Artificial Intelligence-Based Land Spatial Survey Data Acquisition and Monitoring System

Abstract

This invention relates to the field of image processing technology, specifically to an artificial intelligence-based land space survey data acquisition and monitoring system. The system obtains different regional units based on the color distribution characteristics of the monitored images; obtains the decomposition necessity based on the differences in texture intensity distribution within the regional units; obtains an adaptive step size based on the difference in decomposition necessity between the initial downsampled image and the corresponding downsampled object; iteratively downsamples the initial downsampled image using the adaptive step size to obtain sub-images of different scales; obtains the size evaluation feature value of the convolution kernel based on the downsampling order characteristics, decomposition necessity, and pixel count characteristics of the sub-images; and adaptively adjusts the preset convolution kernel size based on the size evaluation feature value. This invention uses a convolutional neural network and adaptive size to acquire features of sub-images and perform feature modeling of land space, improving the accuracy of land space modeling and monitoring.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and more specifically to an artificial intelligence-based land space survey data acquisition and monitoring system. Background Technology

[0002] In land space surveys and dynamic monitoring, spatial image data is typically acquired through multiple sources, including satellite remote sensing, UAV aerial photography, and ground surveying. The collected data generally exhibits characteristics of multi-source heterogeneity, multi-scale distribution, and significant regional differences, with dynamic variations in texture complexity and information contribution across different regions. When extracting features from the acquired land space image data, existing convolutional neural network methods are typically used to extract image features, such as local regional boundaries and trends in landform changes.

[0003] However, while convolution can capture local texture changes and boundary information, traditional fixed convolution kernels cannot adaptively adjust their feature extraction strategies when faced with features of different scales and regions. For example, urban areas usually have complex edge structures and rich details, while farmland or woodland exhibits relatively smooth texture features. Single-scale convolution operations cannot effectively adapt to the feature extraction needs of different regions, which may lead to the loss of detailed information or insufficient smoothness in the overall trend extraction. This limits the ability of subsequent land space models to express spatial information and affects the accuracy and precision of spatial monitoring results. Summary of the Invention

[0004] To address the aforementioned technical problems, the present invention aims to provide an artificial intelligence-based land space survey data acquisition and monitoring system, the specific technical solution of which is as follows:

[0005] The image acquisition module is used to acquire monitoring images of the national land space;

[0006] The image analysis module is used to obtain different region units based on the color distribution characteristics of the monitored image; obtain the decomposition necessity of the region units based on the difference in texture intensity distribution within the region units; determine whether to downsample based on the decomposition necessity; downsample the region units according to a preset minimum step size to obtain an initial downsampled image; obtain a step size adjustment coefficient based on the difference in decomposition necessity between the initial downsampled image and the corresponding downsampled object; obtain an adaptive step size based on the step size adjustment coefficient; and continue iterative downsampling of the initial downsampled image according to the adaptive step size to obtain sub-images of different scales.

[0007] The feature processing module is used to obtain the size evaluation feature value of the convolution kernel based on the downsampling order feature of the sub-image, the decomposition necessity and the number of pixels; and to adaptively adjust the preset convolution kernel size based on the size evaluation feature value to obtain an adaptive size.

[0008] The spatial modeling module is used to obtain the features of the sub-image based on the convolutional neural network and the adaptive size, and to perform feature modeling of the national land space.

[0009] Furthermore, the step of obtaining different region units based on the color distribution characteristics of the monitored image includes:

[0010] The monitoring image is smoothed by mean filtering to obtain a smoothed image; a feature vector is constructed based on the coordinates and color of each pixel in the smoothed image; the pixels are clustered based on the differences in the feature vectors to obtain different clusters; the range of pixels in the same cluster in the monitoring image is taken as a region unit.

[0011] Further, the step of obtaining the decomposition necessity of the region unit based on the difference characteristics of texture intensity distribution in the region unit includes:

[0012] In the formula, S represents the decomposition necessity degree of the region unit. This indicates the preset weight, and N represents the number of pixels within the region unit. Linear normalization is represented; the maximum color difference between a pixel in the region unit and other pixels within a preset neighborhood is calculated to obtain the local color difference feature value of the pixel; This represents the local color difference feature value of the nth pixel. This represents the average value of the color difference feature of pixels within a region unit. This represents the upper limit of the color difference feature value.

