body features, edge detection and shape fitting methods are combined to extract the smooth curve features of the water body, and the characteristics of the smooth curve are used to distinguish water bodies from other landforms.

[0049] For the shape features of each region node, the minimum circumscribed rectangle and the minimum circumscribed circle of the shape features of each region node are obtained, and the minimum circumscribed rectangle and the minimum circumscribed circle are used as similarity indicators. The shape similarity of two regions is measured by comparing the similarity indicators.

[0050] Furthermore, the calculation formula of the influence index model is:

[0051]

[0052] Among them, Q i represents the influence coefficient of the i-th other regional node on the regional node of the confusion region, S i represents the area coverage of the i-th other regional node, d i represents the Euclidean distance between the i-th other region node pair and the region node of the confusion region;

[0053] Then, weighted summation is performed on the shape similarity coefficients of the regional node in the confusion region and other regional nodes and the influence coefficients of other regional nodes on the regional node in the confusion region to obtain the importance coefficients of other regional nodes on the regional node in the confusion region.

[0054] Furthermore, the feature type extraction module constructs a feature type extraction model based on a convolutional neural network, uses a set of determination thresholds for feature types in different regions of a target city in multiple time periods and an additional set of determination thresholds as training sets and test sets, and obtains the feature types of each region of the target city at the current moment based on the high-resolution remote sensing data of the target city at the current moment and the feature type extraction model, and the process includes:

[0055] Obtaining a set of determination thresholds and an additional set of determination thresholds for the types of objects in different areas of a target city in multiple time periods;

[0056] A land feature type extraction model is constructed based on a convolutional neural network, and the land feature type determination threshold sets and additional determination threshold sets in different regions of the target city in multiple time periods are used as training sets and test sets.

[0057] Input the training set into the land feature type extraction model for training until the loss function training is stable, and save the model parameters, test the land feature type extraction model through the test set until it meets the preset requirements, and output the land feature type extraction model;

[0058] The high-resolution remote sensing data of the target city collected by the high-resolution remote sensing satellite at the current moment is obtained and input into the ground feature type extraction model, and the ground feature type of each area of ​​the target city at the current moment is obtained according to the ground feature type extraction model.

[0059] Compared with the prior art, the beneficial effects of the present invention are as follows: marking the types of objects in different areas in a three-dimensional coordinate system, and obtaining the corresponding discrimination coefficients between a number of spectral indices and a number of texture eigenvalues ​​between different types of objects based on the characteristic curve, and judging whether there is a confused area in the three-dimensional coordinate system based on the discrimination coefficient; if there is a confused area and the object type of the confused area cannot be judged by the spectral index or the texture eigenvalue, the confused area is subjected to feature addition judgment processing to obtain an additional judgment threshold set for the confused area, and finally the judgment threshold sets and additional judgment threshold sets of the object types in different areas of a target city in multiple time periods are used as training sets and test sets to construct an object type extraction module, and the object type judgment threshold sets of different areas of the target city are dynamically adjusted by the object type extraction module, thereby significantly improving the accuracy of object classification and target recognition. DETAILED DESCRIPTION

[0060] The following is a clear and complete description of the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without making any creative work are within the scope of protection of the present application.

[0061] A high-resolution remote sensing monitoring system for smart cities, comprising a monitoring center, wherein the monitoring center is communicatively connected to a data acquisition module, a data processing module, a region matching module, a threshold determination module, and a ground object type extraction module;

[0062] The data acquisition module is used to collect high-resolution remote sensing monitoring data of the target city and obtain spectral feature sets and texture feature sets of different areas of the target city;

[0063] The data processing module is used to construct a three-dimensional coordinate system and display characteristic curves of the spectral feature set and the texture feature set of different areas of the target city on the three-dimensional coordinate system;

[0064] The region matching module is used to mark the types of objects in different regions in the three-dimensional coordinate system and determine whether there is a confusing region in the three-dimensional coordinate system based on the characteristic curve;

[0065] The threshold determination module is used to obtain a determination threshold set of the ground object type in all regions and an additional determination threshold set of the confusion region according to the characteristic curve;

[0066] The ground feature type extraction module is used to construct a ground feature type extraction model, and obtain the ground feature type of each area of ​​the target city at the current moment according to the high-resolution remote sensing data of the target city at the current moment and the ground feature type extraction model.

