Land survey quality monitoring system for remote sensing image intelligent segmentation and ground object identification

Abstract

The application discloses a land survey quality monitoring system for remote sensing image intelligent segmentation and ground object identification, relates to the technical field of remote sensing, and comprises an image identification module, a label comparison module and a monitoring module. Specifically, remote sensing images of land to be monitored are acquired through a remote sensing sensor, image regions are divided through structural complexity analysis, and the category probability vectors of the pixel positions of each region in the formed non-uniform region set are analyzed and acquired to construct a label map of the whole remote sensing image. Pixel-by-pixel category comparison is performed on the current label map and a reference label map to form a change mask map. Structural change regions are identified according to the change mask map. Through time sequence analysis, it is determined whether the corresponding candidate label inconsistent regions under different time phases are still structural change regions to acquire a judgment result. Based on the judgment result, an electronic map is updated.

Description

Technical Field

[0001] This invention relates to the field of remote sensing technology, specifically a land survey quality monitoring system for intelligent segmentation and feature identification of remote sensing images. Background Technology

[0002] In the field of land surveys, to accurately acquire and update information on the distribution of land features, intelligent processing of remote sensing imagery is necessary to identify land cover labels such as cultivated land, forest land, water bodies, and construction land, and to determine whether land features have changed and whether the types of changes are accurate and reliable. Therefore, given the complex distribution of land features, significant texture differences, and high uncertainty in model classification within remote sensing imagery, there is an urgent need for a comprehensive monitoring system capable of adapting to regional structural characteristics, assessing label reliability, and interpreting temporal changes.

[0003] Currently, in land surveys and feature identification, conventional methods rely heavily on static remote sensing imagery and classification algorithms based on fixed rules. These methods struggle to adapt to the varying structural complexities of feature distribution, leading to a significant decrease in identification accuracy for complex features such as suburban transition zones and mixed-use land parcels. Secondly, the lack of intelligent identification mechanisms for temporary occupancy, relying solely on category differences to judge changes, easily results in misjudgments or omissions. Summary of the Invention

[0004] In view of the above-mentioned problems existing in the prior art, this application provides a land survey quality monitoring system for intelligent segmentation and land feature identification of remote sensing images.

[0005] This disclosure provides a land survey quality monitoring system for intelligent segmentation and feature identification of remote sensing images, including:

[0006] The image recognition module acquires remote sensing images of the land to be monitored through remote sensing sensors. After structural complexity analysis, it divides the image regions to form a non-homogeneous region set. It then analyzes and obtains the category probability vector of the pixel position in each region within the non-homogeneous region set to construct a label map for the entire remote sensing image.

[0007] The label comparison module performs pixel-by-pixel category comparison between the label image under the current time conditions and the reference label image to form a change mask image. Based on the change mask image, structural change regions are identified, and these structural change regions are used to determine the inconsistency of the land cover category labels.

[0008] The monitoring module uses time-series analysis to determine whether areas with inconsistent candidate labels under different time phases are still areas of structural change, in order to obtain judgment results. Based on the judgment results, the electronic map is updated. The judgment results include non-temporary signs and temporary signs.

[0009] Optionally, the image recognition module includes an acquisition unit, a structural analysis unit, a segmentation scale unit, and a label generation unit;

[0010] The acquisition unit acquires remote sensing images of the land to be monitored through remote sensing sensors and preprocesses the remote sensing images, including standardization processing such as radiometric correction, atmospheric correction and geometric correction.

[0011] The structural analysis unit divides the remote sensing image into regions to obtain multiple regularized initial raster regions and analyzes the structural complexity of each initial raster region. By directionally weighting the structural change intensity of each pixel in each initial raster region and averaging the results, a structural complexity index is obtained. The structural complexity index is used to reflect the degree of local structural disturbance in each initial raster region.

[0012] Optionally, when dividing the scale unit, if the structural complexity index exceeds a preset complexity threshold, it indicates that the corresponding initial raster region has a complex structure, and a first-class region processing mechanism is output. If the structural complexity index does not exceed the preset complexity threshold, it indicates that the corresponding initial raster region has a stable structure, and a second-class region processing mechanism is output. If a first-class region processing mechanism is output, the corresponding initial raster region is further divided. If a second-class region processing mechanism is output, the corresponding initial raster region is merged with initial raster regions of the same type and adjacent positions.

[0013] Specifically, the initial raster region segmentation scale is determined based on two types of region processing mechanisms to form a non-uniform region set. The segmentation path for each region within this set is then determined. If a type of region processing mechanism is output, the corresponding initial raster region is further subdivided, including:

[0014] A minimum granularity limit is preset, and the structural complexity index of the corresponding initial raster region is determined to exceed a preset complexity threshold. Based on the structural complexity index being inversely proportional to the block size, the granularity of the corresponding initial raster region is set, specifically as follows: Among them, S R Let S be the granularity of the initial raster region. min To define the minimum granularity limit, k is the scaling factor. This is used to dynamically adjust the granularity of each initial raster region.

[0015] Optionally, determining the segmentation path for the non-homogeneous region by grouping it into different regions includes,

[0016] Based on the non-homogeneous region set, distinguish between regions obtained through a type-one region processing mechanism and regions obtained through a type-two region processing mechanism within the non-homogeneous region set;

[0017] Regions obtained through the first type of region processing mechanism are processed using a deep attention residual path, and the corresponding segmentation network structure is a deep attention residual network. Regions obtained through the second type of region processing mechanism are processed using a standard semantic segmentation path, and the corresponding segmentation network structure is the standard fully convolutional semantic segmentation model of U-Net.

[0018] Optionally, the label generation unit obtains the category probability vector of each pixel location based on the segmentation network structure to construct a label map for the entire remote sensing image. Specific steps include:

[0019] The remote sensing images corresponding to each region in the non-homogeneous region set are input into the corresponding segmentation network structure. Spatial and semantic information are extracted through the convolutional feature extraction layer to output the feature map. The feature map is then upsampled using the deconvolution method. Each pixel in the upsampled feature map has a multi-dimensional vector.

[0020] The multidimensional vector is mapped to an output vector of length C through a fully connected mapping layer, where C is the total number of land cover categories;

[0021] The output vector is normalized using the Softmax function to obtain the class probability vector;

[0022] The land cover category with the highest probability is selected from the category probability vector as the final land cover category label;

[0023] Traverse each pixel location and reassemble it into the original image according to the region numbers of the non-uniform region set to construct the label map of the entire remote sensing image.

[0024] Optionally, the label comparison module includes pixel-level label difference units and recognition units;

[0025] The pixel-level label difference unit performs spatial resampling and geometric registration on the current label image and the reference label image, and performs pixel-by-pixel category comparison on the label image and the reference label image under the current time condition. If the label is different at the same pixel position, it is recorded as a changed pixel and a change mask image is formed.