[0013] Furthermore, the step of determining whether to downsampling based on the decomposition necessity includes:

[0014] The region units whose decomposition necessity exceeds a preset threshold are downsampled.

[0015] Further, the step of obtaining the step size adjustment coefficient based on the difference in decomposition necessity between the initial downsampled image and the corresponding downsampled object includes:

[0016] In the formula, W represents the step size adjustment coefficient of the initial downsampled image. This indicates the decomposition necessity of the initial downsampled image. This indicates the decomposition necessity of the downsampled image corresponding to the initial downsampled image.

[0017] Further, the step of obtaining the adaptive step size based on the step size adjustment coefficient includes:

[0018] In the formula, L represents the adaptive step size for the next downsampling. This indicates the preset minimum step size. This indicates the preset maximum step size, and W represents the step size adjustment coefficient. This indicates rounding down to the nearest integer.

[0019] Furthermore, the step of iteratively downsampling the initial downsampled image according to the adaptive step size to obtain sub-images of different scales includes:

[0020] The initial downsampled image is downsampled according to the adaptive step size to obtain a new downsampled image; the adaptive step size of the new downsampled image is obtained cyclically and downsampling continues until the decomposition necessity of the finally obtained downsampled image does not exceed a preset threshold; all images obtained by the region unit during the downsampling process are used as sub-images of different scales.

[0021] Further, the step of obtaining the convolution kernel size evaluation feature value based on the downsampling order feature of the sub-image, the decomposition necessity, and the pixel count feature includes:

[0022] In the formula This represents the size evaluation feature value corresponding to the q-th sub-image. This indicates the preset first weighting factor. This indicates a preset second weighting factor. This indicates a preset third weighting factor. This indicates the downsampling order corresponding to the q-th sub-image, and T represents the number of sub-images. Let H represent the decomposition necessity of the q-th sub-image, and let H represent the number of pixels in the region unit. This represents the number of pixels in the q-th sub-image.

[0023] Further, the step of adaptively adjusting the preset convolutional kernel size based on the size evaluation feature value to obtain an adaptive size includes:

[0024] In the formula This represents the adaptive size of the convolution kernel corresponding to the q-th sub-image. This represents the function that takes the nearest odd number. This indicates the minimum size of the preset convolution kernel. Indicates the maximum size of the preset convolution kernel. This represents the size evaluation feature value of the q-th sub-image.

[0025] The present invention has the following beneficial effects:

[0026] In this invention, acquiring different region units enables the segmentation of areas with significant feature differences in the monitored image; acquiring the decomposition necessity degree allows for the determination of the necessity for multi-scale analysis based on the texture features in the region units, thereby improving the accuracy of subsequent feature extraction. Acquiring the initial downsampled image ensures that image details are preserved relatively completely at the start of downsampling; obtaining the step size adjustment coefficient more accurately determines the scale differences during downsampling; acquiring the adaptive step size allows for adaptive downsampling based on image features, improving the accuracy and precision of subsequent feature extraction; acquiring different sub-images allows for the acquisition of effective features at more scales during the convolution process. Acquiring size evaluation feature values ​​and adaptive size allows the size of the convolution kernel to be set based on image features during convolution, further improving the accuracy of feature extraction for sub-images at different scales. Finally, features of sub-images are acquired based on the convolutional neural network and adaptive size, and feature modeling of land space is performed, improving the accuracy of land space modeling and monitoring. Attached Figure Description

[0027] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention 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 only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0028] Figure 1 This is a block diagram of an artificial intelligence-based land space survey data acquisition and monitoring system provided as an embodiment of the present invention. Detailed Implementation

[0029] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of an artificial intelligence-based land space survey data acquisition and monitoring system proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0030] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0031] The following description, in conjunction with the accompanying drawings, details a specific scheme for an artificial intelligence-based land space survey data acquisition and monitoring system provided by the present invention.

[0032] Please see Figure 1 The diagram illustrates a block diagram of an artificial intelligence-based land space survey data acquisition and monitoring system according to an embodiment of the present invention. The system includes the following modules:

[0033] The image acquisition module S1 is used to acquire monitoring images of the national land space.