[0067] It should be further explained that, in the specific implementation process, the data acquisition module collects high-resolution remote sensing monitoring data of the target city and obtains spectral feature sets and texture feature sets of different areas of the target city, and the process includes:

[0068] The high-resolution remote sensing data of the target city collected by the high-resolution remote sensing satellite is obtained, and the spectral reflectance data of each band in the relevant band combination of surface reflectance of different areas of the target city is obtained according to the high-resolution remote sensing data, and a number of spectral indexes of different areas of the target city are obtained according to the spectral reflectance data bands, and a spectral feature set of different areas of the target city is constructed according to the spectral index data, and a grayscale co-occurrence matrix of the high-resolution remote sensing data is constructed at the same time, and a number of texture eigenvalues ​​in the grayscale co-occurrence matrix are extracted, and a texture feature set of different areas of the target city is constructed according to the several texture eigenvalues ​​in the grayscale co-occurrence matrix.

[0069] It should be further explained that, in the specific implementation process, the spectral reflectance data of each band in the relevant band combination of the target city surface reflectance mainly includes red, green and blue bands, which are usually used to obtain the real color information and morphological characteristics of the surface;

[0070] in:

[0071] Infrared band: including near-infrared band and far-infrared band, used to monitor vegetation health, land cover changes, etc.;

[0072] Short-wave infrared band: including short-wave infrared 1 band and short-wave infrared 2 band, used to detect vegetation moisture content, soil characteristics, etc.;

[0073] Microwave band: used to penetrate clouds and vegetation to obtain information such as surface topography and soil moisture;

[0074] Several spectral indices obtained by band calculation include but are not limited to: Normalized Difference Vegetation Index NDVI, Normalized Difference Water Index NDWI, where:

[0075]

[0076] In the formula, NIR represents the reflectance value of the near-infrared band, Red represents the reflectance value of the red band, and the NDVI value range is between -1 and 1, where negative values ​​indicate non-vegetation coverage (such as water bodies or bare soil), values ​​close to 0 indicate sparse vegetation coverage, and values ​​close to 1 indicate dense vegetation coverage;

[0077]

[0078] In this formula, NIR represents the reflectance value of the near-infrared band, and Green represents the reflectance value of the green band;

[0079] It should be further explained that, in the specific implementation process, the process of constructing the texture feature set of the target city includes:

[0080] Using the acquired high-resolution remote sensing data, the grayscale co-occurrence matrix around each pixel is calculated by sliding the window. The specific details include quantizing the grayscale level of the remote sensing image into discrete grayscale levels, dividing the grayscale value range into several levels, for example, 8-bit images can be divided into 16, 32, and 64 levels, and defining the parameters required for the grayscale co-occurrence matrix, including distance (d) and direction (θ), which are determined according to actual application requirements. Then, based on the calculated grayscale co-occurrence matrix, a series of texture feature values ​​are extracted to construct a texture feature set. The texture features in the texture feature set include but are not limited to:

[0081] Contrast: describes the statistical characteristics of the contrast between pixels of different gray levels in an image;

[0082] Correlation: describes the degree of correlation between pixels of different gray levels in an image;

[0083] Energy: describes the uniformity of the grayscale distribution of pixels in an image;

[0084] Homogeneity: describes the similarity of the gray levels of adjacent pixels in an image;

[0085] Entropy: describes the degree of uncertainty of image texture.