[0026] The identification unit clusters adjacent changing pixels in the change mask to extract connected regions as candidate label inconsistency regions. For each candidate label inconsistency region, the ratio of changing pixels within it to the number of pixels in the corresponding clustered region is used to obtain the changing pixel ratio, which is used to identify structurally changing regions.

[0027] Optionally, the monitoring module includes a timing analysis unit and a feedback unit;

[0028] If the proportion of changing pixels exceeds a preset proportion threshold, the temporal analysis unit marks the corresponding candidate label inconsistency region as a structural change region under the current temporal condition; otherwise, no marking is performed. If the corresponding candidate label inconsistency region is marked as a structural change region under the current temporal condition, the temporal analysis is used to determine whether the candidate label inconsistency region under the current temporal condition is still a structural change region under other temporal conditions, so as to obtain the judgment result.

[0029] Optionally, the feedback unit, if the candidate label inconsistency area under the current time phase condition is still a structural change area under other time phase conditions, will use the judgment result as a non-temporary label; otherwise, it will use a temporary label. Based on the judgment result, a structured attribute record table will be generated for the corresponding area in the land to be monitored, and the structured attribute record table will be embedded in the corresponding position in the electronic map to realize the updating of the electronic map. The structured attribute record table includes the corresponding area number and its judgment result.

[0030] The beneficial effects of this invention are:

[0031] (1) The system uses structural analysis units to rasterize remote sensing images and introduces the structural complexity index (ICI) as a quantitative indicator of the degree of regional perturbation. It evaluates each initial raster region and effectively characterizes the intensity of texture changes and edge density within the region. Compared with the traditional fixed-scale segmentation method, the structural complexity index can distinguish the structural expression difficulty of different regions, thereby supporting the dynamic adjustment of the downstream segmentation scale and the adaptive selection of the recognition path to generate a non-uniform region set. This mechanism enables the system to achieve more refined feature extraction and label discrimination in complex regions without increasing the overall computational load.

[0032] (2) By setting up a label generation unit, based on the structural differences corresponding to the non-homogeneous region set, and combined with the high-dimensional semantic feature vector output by the differential segmentation network structure, the pixel-level land cover category identification and label map construction of the entire remote sensing image are accurately realized, thereby improving the semantic expression ability and classification accuracy of the remote sensing image. The system first inputs each region in the non-homogeneous region set after structural complexity-driven segmentation into the matching segmentation network structure. For each pixel location, there is a corresponding high-dimensional semantic feature vector in the feature map, ensuring that the land cover determination at each location is based on the depth representation under sufficient contextual understanding. Finally, the output vector is converted into a probabilistic category probability vector. This vector can not only reflect the most likely label of the land cover type to which the pixel belongs, but also be used to obtain the model confidence score. By processing all pixels one by one and combining the region number information, the system can spatially reconstruct the classification results according to the non-homogeneous region structure, and finally generate a complete remote sensing label map with good spatial integrity and semantic consistency.

[0033] (3) By setting up a label comparison module, a pixel-level difference calculation, mask construction, and structural change recognition mechanism are introduced, enabling high-precision comparison of remote sensing image labels at different times and automatic extraction of abnormal areas of land cover changes. This improves the traceability, reliability, and automatic recognition capability of land cover label changes in land surveys. In the pixel-level label difference unit, spatial resampling and geometric registration are performed on the label image generated in the current time phase and the historical reference label image to ensure that the two images are highly consistent in terms of spatial coordinates, resolution, and rotation angle, thereby effectively avoiding misjudgment and false changes caused by pixel misalignment or scale inconsistency. Based on coordinate alignment, the system performs point-by-point comparison of the land cover category label at each pixel location. If the current label is inconsistent with the historical label, it is determined to be a changed pixel, and a change mask image is generated. This mask image marks the change state of each pixel in a binary manner of 0 / 1, providing a good foundation for visualization and structural processing. Subsequently, in the identification unit, the system performs connectivity clustering on the changed pixels in the change mask image, aggregating spatially adjacent changed pixels into several candidate label inconsistency regions. This clustering process not only improves the spatial representation capability of change detection but also provides logical entity support for subsequent regional analysis. More importantly, for each candidate region, the system further calculates the proportion of changed pixels, using this as a key basis for identifying structural changes. When the proportion of change is low, it indicates that only a few pixels in the region show differences, which may be due to occasional model misjudgments, edge jitter, or other non-real changes. Such regions will be automatically removed by the system to prevent false alarms. When the proportion of change is significant, the system marks it as a structural change region for subsequent ground cover label inconsistency analysis and explanation of the causes of change.

[0034] (4) By setting up a monitoring module and introducing a mechanism based on temporal change trend analysis and result feedback modeling, dynamic judgment and interpretation classification of areas with inconsistent remote sensing labels are realized, thereby effectively distinguishing the actual causes of temporary and long-term changes and enhancing the reliability, traceability, and interpretability of the land survey quality monitoring system's change judgment. Based on the current structural change area, the system tracks subsequent temporal data and compares the continuity of changes in the same area over time. If the area remains a structural change area in multiple subsequent temporal phases, it indicates that the change in the land feature is continuous and stable, and the system judges it as a non-temporary marker, i.e., a real and stable land use change. Conversely, if the area reverts to its original label or is no longer marked as a change in subsequent temporal phases, it indicates that the change is temporary or periodic, such as temporary dumping or agricultural rotation, and the system judges it as a temporary marker, preventing short-term anomalies from being misjudged as long-term land use changes, thereby ensuring the authenticity and timeliness of the label judgment. Finally, the feedback unit generates a structured attribute record table for each candidate area based on the above temporal judgment results and embeds it into the corresponding spatial location in the electronic map. This record table contains key attributes such as region number, change status (temporary / non-temporary), evolution phase trajectory, and change density. It can be used for map-based click queries and result tracing, and also supports filtering by conditions, classification statistics, and visualization of change distribution, providing an efficient data foundation for subsequent manual review, anomaly assessment, and policy response. Attached Figure Description

[0035] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application.

[0036] Figure 1 This is a block diagram of the present invention. Detailed Implementation

[0037] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0038] like Figure 1 As shown, the present invention proposes a land survey quality monitoring system for intelligent segmentation and feature identification of remote sensing images, including an image recognition module, a label comparison module and a monitoring module;

[0039] Specifically, the image recognition module acquires remote sensing images of the land to be monitored through remote sensing sensors, analyzes the structural complexity, divides the image regions to form a non-uniform region set, and analyzes and obtains the category probability vector of the pixel position in each region within the non-uniform region set to construct a label map of the entire remote sensing image.