[0034] This invention primarily targets multi-source heterogeneous image data acquired during land space surveys. In the feature extraction stage, a multi-scale processing mechanism and an adaptive adjustment method for the kernel size parameter in a convolutional neural network are introduced to improve the final land space model construction capability and spatial information expression accuracy based on the convolutional neural network. First, monitoring images of the target area's land space are acquired using image acquisition equipment.

[0035] The image analysis module S2 is used to obtain different region units based on the color distribution characteristics of the monitored image; obtain the decomposition necessity of the region units based on the difference in texture intensity distribution within the region units; determine whether to downsample based on the decomposition necessity; downsample the region units according to a preset minimum step size to obtain an initial downsampled image; obtain a step size adjustment coefficient based on the difference in decomposition necessity between the initial downsampled image and the corresponding downsampled object; obtain an adaptive step size based on the step size adjustment coefficient; and continue iterative downsampling of the initial downsampled image according to the adaptive step size to obtain sub-images of different scales.

[0036] In the process of land spatial surveys and dynamic monitoring, the acquired multi-source monitoring images typically have wide coverage and strong spatial heterogeneity. Different regions exhibit significant differences in surface type, degree of change, and texture structure. For example, urban areas usually contain a large amount of edge and high-frequency detail information, while farmland and forest areas are mainly characterized by large-scale continuous distribution and relatively gentle texture changes. In this case, if a uniform modeling and processing method is used for the entire monitoring image, especially when using convolutional neural networks for feature extraction, the convolution operation relies on fixed perception and weight sharing mechanisms, which can easily lead to convergent responses in different regions. This makes it difficult to effectively adjust for regional differences, thereby weakening the ability to depict local details and potentially introducing redundant features, reducing the effectiveness and discriminative power of overall feature representation. Therefore, it is necessary to divide the monitoring image at the initial stage of image analysis, dividing regions with similar feature attributes into several relatively unified regional units. This facilitates subsequent feature extraction and multi-source data processing within a more locally consistent range, thus providing a foundation for adaptive multi-scale modeling and convolutional feature learning, while improving the ability to express complex spatial information and the accuracy of modeling. Therefore, different regional units are obtained based on the color distribution characteristics of the monitored images.

[0037] Preferably, in this embodiment of the invention, the step of obtaining different region units includes: smoothing the monitoring image by mean filtering to obtain a smoothed image; mean filtering can reduce the interference of noise and local texture fluctuations on clustering, thereby improving the stability and accuracy of region division. A feature vector is constructed based on the coordinates and color of each pixel in the smoothed image, and the pixels are clustered based on the differences in the feature vectors to obtain different clusters; specifically, this includes: linearly normalizing the coordinates and RGB channel values ​​respectively, constructing a feature vector based on the normalized result, measuring the similarity between pixels using the Euclidean distance between feature vectors, and using the existing K-means++ clustering algorithm to cluster the pixels. The number of clusters is obtained using the elbow method. The positions and colors of the pixels within each cluster are relatively similar. Therefore, the area formed by pixels in the same cluster in the monitoring image is considered as a region unit.

[0038] Furthermore, considering the differences in texture complexity and intensity among different regions, their corresponding multi-scale decomposition requirements also differ significantly. Therefore, adaptive multi-scale decomposition of region units is necessary. Thus, the decomposition necessity of a region unit can be obtained based on the differences in texture intensity distribution within that unit. Preferably, in this embodiment of the invention, the step of obtaining the decomposition necessity includes:

[0039]

[0040] In the formula, S represents the decomposition necessity degree of the region unit. The preset weight is 0.4 in this embodiment of the invention, but the implementer can determine it according to the implementation scenario. N represents the number of pixels in the region unit. This indicates linear normalization. It calculates the maximum color difference between a pixel in a given region and other pixels within a preset neighborhood, obtaining the local color difference feature value of that pixel. The preset neighborhood is the eight-neighborhood of that pixel, and the color difference refers to the average of the absolute values ​​of the differences between the RGB channels between two pixels. This represents the local color difference feature value of the nth pixel. This represents the average value of the color difference feature of pixels within a region unit. The larger the average value, the greater the texture intensity within that region unit, and the more necessary it is to decompose it into multiple scales in order to obtain local details and overall trends. This represents the upper limit of the color difference eigenvalue. The upper limit refers to the maximum value that the color difference eigenvalue can theoretically reach, and it is used to normalize molecules. The larger the value, the greater the dispersion of texture intensity distribution within the region unit, and the wider the distribution of texture strength, thus increasing the necessity for multi-scale decomposition. The decomposition necessity value can simultaneously characterize both texture strength and distribution differences, thereby more accurately measuring the necessity of multi-scale decomposition for the region unit. After obtaining the decomposition necessity value of the region unit, it can be determined whether to downsample. Specifically, this includes downsampling region units with a decomposition necessity value exceeding a preset threshold. In this embodiment of the invention, the preset threshold is 0.2, which can be determined by the implementer according to the implementation scenario.