[0086] It should be further explained that, in the specific implementation process, the data processing module constructs a three-dimensional coordinate system and displays the characteristic curves of the spectral feature set and the characteristic curves of the texture feature set of different areas of the target city on the three-dimensional coordinate system, and the process includes:

[0087] A three-dimensional coordinate system is established, a city map of the target city is obtained, and the city map of the target city is mapped to the xy-axis plane in the three-dimensional coordinate system, and the xy-axis mapping coordinates of the spectral feature sets and texture feature sets of different areas of the target city on the GIS surface plan of the target city are obtained, and according to the xy-axis mapping coordinates, a number of spectral indices in the spectral feature set and a number of texture feature values ​​in the texture feature set are input into the z-axis corresponding to the xy-axis mapping coordinates, and then the z-axis values ​​corresponding to the several spectral indices in the spectral feature set on different xy-axis mapping coordinates are plotted into a curve to obtain the characteristic curves of several spectral indices of the target city. Similarly, the characteristic curves of several texture feature values ​​of the target city are obtained.

[0088] It should be further explained that, in the specific implementation process, the region matching module labels the types of objects in different regions in the three-dimensional coordinate system, obtains the discrimination coefficients corresponding to a number of spectral indices and a number of texture feature values ​​between different types of objects according to the characteristic curve, and judges whether there is a confusing area in the three-dimensional coordinate system according to the discrimination coefficient. The process includes:

[0089] Annotate the xy-axis plane in the three-dimensional coordinate system with the type of the object, obtain the xy-axis mapping coordinate sets corresponding to the different object types in the target city on the xy-axis plane, the object types include but are not limited to park green space, construction land, buildings, water bodies, bare land, roads, etc., select any one of the characteristic curves of the spectral index from the characteristic curves of the spectral index, obtain the z-axis value corresponding to the xy-axis mapping coordinate sets of the characteristic curves of the different object types, compare the z-axis values ​​corresponding to the xy-axis mapping coordinate sets of the different object types in pairs, obtain the z-axis mean square error value corresponding to each object type and other different object types, and normalize the z-axis mean square error values ​​corresponding to the spectral indexes between the different object types to convert them into discrimination coefficients. The threshold range of the discrimination coefficient is [0, 1]. The larger the difference in the z-axis values, the larger the discrimination coefficient. By analogy, obtain the discrimination coefficients corresponding to the several spectral indexes between the different object types. Similarly, obtain the discrimination coefficients corresponding to the several texture feature values ​​between the different object types.

[0090] A discrimination coefficient threshold is preset, and the discrimination coefficients of a number of spectral indices between different types of land objects and the discrimination coefficients of a number of texture eigenvalues ​​between different types of land objects are compared with the discrimination coefficient threshold, and the spectral indices or texture eigenvalues ​​whose discrimination coefficients are less than the discrimination coefficient threshold are judged and screened out, and a set of xy-axis mapping coordinate sets of different types of land objects corresponding to the spectral indices or texture eigenvalues ​​are obtained, and the areas corresponding to the set of xy-axis mapping coordinate sets of different types of land objects in the three-dimensional coordinate system are marked as a set of confusing areas.

[0091] It should be further explained that the significance of obtaining the discrimination coefficients of several spectral indices and several texture feature values ​​between different types of land features is:

[0092] When processing high-resolution remote sensing data, it is necessary to perform object classification and target recognition, such as building recognition, road extraction and other tasks. However, the urban environment is complex and changeable, and there are occlusions and similarities between objects, which pose challenges to object classification and target recognition. In addition, there are various interference sources in the urban environment, such as weather, clouds, and light. These interferences will affect the acquisition and processing of remote sensing data and reduce the accuracy of monitoring results. Therefore, for the high-resolution remote sensing data of the same target city collected by high-resolution remote sensing satellites at different times, there will be different data analysis results under different interference source conditions. For example, under strong sunlight, water bodies and buildings may have similar spectral characteristics in remote sensing images. For example, the surfaces of water bodies and buildings may reflect strong light, however , water bodies usually present relatively smooth texture features, while buildings present complex structures and textures. There are obvious differences in texture features between the two, that is, spectral features interfere with the discrimination of the types of the two objects. Similarly, in the process of regional type classification based on spectral and texture features, for example, on rainy days, the characteristic curves of green land and construction land in several texture feature indexes may be similar, and it is difficult to distinguish the two through their texture features, that is, texture features interfere with the discrimination of the types of the two objects, but the difference in spectral index is significant, and the two can be distinguished through spectral features. On the other hand, if the buildings and roads are highly similar in spectral and texture features and are difficult to distinguish, other features need to be added to distinguish them.