[0040] The image recognition module includes an acquisition unit, a structural analysis unit, a segmentation scale unit, and a label generation unit;

[0041] The acquisition unit acquires remote sensing images of the land to be monitored using remote sensing sensors and preprocesses the images, including standardization processes such as radiometric correction, atmospheric correction, and geometric correction, to ensure physical consistency of spectral information in different bands and eliminate the effects of atmospheric scattering and differences in solar altitude angle.

[0042] The purpose of radiometric correction is to convert the original digital values ​​(DN values) in remote sensing images into physically meaningful radiance, so as to ensure that images acquired at different times or by different sensors are numerically comparable.

[0043] Atmospheric correction aims to eliminate radiation deviations caused by atmospheric aerosols, molecular scattering, water vapor absorption, etc., and restore the spectral reflectance characteristics of ground objects under atmospheric conditions. This is especially crucial in the visible and near-infrared bands. It can be achieved by using empirical model methods (such as the DOS method), assuming the existence of dark pixels with near-zero reflectance in the image (such as deep water), estimating atmospheric path radiation and making corrections.

[0044] The purpose of geometric correction is to correct the position of each pixel in a remote sensing image to its position in the real geographic coordinate system, eliminate spatial distortions caused by the curvature of the earth, topographic relief, sensor perspective, attitude error, etc., and achieve consistent spatial alignment with maps or historical images.

[0045] The structural analysis unit divides remote sensing images into regions to obtain multiple regularized initial raster regions, such as 128×128 pixels each. Although this division facilitates parallel processing and a unified input format, it lacks adaptability to the difficulty of recognition due to the complex distribution of ground features.

[0046] Therefore, it is necessary to analyze the structural complexity of each initial grid region. By directionally weighting the intensity of structural changes of each pixel in each initial grid region and averaging the results, a structural complexity index can be obtained. The structural complexity index is used to reflect the degree of local structural disturbance in each initial grid region.

[0047] The specific method for obtaining the structural complexity index is as follows: Wherein, ICI(R) is the structural complexity index of the R-th initial raster region, R is an initial raster region in the remote sensing image, (x, y) are the pixel coordinates, I(x, y) is the brightness of the pixel, and w(x, y) is the orientation weight factor, reflecting whether the point is located on the principal axis of the structural orientation, thus improving the structural sensitivity of the response. The Laplacian operator value for remotely sensed imagery measures the second-order rate of change of the image at that point, i.e., edge intensity; using... This is used to measure the second derivative of the grayscale change of each pixel. In an image, the changes are most drastic at the edges, resulting in the largest Laplacian value. Therefore, this part is used to measure the complexity of the texture structure within a region and whether it has a clear outline. Its mathematical definition is: Where I is the pixel value at a certain location in the image, θ is the partial derivative sign, indicating the derivative of the multivariable function with respect to one variable, θ 2 The symbol for the second-order partial derivative indicates the second-order partial derivative of a variable. It is the second derivative of the image in the horizontal x-direction, a second-order feature used to measure the degree of drastic change in the image in the horizontal direction, and can detect horizontal edges, stripes or abrupt changes in the horizontal structure; It is the second derivative of the image in the vertical direction y, which measures the degree of structural perturbation of the image in the vertical direction and can detect structural features such as vertical edges and textures.

[0048] use The purpose is to enhance the contribution of high regions and reduce background or small fluctuations. In an image, there are many small fluctuations, but the regions with real structural transitions have a large range of changes. Squaring can weight and amplify the influence of structural transition regions.

[0049] To improve the response values ​​of pixels with significant boundary structures in certain directions, the system also calculates a directional weight factor for each pixel. This weight is set based on whether the current pixel is located in a main texture direction or an edge direction in the image. If the pixel is located in a main structure direction, its weight value is higher; if it is in a blurred edge area or has an irregular direction, its weight value is lower. By introducing this weight coefficient, the system can more effectively highlight pixels that contribute to the structural contour, thereby enhancing the response value of complex structural areas.

[0050] Directional weighting factors are evaluation coefficients for the importance or sensitivity of structural perturbations in a certain direction, reflecting the contribution of that direction to the structural expression of regional texture, edges, and arrangement. The acquisition methods can be divided into three categories: empirical setting method, adaptive statistical method (based on local directional distribution), and deep learning driven method (training and learning). Among these, the empirical setting method is based on the main direction of common land cover structures in remote sensing images (such as roads and water bodies, which are mostly arranged horizontally or vertically), which is simple, efficient, and suitable for large-scale rapid processing. The adaptive statistical method analyzes the directional distribution using a gradient direction histogram (such as HoG) in each initial region, and the weights are automatically adjusted according to the pixel proportion or energy density distribution in a certain direction. The specific steps are: calculating the gradient direction of each pixel, constructing a regional direction histogram, and normalizing it by frequency to obtain the weighting factor. This method is suitable for land cover environments with large differences in structural complexity. The deep learning driven method analyzes the main direction of regional texture based on a trained neural network and automatically outputs the directional response intensity as the weight.

[0051] In this embodiment of the invention, by setting a collaborative processing mechanism between the structure analysis unit and the segmentation scale unit in the image recognition module, the structural complexity identification and dynamic segmentation scale control of complex areas of ground features in remote sensing images are realized. This effectively solves the problems of classification confusion, boundary misjudgment, and label instability when identifying high-complexity areas using traditional unified scale processing methods.

[0052] Traditional methods for remote sensing image segmentation typically employ a uniform grid scale, such as fixed 128×128 pixel blocks, to segment the entire image in parallel. However, this ignores the significant differences in texture complexity, edge density, and structural perturbation among different regions in remote sensing images. This leads to boundary misclassification, type confusion, or jigsaw puzzle effects, often occurring in areas with high structural complexity, such as urban boundaries, fragmented farmland, or mountain slopes. By introducing the Structural Complexity Index (ICI) as an adaptive criterion for region segmentation, the system can identify regions with significant structural perturbation and perform further segmentation, such as refining them into 64×64 or even 32×32 pixel blocks. For structurally stable regions, such as large areas of farmland or water bodies, region merging can be performed to form larger-granularity processing units. In this way, the system can form a non-uniform region set, improving the balance between the overall segmentation's structural adaptability and computational efficiency.

[0053] For example, in a remote sensing image, the upper left corner is a large area of ​​woodland with smooth boundaries and uniform texture, corresponding to a low structural complexity index; the lower right corner is a peri-urban area with a large number of buildings and road intersections, corresponding to a high structural complexity index. The system will merge the former into a large area and process it with a simple network; for the latter, the regional granularity will be refined and an enhanced path network will be used to improve the accuracy of boundary recognition.