[0041] Furthermore, due to the varying texture complexity of different regions, using a fixed downsampling stride when downsampling different regions makes it difficult to balance detail preservation and scale differentiation. For example, in regions with strong texture, a smaller stride may result in insignificant feature differences between different scales, reducing the sensitivity of subsequent convolutional feature extraction to scale changes. Conversely, in regions with weak texture, a larger stride may lead to excessive loss of detail information, resulting in excessive feature differences between different scales and affecting the stable modeling of multi-scale transformation patterns. Therefore, an adaptive adjustment mechanism for the downsampling stride is needed. For regions that meet the downsampling conditions, the regions are first downsampled according to a preset minimum stride to obtain an initial downsampled image. A basic scale hierarchy is then established while preserving as much detail information as possible. In this embodiment, the preset minimum stride is 1, which can be determined by the implementer based on the implementation scenario. After obtaining the initial downsampled image, a stride adjustment coefficient can be obtained based on the difference in decomposition necessity between the initial downsampled image and the corresponding downsampled object. Preferably, in this embodiment, the step of obtaining the stride adjustment coefficient includes:

[0042]

[0043] In the formula, W represents the step size adjustment coefficient of the initial downsampled image. This indicates the decomposition necessity of the initial downsampled image. This represents the decomposition necessity of the downsampled image corresponding to the initial downsampled image. The downsampled image corresponding to the initial downsampled image is an image of a region unit; the method for obtaining the decomposition necessity of different images is the same, and will not be repeated here. The higher the decomposition necessity of the initial downsampled image, the stronger its potential texture features, and the larger the decomposition step size may need to be. The smaller the value, the smaller the difference in feature changes between adjacent scales for that region unit, and the larger the step size adjustment coefficient will be for the next downsampling decomposition. Therefore, the larger the step size coefficient, the larger the step size is required; hence, an adaptive step size is obtained based on the step size adjustment coefficient; preferably, in this embodiment of the invention, the step of obtaining the adaptive step size includes:

[0044]

[0045] In the formula, L represents the adaptive step size for the next downsampling. This indicates the preset minimum step size. This indicates the preset maximum step size, and W represents the step size adjustment coefficient. This indicates rounding down. In this embodiment of the invention, the preset minimum step size is 1, and the preset maximum step size is 5. The implementer can determine the step size according to the implementation scenario. The larger the step size adjustment coefficient, the larger the adaptive step size of the next downsampling.

[0046] Furthermore, the initial downsampled image can be iteratively downsampled using an adaptive step size to obtain sub-images at different scales. Specifically, this involves: downsampling the initial downsampled image using an adaptive step size to obtain a new downsampled image; obtaining a corresponding step size adjustment coefficient based on the new downsampled image; and iteratively obtaining an adaptive step size for the new downsampled image based on the new step size adjustment coefficient, continuing downsampling until the decomposition necessity of the final downsampled image does not exceed a preset threshold; all images obtained by this region unit during the downsampling process are used as sub-images at different scales. This downsampling strategy can gradually enhance the differences between scales while avoiding excessive information loss in the initial stage, making the feature representations at different scales more discriminative, thereby providing more stable and effective data support for subsequent multi-scale feature extraction and convolutional adaptive processing.

[0047] The feature processing module S3 is used to obtain the size evaluation feature value of the convolution kernel based on the downsampling order feature, decomposition necessity and pixel number feature of the sub-image; and to adaptively adjust the preset convolution kernel size based on the size evaluation feature value to obtain an adaptive size.