[0093] It should be further explained that, in the specific implementation process, the threshold determination module obtains the determination threshold set of the ground object type in all regions according to the characteristic curve of the spectral feature set and the characteristic curve of the texture feature set in the three-dimensional coordinate system, and the process includes:

[0094] The xy-axis mapping coordinate sets of the same land feature type in all regions in the three-dimensional coordinate system are divided into a group, and the z-axis value ranges of the characteristic curves of several spectral indices and the z-axis value ranges of the characteristic curves of several texture eigenvalues ​​corresponding to the xy-axis mapping coordinate sets of any group of land feature types in the three-dimensional coordinate system are obtained. The z-axis value ranges of the characteristic curves of the several spectral indices and the z-axis value ranges of the characteristic curves of the several texture eigenvalues ​​are used as the judgment threshold set of the group of land feature types, and so on, to obtain the judgment threshold sets of different groups of land feature types in the three-dimensional coordinate system.

[0095] It should be further explained that, in a specific implementation process, the threshold determination module obtains an additional determination threshold set for the confusion region according to the discrimination coefficients of several spectral indices and several texture feature values ​​in the confusion region, and the process includes:

[0096] Determine whether there is a spectral index or texture feature value with a discrimination coefficient greater than or equal to a discrimination coefficient threshold in a set of confusion regions; if there is a spectral index or texture feature value with a discrimination coefficient greater than or equal to the discrimination coefficient threshold in the confusion region, then use the z-axis value range corresponding to the spectral index or texture feature value with a discrimination coefficient greater than or equal to the discrimination coefficient threshold in the confusion region in the three-dimensional coordinate system as an additional determination threshold set for the confusion region;

[0097] If there is no spectral index or texture feature value whose discriminant coefficient is greater than or equal to the discriminant coefficient threshold in the confusion region, feature addition judgment processing is performed on the confusion region.

[0098] It should be further explained that, in the specific implementation process, the feature addition judgment process of the confusion area includes:

[0099] According to the xy-axis mapping coordinate sets corresponding to different types of land objects in the target city on the xy-axis plane, the three-dimensional coordinate system is segmented into regions, and the regions corresponding to each segmented type of land object are used as regional nodes. Shape features of each regional node are extracted according to high-resolution remote sensing data, and the geometric shape features of each regional node are extracted. The geometric shape features of the regional nodes in the confused area are compared with the geometric shape features of other regional nodes for similarity, and the shape similarity coefficients of the regional nodes in the confused area and other regional nodes are obtained;

[0100] Obtain the two-dimensional coordinates of the geometric center and the regional coverage area of ​​each regional node, obtain the Euclidean distance between the regional node of the confusion area and other regional nodes according to the two-dimensional coordinates of the geometric center of each regional node, build an influence index model, input the Euclidean distance of other regional nodes and the regional coverage area of ​​other regional nodes into the influence index model, and obtain the influence coefficient of other regional nodes on the regional node of the confusion area;

[0101] According to the shape similarity coefficient of the regional node of the confused area and other regional nodes and the influence coefficient of other regional nodes on the regional nodes of the confused area, the importance coefficient of other regional nodes is obtained, the regional node with the largest importance coefficient among other regional nodes is screened out, and the judgment threshold set of the regional nodes is used as the additional judgment threshold set of the confused area.