[0054] The Structural Complexity Index (ICI) is used to measure the degree of local texture perturbation in each initial raster region of a remote sensing image. Specifically, it calculates by applying the Laplacian operator to each pixel within the region to extract the structural change response, multiplying it by a directional weighting factor, and then averaging the results to form a directional comprehensive perturbation value. This index accurately reflects the edge strength and structural complexity of the texture within the region and is a core basis for controlling segmentation granularity. The introduction of the ICI provides the system with a region-level discriminative index, meaning that instead of blindly dividing the image, it rationally determines which regions need refinement and which can be merged. Based on this, different semantic segmentation path structures, such as deep attention residual networks, are selected in subsequent processes to achieve differentiated allocation of model resources, improve accuracy in complex regions, and reduce computational redundancy in stable regions.

[0055] The aforementioned mechanism of the image recognition module is ultimately used to construct a label map for the entire remote sensing image. Because the system performs structure-sensitive region processing on the image and uses different network paths for pixel-level category probability calculation in each region, the final label map not only has higher boundary consistency and spatial continuity but also possesses information integrity supported by pixel confidence scores. This provides a robust input foundation for subsequent change detection, label comparison, and monitoring modules. By outputting the category probability vector for each pixel and applying Softmax normalization, the system can further provide a confidence score for each predicted label.

[0056] Furthermore, when dividing the scale unit, if the structural complexity index exceeds a preset complexity threshold, it indicates that the corresponding initial raster region has a complex structure, and a first-class region processing mechanism is output. If the structural complexity index does not exceed the preset complexity threshold, it indicates that the corresponding initial raster region has a stable structure, and a second-class region processing mechanism is output. If a first-class region processing mechanism is output, the corresponding initial raster region is further divided. If a second-class region processing mechanism is output, the corresponding initial raster region is merged with initial raster regions of the same type and adjacent positions.

[0057] The initial grid region segmentation scale is determined based on two types of region processing mechanisms to form a non-uniform region set. The segmentation path is then determined for each region in the non-uniform region set. The segmentation path refers to the neural network processing method used in the system for different regions to extract regional features and achieve pixel-level land cover classification. Different paths usually correspond to different network structures or processing intensities. This segmentation mechanism is used to adaptively select the corresponding network processing path based on the structural feature intensity of each image sub-region, thereby effectively controlling computational complexity and enhancing regional adaptation capability while ensuring recognition accuracy.

[0058] If a certain type of region processing mechanism is output, then the corresponding initial raster region will be further subdivided, including:

[0059] A minimum granularity limit is preset, and the structural complexity index of the corresponding initial raster region is determined to exceed a preset complexity threshold. Based on the structural complexity index being inversely proportional to the block size, the granularity of the corresponding initial raster region is set, specifically as follows: Among them, S R Let S be the granularity of the initial raster region. min The minimum granularity limit, i.e., the minimum allowed subdivision size, is used to prevent over-subdivision, which can lead to excessive subdivision and a significant increase in the system's computational burden. k is a scaling factor used to control the overall scale level, reflecting the granularity of the baseline region (typically dependent on the original image size). It is used to dynamically adjust the granularity of each initial raster region. When the ICI(R) is higher, the denominator will be larger and the whole will be smaller, and the region division will be more detailed. Conversely, the region division will be coarser.

[0060] The choice of scaling factor is essentially to set the average granularity baseline of the entire image across the blocks. This can be obtained through resolution normalization (image adaptation): based on the image spatial resolution (e.g., how many meters each pixel represents) and the smallest recognizable unit of the target feature, k = the desired physical size of the blocks in the target application (e.g., 20m) divided by the image spatial resolution (e.g., 1m / pixel). Alternatively, a fixed empirical method (static setting) can be used, setting a fixed value during system initialization based on the original resolution of the remote sensing image and the target application (e.g., 10m-level land survey or 0.5m-level urban monitoring). For example, if the original image is 5120×5120 pixels and the initial block division is 128×128, then k = 128 is recommended.

[0061] The determination of the segmentation path for non-homogeneous regions is carried out by grouping them into different regions, including:

[0062] Based on the non-homogeneous region set, distinguish between regions obtained through a type-one region processing mechanism and regions obtained through a type-two region processing mechanism within the non-homogeneous region set;

[0063] Regions obtained through the first type of region processing mechanism are processed using a deep attention residual path. The segmentation network structure corresponding to this path is a deep attention residual network, which is used to capture microstructures. Regions obtained through the second type of region processing mechanism are processed using a standard semantic segmentation path. The segmentation network structure corresponding to this path is the standard fully convolutional semantic segmentation model of U-Net, which is used to improve efficiency and suppress overfitting.

[0064] The deep attention residual path refers to a data processing path adopted by the system when dealing with regions of high structural complexity. The deep attention residual network is the specific network structure implementation upon which the deep attention residual path relies. This structure integrates the following: Residual Blocks: Inputs can bypass some convolutional layers and directly fuse with the output; Attention Modules: Guiding the network to focus on key regions or key channels, such as SE modules or CBAM modules; Deep convolutional stacked structures: The network depth is typically ≥20 layers, supporting multi-scale semantic information extraction. The standard semantic segmentation path corresponds to the processing strategy for regions with relatively stable structures and relatively homogeneous textures. The U-Net standard fully convolutional semantic segmentation model is the specific neural network architecture actually used in the standard semantic segmentation path. U-Net, a classic fully convolutional network architecture, is widely used in medical and remote sensing image segmentation tasks due to its symmetrical encoder and decoder structure and important skip connection design. This structure integrates the following: the encoder extracts image features layer by layer and performs spatial downsampling; the decoder uses upsampling operations to gradually restore spatial resolution, and at the same time, it fuses high-resolution features with decoded features through skip connections, thereby better restoring target boundaries and details.

[0065] Segmentation path refers to the neural network processing method used in the system for different regions to extract regional features and achieve pixel-level land cover classification. Different paths correspond to different network structures or processing intensities. Therefore, the system dynamically determines which segmentation network structure to use based on the structural complexity index of each image region.

[0066] The standardized image undergoes initial rasterization, dividing it into several equal-sized basic analysis regions. While this division facilitates parallel computation, it fails to reflect the structural differences within each region. Therefore, this system further evaluates the structural complexity of each initial region, constructing a structure-driven adaptive region system. The evaluation criteria for structural complexity include local edge density, texture directionality, and pixel grayscale variability. Based on the complexity level, the system automatically determines whether to refine a region into smaller sub-regions or merge it with neighboring structurally stable regions to form new, heterogeneous, adaptive spatial units. The core objective is to retain more recognition details in structurally complex regions while improving processing efficiency and recognition robustness in structurally simple regions. The output of this step is a structural complexity evaluation map and a hierarchical region partitioning map, providing a basic spatial analysis structure for subsequent segmentation and recognition modules.

[0067] In this embodiment of the invention, by introducing an adaptive region processing mechanism guided by structural complexity into the segmentation scale unit, the image segmentation granularity and semantic segmentation path can be dynamically adjusted according to the degree of structural disturbance in different regions of remote sensing images. This effectively solves the problems of difficult subdivision of structurally complex regions and waste of resources in structurally stable regions in traditional remote sensing segmentation, thereby improving the accuracy of ground feature identification and the overall processing efficiency of the system.

[0068] The system first uses the Structural Complexity Index (ICI) to determine the degree of structural perturbation in each initial raster region, enabling the system to possess content-sensitive segmentation preprocessing capabilities. If the ICI value of a region exceeds a preset complexity threshold, it indicates that the region has characteristics such as dense texture, interlaced edges, or complex semantic boundaries. The system then outputs a region processing mechanism to initiate the region re-segmentation process. In this process, a new refinement granularity is dynamically set according to a formula that shows an inverse relationship between the ICI value and the segmentation scale. For example, if the ICI value of a region is extremely high, the segmentation granularity will be close to S. min The system sets a limit to the fine-grained segmentation value, effectively improving semantic segmentation capabilities. If the ICI value is in the medium-high range, the system will automatically adjust to medium granularity to maintain a balance between segmentation efficiency and accuracy. Conversely, if the structural complexity index of a region is below a preset threshold, the system determines that its structure is stable, outputs a second-class region processing mechanism, and automatically triggers an adjacent region merging strategy: merging spatially continuous raster regions with consistent land cover types into larger regional units, thereby improving segmentation processing efficiency and reducing model inference redundancy.

[0069] After the non-homogeneous region set is formed, the system further allocates paths and selects network structures based on the processing mechanism corresponding to each region: For regions obtained through the first type of mechanism, due to their high structural complexity, the system adopts a deep attention residual path, which has strong perception and boundary preservation capabilities and is suitable for semantic extraction tasks in complex image scenes; for the second type of region, the standard semantic segmentation path is adopted, namely a fully convolutional neural network based on the U-Net structure, which has advantages such as high computational efficiency and stable convergence, and is suitable for regions with regular structures and stable semantics. For example, in a remote sensing image containing urban and farmland, the core urban area contains a large number of buildings and interwoven roads with complex boundaries, and its corresponding ICI value is usually higher than the threshold. Therefore, this region will trigger the first type of mechanism, which is finely segmented and processed by the deep attention residual network; while the large areas of farmland in the suburbs have clear boundaries and simple textures, with lower ICI values, which will trigger the second type of mechanism, where multiple plots are automatically merged and processed by U-Net, further improving the model's inference speed.

[0070] Through the above strategies, the system ensures the accuracy of identification in structurally complex areas while reducing the processing overhead in structurally simple areas, forming a collaborative identification mechanism of precise segmentation of difficult areas and rapid segmentation of easy areas. This enhances the system's ability to process large-scale remote sensing images and improves the classification consistency, boundary clarity, and system execution efficiency of the final labeled image.

[0071] Furthermore, the label generation unit, based on the segmentation network structure, obtains the category probability vector for each pixel location to construct a label map for the entire remote sensing image. Specific steps include:

[0072] The remote sensing images corresponding to each region in the non-uniform region set are input into the corresponding segmentation network structure. Spatial and semantic information is extracted through the convolutional feature extraction layer to output a feature map. The feature map is then upsampled using the deconvolution method to restore it to the same spatial size as the input image. In this way, each pixel position in the feature map corresponds to a set of high-dimensional semantic feature vectors, ensuring that each pixel has complete classification basis. Each pixel point after upsampling has a multi-dimensional vector in the feature map.

[0073] Spatial information refers to the location, structure, and boundary features of pixels in an image, such as the spatial arrangement of features, texture variations, edges, and shapes. Semantic information refers to the higher-order category meaning contained in an image. For example, if an area is farmland rather than woodland, it means that it is a "building structure" rather than a "natural landscape." Semantic information reflects what something is rather than where it is or what it looks like.

[0074] Feature maps are intermediate image representations of the output of each layer in a convolutional neural network. They describe the activation response of an image in a specific channel. In layman's terms, feature maps are an abstract description of the original image. Each layer's output extracted through convolution operations forms a set of feature maps.

[0075] A multidimensional vector is a set of high-dimensional numbers output by a neural network at each pixel, representing the semantic description of that pixel. For example, if the system needs to classify land cover into 7 categories, then each pixel will eventually correspond to a 7-dimensional vector, with each dimension representing the probability value of the pixel belonging to a certain category.

[0076] The multidimensional vector is mapped to an output vector of length C through a fully connected mapping layer (usually a 1×1 convolution), where C is the total number of land cover categories. For example, if 7 land cover categories are set (farmland, forest, water, buildings, roads, bare land, and grassland), the output is a 7-dimensional vector per pixel, with each dimension representing the score value of the corresponding category.

[0077] The output vector is normalized using the Softmax function to convert it into a probability form in order to obtain the class probability vector.

[0078] The land cover category with the highest probability is selected from the category probability vector as the final land cover category label; the land cover category label is the land surface category identified in the remote sensing image.

[0079] Traverse each pixel location and reassemble it into the original image according to the region numbers of the non-uniform region set to construct the label map of the entire remote sensing image.

[0080] The raw images here refer to the results of remote sensing imagery acquisition of the land to be monitored using remote sensing sensors;

[0081] Softmax guarantees that the sum of the C values ​​of each pixel is 1, and each value is in the range [0, 1], representing the relative probability that the pixel belongs to a certain type of land cover.

[0082] Based on the structural segmentation results, multiple land cover discrimination features are extracted for each segmented region and input into the corresponding segmentation network structure to obtain specific land cover label results. Feature extraction covers spectral information, such as multi-band reflectance values; texture information, such as the contrast and entropy of the gray-level co-occurrence matrix; geometric information, such as shape compactness and boundary curvature; and remote sensing index information, such as NDVI, NDBI, and NDWI. Multi-dimensional fusion modeling ensures that the system has strong discrimination capabilities. The system dynamically allocates different recognition model paths according to the structural complexity of the region. For example, a deep multi-scale fusion network is called in structurally complex regions, while a lightweight model is used in structurally simple regions. The model recognition process is a fully supervised semantic segmentation, which outputs a label map for each region through pixel-level prediction, thereby forming land cover classification results. Furthermore, the system maps land cover labels to a standard land use classification system, such as agricultural land, construction land, and water area. This mapping can be based on administrative planning layers or artificial intelligence rules. The output includes pixel label maps, areal vector feature patches, and regional land use statistics, providing a semantic basis for subsequent quality monitoring and consistency analysis.