[0102] It should be further explained that, in the specific implementation process, the process of extracting shape features of each regional node based on high-resolution remote sensing data is as follows:

[0103] Perform edge detection on the remote sensing image to identify the boundary features of the target. The edge detection algorithm used in the present invention includes but is not limited to Sobel, Canny, etc. The external shape information of the target can be obtained by detecting the boundary;

[0104] Fit the detected boundaries and use simple geometric shapes such as rectangles and circles to approximate the actual shape. For example, the minimum bounding box can be used to fit a rectangular shape, or the least squares method can be used to fit a straight line, circle, etc.

[0105] Extract the contour of the target, obtain the shape features of the target, and use the contour extraction algorithm, such as the findContours function in OpenCV, to extract contour information on the boundary of the target;

[0106] For regular convex shapes such as rectangles, the convex hull detection algorithm is used to enclose the target in the smallest convex polygon. For water body features, edge detection and shape fitting methods are combined to extract the smooth curve features of the water body, and the characteristics of the smooth curve are used to distinguish water bodies from other landforms.

[0107] For the shape features of each region node, the minimum circumscribed rectangle and the minimum circumscribed circle of the shape features of each region node are obtained, and the minimum circumscribed rectangle and the minimum circumscribed circle are used as similarity indicators. The shape similarity of two regions is measured by comparing the similarity indicators.

[0108] It should be further explained that, in the specific implementation process, the calculation formula of the influence index model is:

[0109] Q i =S i 1-di ;

[0110] Among them, Q i represents the influence coefficient of the i-th other regional node on the regional node of the confusion region, S i represents the area coverage of the i-th other regional node, d i represents the Euclidean distance between the i-th other region node pair and the region node of the confusion region;

[0111] Then, weighted summation is performed on the shape similarity coefficients of the regional node in the confusion region and other regional nodes and the influence coefficients of other regional nodes on the regional node in the confusion region to obtain the importance coefficients of other regional nodes on the regional node in the confusion region.

[0112] It should be further explained that, in the specific implementation process, the land feature type extraction module constructs a land feature type extraction model based on a convolutional neural network, uses the determination threshold set of land feature types in different areas of the target city in multiple time periods and the additional determination threshold set as a training set and a test set, and obtains the land feature type of each area of ​​the target city at the current moment according to the high-resolution remote sensing data of the target city at the current moment and the land feature type extraction model, and the process includes:

[0113] Obtaining a set of determination thresholds and an additional set of determination thresholds for the types of objects in different areas of a target city in multiple time periods;

[0114] A land feature type extraction model is constructed based on a convolutional neural network, and the land feature type determination threshold sets and additional determination threshold sets in different regions of the target city in multiple time periods are used as training sets and test sets.

[0115] Input the training set into the land feature type extraction model for training until the loss function training is stable, and save the model parameters, test the land feature type extraction model through the test set until it meets the preset requirements, and output the land feature type extraction model;

[0116] The high-resolution remote sensing data of the target city collected by the high-resolution remote sensing satellite at the current moment is obtained and input into the ground feature type extraction model, and the ground feature type of each area of ​​the target city at the current moment is obtained according to the ground feature type extraction model.

[0117] By comparing the land feature types of various areas of the target city at different times generated by the land feature type extraction model, the areas where the land feature types of the target city have changed are obtained.

[0118] The above embodiments are only used to illustrate the technical method of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present invention.