[0083] In this embodiment of the invention, a label generation unit is set up to realize the process of automatically generating a complete label map from remote sensing images concentrated in non-homogeneous regions. Furthermore, the precision and reliability of remote sensing feature identification are improved through pixel-by-pixel category probability calculation and maximum value extraction mechanisms. Specifically, the system inputs each remote sensing image segment in a non-homogeneous region into its corresponding segmentation network structure, such as a deep attention residual network or a U-Net network. First, a convolutional feature extraction layer is used to extract a high-dimensional feature map containing spatial structure and semantic context. This step ensures that the model can capture the semantic correlation and local geometric features between pixels, avoiding misclassification of labels due to insufficient information at a single point. Subsequently, a deconvolution (upsampling) operation restores the feature map to the same spatial size as the original image, ensuring that each pixel obtains a complete semantic feature vector in the feature map. This vector represents the model's depth judgment basis for the pixel's category. Next, the system uses a 1×1 convolutional layer as a fully connected mapping layer to map the high-dimensional semantic vector of each pixel into an output vector of length C. For example, let C=7, representing seven categories: cultivated land, forest land, water body, building, road, bare land, and grassland. Each dimension of this output vector represents the score value of the corresponding category. The higher the score, the more likely the pixel is to be classified as that type of land feature.

[0084] To ensure the output has a clear classification probability, the system introduces a Softmax normalization function, transforming the scoring vector into a category probability vector with a sum of 1 and each dimension ranging from [0, 1]. Each component of this vector represents the relative probability that a pixel belongs to a certain type of land cover. For example, if the output of a pixel is {0.1, 0.05, 0.02, 0.72, 0.03, 0.05, 0.03}, then the land cover category corresponding to this pixel is building (category 4), with a probability of 0.72, indicating a classification confidence level of 72%. Subsequently, the system selects the category with the highest probability value from the category probability vector as the final label for the pixel.

[0085] Finally, the system reassembles all the processing results in the non-homogeneous region set by numbering, that is, it maps the classification results to the original remote sensing image locations according to the region index, thereby forming a complete remote sensing label map that is spatially consistent, label continuous, and has controllable confidence.

[0086] Through the above steps, this system not only achieves high-precision ground feature identification at the pixel level, but also establishes an interpretable structure with a probability set for each pixel. This enables the label image to possess high-resolution expressive capabilities and quantifiable confidence attributes for classification results, which is of great significance for land use supervision, critical area verification, and label mismatch tracking. For example, in rural construction land occupation identification, when the probability of building-related pixels is less than 0.5, it can be judged as low confidence and thus enters the queue for manual review, effectively avoiding false alarms and missed judgments. Therefore, the label generation unit of this invention, combined with a structurally adapted network path, accurate probability vector output, a reliable confidence quantification mechanism, and a regional-level restoration and stitching strategy, improves the overall performance of the remote sensing ground feature identification system in terms of label accuracy, stability, and traceability.

[0087] Specifically, the label comparison module performs pixel-by-pixel category comparison between the label image under the current time conditions and the reference label image to form a change mask image. Based on the change mask image, structural change regions are identified, and these structural change regions are used to determine the inconsistency of the land cover category labels.

[0088] The reference label map relies on historical remote sensing images or manually labeled land cover layers as a reference to ensure spatial consistency between the two label maps;

[0089] Furthermore, the label comparison module includes a pixel-level label difference unit and a recognition unit;

[0090] The pixel-level label difference unit performs spatial resampling and geometric registration on the current label image and the reference label image to ensure that the two have a consistent image coordinate system, avoid misjudgment due to positional errors, and performs pixel-by-pixel category comparison on the label image and the reference label image under the current time conditions. If the label is different at the same pixel position, it is recorded as a changed pixel and a change mask image is formed.

[0091] Resampling refers to the re-interpolation and calculation of image pixels during image geometric transformation to adapt to the target spatial resolution or transformed position. Registration refers to the spatial alignment of two or more images so that the same feature has the same position coordinates in different images (i.e., pixel alignment).

[0092] The change mask is a binary image. Specifically, if a pixel's current label is different from its historical label, the mask value is 1 (change); if the labels are the same, the mask value is 0 (no change).

[0093] The recognition unit clusters adjacent changing pixels in the change mask to extract connected regions as candidate label inconsistency regions. These regions represent the spatial range where the model recognition result is inconsistent with the historical label. For each candidate label inconsistency region, the ratio of changing pixels within it to the number of pixels in the corresponding clustered region is used to obtain the changing pixel ratio. This is to eliminate invalid regions where only a few pixels cause deviations and retain structurally changing regions. The changing pixel ratio is used to identify structurally changing regions.

[0094] The clustering process in the identification unit mainly includes two steps: connectivity detection and region generation. The specific process is as follows:

[0095] Step 1, Scan the change mask map: Traverse each pixel in the change mask map. If the mask value of the current pixel is 1, it means that it is a change pixel and is marked as an object to be clustered.

[0096] Step 2, Adjacent pixel connectivity analysis: For each changing pixel, search its neighboring pixels (usually using 4-neighborhood or 8-neighborhood). Where 4-neighborhood: up, down, left, right; 8-neighborhood: up, down, left, right + top left, top right, bottom left, bottom right. If the neighboring pixel is also a changing pixel (value 1), then classify it into the same connected region to obtain candidate label inconsistent regions.

[0097] To improve spatial robustness and generalization ability, the system performs neighborhood expansion to obtain region blocks containing the core pixel of the change and its neighboring pixels. This technique is common in image analysis systems and is called dilatational clustering. The final clustered region includes not only pixels with a value of 1 in the initial change mask but may also include some neighboring pixels with a value of 0. For example, if a group of changed pixels clusters into a 10×10 region, and there are 50 pixels with a value of 1 in the original change mask, the system will expand the boundary through dilatation to ensure coverage of the entire suspected region. In the end, the clustered region of changed pixels will contain 80 pixels, and there will be 60 pixels with a value of 1 in the change mask. At this point, the ratio of changed pixels is 60 to 80.

[0098] The clustering process here is not the traditional K-means, but connectivity-based clustering, such as FloodFill and Connected Component Labeling (CCL).

[0099] A connected region is a set of spatially connected pixels that form a closed or recognizable area.

[0100] Regions with inconsistent candidate labels may represent real changes in land cover, or they may be caused by model errors or label shifts. Therefore, they cannot be directly treated as identification errors. To improve analysis efficiency, the system further calculates the label change density of each candidate region, which is the proportion of pixels with label changes in the region. If the change density is too low, the region will be ignored or marked as noise change. Regions with significant label change density will be extracted as key detection objects. The system uses these regions as input targets for the next stage of label interpretation analysis to distinguish whether they belong to temporary occupancy or non-temporary occupancy, thus building the basis for label stability assessment.

[0101] In this embodiment of the invention, by setting pixel-level label difference units and structural change recognition units in the label comparison module, and combining spatial registration, change mask construction and regional clustering analysis mechanisms, the ability to identify areas with inconsistent land cover category labels in remote sensing images and the accuracy of change interpretation are improved, thereby providing a more robust data foundation and semantic support for land use monitoring and land cover change determination.

[0102] Specifically, the system first uses pixel-level label difference units to perform spatial resampling and geometric registration on the label image automatically generated in the current time phase and the historical reference label image. This ensures that the two images are aligned at the pixel level in the same coordinate system, eliminating spatial errors caused by sensor attitude, trajectory deviation, or coordinate system differences. This step is the basis for subsequent difference comparison and can effectively avoid the problem that the labels are actually consistent but are misjudged as changes due to position offset, thus improving the robustness of the comparison.

[0103] After spatial alignment is completed, the system performs a pixel-by-pixel category comparison operation for each pixel location: if the current temporal label is consistent with the historical label, it is judged as unchanged, and the corresponding pixel in the mask is assigned a value of 0; if they are inconsistent, it is judged as changed, and the corresponding pixel in the mask is assigned a value of 1, thus forming a binary change mask map. This map reflects the temporal distribution of category labels in the entire image and serves as the input basis for subsequent region clustering analysis.

[0104] Furthermore, the recognition unit clusters spatially adjacent changed pixels in the change mask image to extract regions with boundary continuity and spatial connectivity, called candidate label inconsistency regions. These candidate regions represent the spatial range where the model classification results differ from historical labels, which may be caused by factors such as actual ground feature changes, label offsets, identification mismatches, or temporary occupation. To avoid misjudgments caused by prediction deviations of individual pixels, the system calculates the proportion of changed pixels within each candidate region, i.e., the number of changed pixels divided by the total number of pixels in the region. If the proportion is lower than a preset threshold, the region is determined to be composed only of scattered noise pixels and is considered an invalid change region, which is then removed. If the proportion is higher, it indicates that the region has spatial structural continuity and regionality, and is thus identified as a structural change region. This mechanism helps to filter out isolated change points caused by occasional model errors, ensuring that the retained change regions have a real and reasonable evolution trend in both spatial and semantic aspects. For example, in a multi-temporal remote sensing image, if an agricultural plot changes from cultivated land to construction land, a large area of ​​pixels in the label image will change. By clustering the change mask image to form obvious connected components, the proportion of changed pixels exceeds 0.75, and the system will identify it as a structural change area. However, if only a few pixel labels in the neighboring area change from grassland to bare land, accounting for only 5% of the area, the system will judge it as an occasional error and remove it.

[0105] In summary, the tag comparison module of this invention achieves a closed loop of change recognition, from pixel-level error elimination to regional-level anomaly focusing, through key steps such as constructing a change mask map, extracting connected regions, and calculating change density. This improves the system's spatial positioning and change interpretation capabilities for inconsistent land use tags, ensuring high accuracy and interpretability in subsequent change cause analysis, temporal stability determination, and structured attribute recording.

[0106] Specifically, the monitoring module uses time-series analysis to determine whether areas with inconsistent candidate labels under different time phases are still areas of structural change, in order to obtain a judgment result. Based on the judgment result, the electronic map is updated. The judgment result includes non-temporary signs and temporary signs.

[0107] Furthermore, the monitoring module includes a timing analysis unit and a feedback unit;

[0108] If the proportion of changing pixels exceeds a preset proportion threshold, the temporal analysis unit marks the corresponding candidate label inconsistency region as a structural change region under the current temporal condition; otherwise, no marking is performed. If the corresponding candidate label inconsistency region is marked as a structural change region under the current temporal condition, the temporal analysis is used to determine whether the candidate label inconsistency region under the current temporal condition is still a structural change region under other temporal conditions, so as to obtain the judgment result.

[0109] It should be noted that the other time phase conditions here refer to the time phases after the current time phase. Time phases represent different points in time. Through time series analysis, the fluctuation of time series labels can be effectively judged and the "round trip" phenomenon of labels can be identified.

[0110] The feedback unit determines whether a region with inconsistent candidate labels under the current time phase is still a region of structural change under other time phases. If so, the region is determined to be a non-temporary label. Otherwise, it is determined to be a temporary label. Based on the determination result, a structured attribute record table is generated for the corresponding region in the land to be monitored. The structured attribute record table is then embedded in the corresponding location in the electronic map to update the electronic map. The structured attribute record table includes the corresponding region number and its determination result.

[0111] The structured attribute record table is embedded in the corresponding location in the electronic map to interpret the content of the corresponding area.

[0112] In this embodiment of the invention, by setting up a monitoring module and introducing a temporal analysis mechanism and a label change feedback mechanism, the behavior patterns of candidate label inconsistent areas under multiple temporal conditions can be dynamically tracked and reliably judged, realizing intelligent differentiation between temporary changes and real changes, and improving the land survey system's ability to discern the causes of label anomalies and its level of structured expression of change information.

[0113] Specifically, the monitoring module consists of a temporal analysis unit and a feedback unit. Its core logic is as follows: First, in the current time phase, the system identifies several candidate label inconsistency areas through the preceding label comparison module, and calculates the proportion of changed pixels in each area (i.e., the ratio of the number of changed pixels in the area to the total number of pixels). If the proportion exceeds the proportion threshold set by the system (e.g., 20%), the area is marked as a structural change area in the current time phase; otherwise, it is considered that the change is weak and does not need to be tracked.

[0114] Furthermore, to determine whether a change is a temporary disturbance or a genuine transformation, the temporal analysis unit dynamically tracks whether the structurally changed area continues to exhibit a changing state in subsequent time phases. If the area remains structurally changed in two or more subsequent time phases, it is considered to have a stable change trend and is marked as a non-temporary marker (genuine change); conversely, if the area returns to a stable state in subsequent time phases (i.e., the mask value returns to 0), it is determined to be a short-term change or label drift and marked as a temporary marker. For example, in practical applications, if an area's label changes from street to commercial stall in the current time phase, but the changed area is identified as a street again in subsequent daytime remote sensing imagery, it can be determined that the change in the area is not continuous. Based on this, the system automatically marks it as a temporarily occupied area. This mechanism effectively avoids misjudgment of land use due to a one-time change, thereby ensuring the stability and reliability of label map updates.

[0115] Finally, the system encodes the above judgment results into a structured attribute record table through the feedback unit and binds it with the spatial information of the changed area to achieve an integrated expression of attributes and space. The structured record table includes key fields such as area number, judgment mark (temporary / non-temporary), and time phase range, and is embedded in the corresponding position on the electronic map. This allows supervisors to click on any changed area in the map interface to view its change nature, duration, and judgment basis, thereby improving the system's interpretability, traceability, and user interaction efficiency.