Claims

1. A high-resolution remote sensing monitoring system for smart cities, comprising a monitoring center, characterized in that: The monitoring center is communicatively connected to a data acquisition module, a data processing module, a region matching module, a threshold determination module and a ground feature type extraction module; The data acquisition module is used to collect high-resolution remote sensing monitoring data of the target city and obtain spectral feature sets and texture feature sets of different areas of the target city; The data processing module is used to construct a three-dimensional coordinate system and display characteristic curves of the spectral feature set and the texture feature set of different areas of the target city on the three-dimensional coordinate system; The region matching module is used to mark the types of objects in different regions in the three-dimensional coordinate system and determine whether there is a confusing region in the three-dimensional coordinate system according to the characteristic curve, including: Annotate the xy-axis plane in the three-dimensional coordinate system with the type of land object, obtain the xy-axis mapping coordinate set corresponding to the different land object types in the target city on the xy-axis plane, select any characteristic curve of a spectral index from the characteristic curves of a plurality of spectral indexes, obtain the z-axis value corresponding to the xy-axis mapping coordinate set of the characteristic curve in the different land object types, compare the z-axis values ​​corresponding to the xy-axis mapping coordinate sets of each different land object type in pairs, obtain the z-axis mean square error value corresponding to each land object type and other different land object types, and normalize the z-axis mean square error values ​​corresponding to the spectral index between each different land object type and convert them into a discrimination coefficient. The larger the z-axis value difference is, the larger the discrimination coefficient is. By analogy, obtain the discrimination coefficients corresponding to the plurality of spectral indexes between each different land object type. Similarly, obtain the discrimination coefficients corresponding to the plurality of texture feature values ​​between each different land object type. Preset a discrimination coefficient threshold, compare the discrimination coefficients of a plurality of spectral indices between different types of land objects and the discrimination coefficients of a plurality of texture feature values ​​between different types of land objects with the discrimination coefficient threshold, judge and screen out the spectral indices or texture feature values ​​whose discrimination coefficients are less than the discrimination coefficient threshold, obtain a set of xy-axis mapping coordinate sets of different types of land objects corresponding to the spectral indices or texture feature values, and mark the areas corresponding to the set of xy-axis mapping coordinate sets of different types of land objects in the three-dimensional coordinate system as a set of confusion areas; The threshold determination module is used to obtain a determination threshold set of the ground object type in all regions and an additional determination threshold set of the confusion region according to the characteristic curve; The ground object type extraction module is used to construct a ground object type extraction model, and obtain the ground object type of each area of ​​the target city at the current moment according to the high-resolution remote sensing data of the target city at the current moment and the ground object type extraction model, including: Obtaining a set of determination thresholds and an additional set of determination thresholds for the types of objects in different areas of a target city in multiple time periods; A land feature type extraction model is constructed based on a convolutional neural network, and the judgment threshold sets of land feature types in different areas of the target city in multiple time periods and additional judgment threshold sets are used as training sets and test sets; Input the training set into the land feature type extraction model for training until the loss function training is stable, and save the model parameters, test the land feature type extraction model through the test set until it meets the preset requirements, and output the land feature type extraction model; The high-resolution remote sensing data of the target city collected by the high-resolution remote sensing satellite at the current moment is obtained and input into the ground feature type extraction model, and the ground feature type of each area of ​​the target city at the current moment is obtained according to the ground feature type extraction model.

2. A high-resolution remote sensing monitoring system for smart cities according to claim 1, characterized in that: The data acquisition module collects high-resolution remote sensing monitoring data of the target city and obtains spectral feature sets and texture feature sets of different areas of the target city. The process includes: The high-resolution remote sensing data of the target city collected by the high-resolution remote sensing satellite is obtained, and the spectral reflectance data of each band in the relevant band combination of surface reflectance of different areas of the target city is obtained according to the high-resolution remote sensing data, and a number of spectral indexes of different areas of the target city are obtained according to the spectral reflectance data bands, and a spectral feature set of different areas of the target city is constructed according to the spectral index data, and a grayscale co-occurrence matrix of the high-resolution remote sensing data is constructed at the same time, and a number of texture eigenvalues ​​in the grayscale co-occurrence matrix are extracted, and a texture feature set of different areas of the target city is constructed according to the several texture eigenvalues ​​in the grayscale co-occurrence matrix.