[0116] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A land survey quality monitoring system for intelligent segmentation and feature identification of remote sensing images, characterized in that: include, The image recognition module acquires remote sensing images of the land to be monitored through remote sensing sensors. After structural complexity analysis, it divides the image regions to form a non-homogeneous region set. It then analyzes and obtains the category probability vector of the pixel position in each region within the non-homogeneous region set to construct a label map for the entire remote sensing image. The image recognition module includes an acquisition unit, a structural analysis unit, a segmentation scale unit, and a label generation unit; If the structural complexity index of the segmented scale unit exceeds the preset complexity threshold, it indicates that the structure of the corresponding initial raster region is complex, and a first-class region processing mechanism is output. If the structural complexity index does not exceed the preset complexity threshold, it indicates that the structure of the corresponding initial raster region is stable, and a second-class region processing mechanism is output. If a first-class region processing mechanism is output, the corresponding initial raster region is further subdivided. If a second-class region processing mechanism is output, the corresponding initial raster region is merged with the initial raster regions of the same type and adjacent positions. Specifically, the initial raster region segmentation scale is determined based on two types of region processing mechanisms to form a non-uniform region set. The segmentation path for each region within this set is then determined. If a type of region processing mechanism is output, the corresponding initial raster region is further subdivided, including: A minimum granularity limit is preset, and the structural complexity index of the corresponding initial grid region is determined to exceed a preset complexity threshold; Based on the structural complexity index being inversely proportional to the block scale, the granularity of the initial raster region is set as follows: ,in, Let R be the initial grid region's granularity. To define the minimum granularity limit, k is the scaling factor. This is used to dynamically adjust the granularity of each initial raster region. is the structural complexity index of the initial R-th raster region; The label comparison module performs pixel-by-pixel category comparison between the label image under the current time conditions and the reference label image to form a change mask image. Based on the change mask image, structural change regions are identified, and structural change regions are used to determine the inconsistency of land cover category labels. The monitoring module uses time-series analysis to determine whether areas with inconsistent candidate labels under different time phases are still areas of structural change, in order to obtain judgment results. Based on the judgment results, the electronic map is updated. The judgment results include non-temporary signs and temporary signs.

2. The land survey quality monitoring system for intelligent segmentation and feature identification of remote sensing images according to claim 1, characterized in that: The acquisition unit acquires remote sensing images of the land to be monitored through remote sensing sensors and preprocesses the remote sensing images, including standardization processing such as radiometric correction, atmospheric correction and geometric correction. The structural analysis unit divides the remote sensing image into regions to obtain multiple regularized initial raster regions. It then analyzes the structural complexity of each initial raster region. By directionally weighting the structural change intensity of each pixel within each initial raster region and averaging the results, a structural complexity index is obtained. This index reflects the degree of local structural perturbation in each initial raster region. The specific method for obtaining the structural complexity index is as follows: R represents an initial raster region in the remote sensing image. For pixel coordinates, The brightness of a pixel. As the directional weighting factor, is the Laplacian operator value for the remote sensing image.

3. The land survey quality monitoring system for intelligent segmentation and feature identification of remote sensing images according to claim 2, characterized in that: The determination of the segmentation path for non-homogeneous regions is carried out by grouping them into different regions, including: Based on the non-homogeneous region set, distinguish between regions obtained through a type-one region processing mechanism and regions obtained through a type-two region processing mechanism within the non-homogeneous region set; Regions obtained through the first type of region processing mechanism are processed using a deep attention residual path, and the corresponding segmentation network structure is a deep attention residual network. Regions obtained through the second type of region processing mechanism are processed using a standard semantic segmentation path, and the corresponding segmentation network structure is the standard fully convolutional semantic segmentation model of U-Net.

4. The land survey quality monitoring system for intelligent segmentation and feature identification of remote sensing images according to claim 3, characterized in that: The label generation unit, based on the segmentation network structure, obtains the category probability vector for each pixel location to construct a label map for the entire remote sensing image. Specific steps include: The remote sensing images corresponding to each region in the non-homogeneous region set are input into the corresponding segmentation network structure. Spatial and semantic information are extracted through the convolutional feature extraction layer to output the feature map. The feature map is then upsampled using the deconvolution method. Each pixel in the upsampled feature map has a multi-dimensional vector. The multidimensional vector is mapped to an output vector of length C through a fully connected mapping layer, where C is the total number of land cover categories; The output vector is normalized using the Softmax function to obtain the class probability vector; The land cover category with the highest probability is selected from the category probability vector as the final land cover category label; Traverse each pixel location and reassemble it into the original image according to the region numbers of the non-uniform region set to construct the label map of the entire remote sensing image.

5. The land survey quality monitoring system for intelligent segmentation and feature identification of remote sensing images according to claim 4, characterized in that: The label comparison module includes pixel-level label difference units and recognition units; The pixel-level label difference unit performs spatial resampling and geometric registration on the current label image and the reference label image, and performs pixel-by-pixel category comparison on the label image and the reference label image under the current time condition. If the label is different at the same pixel position, it is recorded as a changed pixel and a change mask image is formed. The identification unit clusters adjacent changing pixels in the change mask to extract connected regions as candidate label inconsistency regions. For each candidate label inconsistency region, the ratio of changing pixels within it to the number of pixels in the corresponding clustered region is used to obtain the changing pixel ratio, which is used to identify structurally changing regions.

6. The land survey quality monitoring system for intelligent segmentation and feature identification of remote sensing images according to claim 5, characterized in that: The monitoring module includes a timing analysis unit and a feedback unit; If the proportion of changing pixels exceeds a preset proportion threshold, the temporal analysis unit marks the corresponding candidate label inconsistency region as a structural change region under the current temporal condition; otherwise, no marking is performed. If the corresponding candidate label inconsistency region is marked as a structural change region under the current temporal condition, the temporal analysis is used to determine whether the candidate label inconsistency region under the current temporal condition is still a structural change region under other temporal conditions, so as to obtain the judgment result.

7. The land survey quality monitoring system for intelligent segmentation and feature identification of remote sensing images according to claim 6, characterized in that: The feedback unit determines whether a region with inconsistent candidate labels under the current time phase is still a region of structural change under other time phases. If so, the region is determined to be a non-temporary label. Otherwise, it is determined to be a temporary label. Based on the determination result, a structured attribute record table is generated for the corresponding region in the land to be monitored. The structured attribute record table is then embedded in the corresponding location in the electronic map to update the electronic map. The structured attribute record table includes the corresponding region number and its determination result.

Images