3. A high-resolution remote sensing monitoring system for smart cities according to claim 2, characterized in that: The data processing module constructs a three-dimensional coordinate system and displays characteristic curves of spectral feature sets and texture feature sets of different regions of the target city on the three-dimensional coordinate system. The process includes: A three-dimensional coordinate system is established, a city map of the target city is obtained, and the city map of the target city is mapped to the xy-axis plane in the three-dimensional coordinate system, and the xy-axis mapping coordinates of the spectral feature sets and texture feature sets of different areas of the target city on the GIS surface plan of the target city are obtained, and according to the xy-axis mapping coordinates, a number of spectral indices in the spectral feature set and a number of texture feature values ​​in the texture feature set are input into the z-axis corresponding to the xy-axis mapping coordinates, and then the z-axis values ​​corresponding to the several spectral indices in the spectral feature set on different xy-axis mapping coordinates are plotted into a curve to obtain the characteristic curves of several spectral indices of the target city. Similarly, the characteristic curves of several texture feature values ​​of the target city are obtained.

4. The high-resolution remote sensing monitoring system for smart cities according to claim 3 is characterized in that: The threshold determination module obtains a determination threshold set of the ground object type in all regions according to the characteristic curve, and the process includes: The xy-axis mapping coordinate sets of the same land feature type in all regions in the three-dimensional coordinate system are divided into a group, and the z-axis value ranges of the characteristic curves of several spectral indices and the z-axis value ranges of the characteristic curves of several texture eigenvalues ​​corresponding to the xy-axis mapping coordinate sets of any group of land feature types in the three-dimensional coordinate system are obtained. The z-axis value ranges of the characteristic curves of the several spectral indices and the z-axis value ranges of the characteristic curves of the several texture eigenvalues ​​are used as the judgment threshold set of the group of land feature types, and so on, to obtain the judgment threshold sets of different groups of land feature types in the three-dimensional coordinate system.

5. A high-resolution remote sensing monitoring system for smart cities according to claim 4, characterized in that: The threshold determination module obtains an additional determination threshold set for the confusion area according to the characteristic curve, and the process includes: Determine whether there is a spectral index or texture feature value with a discrimination coefficient greater than or equal to a discrimination coefficient threshold in a set of confusion regions; if there is a spectral index or texture feature value with a discrimination coefficient greater than or equal to the discrimination coefficient threshold in the confusion region, then use the z-axis value range corresponding to the spectral index or texture feature value with a discrimination coefficient greater than or equal to the discrimination coefficient threshold in the confusion region in the three-dimensional coordinate system as an additional determination threshold set for the confusion region; If there is no spectral index or texture feature value whose discriminant coefficient is greater than or equal to the discriminant coefficient threshold in the confusion region, feature addition judgment processing is performed on the confusion region.

6. A high-resolution remote sensing monitoring system for smart cities according to claim 5, characterized in that: The confusion area is processed for feature addition and judgment, and the process includes: According to the xy-axis mapping coordinate sets corresponding to different types of land objects in the target city on the xy-axis plane, the three-dimensional coordinate system is segmented into regions, and the regions corresponding to each segmented type of land object are used as regional nodes. Shape features of each regional node are extracted according to high-resolution remote sensing data, and the geometric shape features of each regional node are extracted. The geometric shape features of the regional nodes in the confused area are compared with the geometric shape features of other regional nodes for similarity, and the shape similarity coefficients of the regional nodes in the confused area and other regional nodes are obtained; Obtain the two-dimensional coordinates of the geometric center and the regional coverage area of ​​each regional node, obtain the Euclidean distance between the regional node of the confusion area and other regional nodes according to the two-dimensional coordinates of the geometric center of each regional node, build an influence index model, input the Euclidean distance of other regional nodes and the regional coverage area of ​​other regional nodes into the influence index model, and obtain the influence coefficient of other regional nodes on the regional node of the confusion area; According to the shape similarity coefficient of the regional node of the confused area and other regional nodes and the influence coefficient of other regional nodes on the regional nodes of the confused area, the importance coefficient of other regional nodes is obtained, the regional node with the largest importance coefficient among other regional nodes is screened out, and the judgment threshold set of the regional nodes is used as the additional judgment threshold set of the confused area.

Images