[0048] After obtaining sub-images at different scales, due to significant differences in spatial resolution and information representation capabilities, the degree of detail preservation and texture richness vary across different scales. Therefore, the structural parameters of the convolution kernel need to be adaptively adjusted during convolutional feature extraction. Larger-scale, low-resolution images focus more on expressing overall structure and trend information; using a larger convolution kernel in this case is beneficial for capturing the overall trend, thereby enhancing the extraction capability of macroscopic spatial distribution features. Conversely, smaller-scale, high-resolution images have richer detail and texture features; using a smaller convolution kernel can more effectively preserve local structural information, thereby improving the ability to characterize microscopic changes. Therefore, by constructing an adaptive adjustment mechanism for the convolution kernel, differentiated configurations of feature extraction strategies at different scales can be achieved, improving the targeting and effectiveness of multi-scale feature expression. Thus, the size evaluation feature value of the convolution kernel is obtained based on the downsampling order feature, decomposition necessity, and pixel count feature of the sub-image. Preferably, in this embodiment of the invention, the step of obtaining the size evaluation feature value includes:

[0049]

[0050] In the formula, This represents the size evaluation feature value corresponding to the q-th sub-image. This indicates the preset first weighting factor. This indicates a preset second weighting factor. This indicates a preset third weighting factor. In this embodiment of the invention, the weights are 0.2, 0.5, and 0.3 in sequence, with a total of 1. The implementer can determine the value of the weighting factor according to the implementation scenario. The downsampling order corresponds to the q-th sub-image. For the original region unit image, the downsampling order is 1, and for the initial downsampled image, the downsampling order is 2. T represents the number of sub-images. It should be noted that the original region unit image is also one of the sub-images. The smaller the value, the closer the sub-image is to the original high-resolution scale, and the richer the details in the image, so the required convolution kernel size is relatively smaller. represents the decomposition necessity of the q-th sub-image. A higher decomposition necessity indicates higher texture complexity and richer detail information, therefore requiring a smaller convolution kernel size. H represents the number of pixels in a region unit. This represents the number of pixels in the q-th sub-image. Fewer pixels mean greater information compression relative to the original scale and more severe loss of detail. Therefore, a larger convolutional kernel size is needed to capture the overall trend features, thereby enhancing the ability to extract overall structural information. By fusing these three features, the evaluation of the convolutional kernel size can simultaneously reflect multiple dimensions such as scale level, regional texture feature intensity, and information compression, achieving a comprehensive evaluation. Furthermore, the preset convolutional kernel size can be adaptively adjusted based on the size evaluation feature value to obtain an adaptive size. Preferably, in this embodiment of the invention, the step of obtaining the adaptive size includes:

[0051]

[0052] In the formula, This represents the adaptive size of the convolution kernel corresponding to the q-th sub-image. This represents the function that takes the nearest odd number. This indicates the minimum size of the preset convolution kernel. Indicates the maximum size of the preset convolution kernel. This represents the size evaluation feature value of the q-th sub-image. In this embodiment of the invention, the preset minimum size of the convolution kernel is 3, and the preset maximum size of the convolution kernel is 11. The implementer can determine the size according to the implementation scenario.

[0053] The spatial modeling module S4 is used to obtain features of sub-images based on convolutional neural networks and adaptive sizes, and to perform feature modeling of national land space.

[0054] By acquiring an adaptive size for the convolutional kernel, local feature expression can be highlighted in areas rich in detail, while global trend extraction can be strengthened in structurally dominant areas. This significantly improves the accuracy and effectiveness of multi-scale feature extraction, providing high-quality feature input for subsequent multi-source data fusion and spatial information modeling. Furthermore, features of sub-images are obtained based on the convolutional neural network and the adaptive size, and land space feature modeling is performed. The adaptive-sized convolutional kernel is embedded into the convolutional layer, replacing the traditional fixed convolutional kernel for feature learning. After inputting sub-images of corresponding scales for each region into the convolutional neural network, an adaptive response can be achieved, allowing for the progressive extraction of low-level texture features and high-level semantic features during multi-layer convolution, activation, and pooling processes, thereby obtaining a multi-scale feature representation with strong discriminative capabilities. It should be noted that the convolutional neural network is existing technology, and the specific steps will not be elaborated further. Compared to traditional methods, this embodiment of the invention, through dynamic adjustment of the convolutional kernel parameters, enables the network to optimize feature extraction strategies for spatial heterogeneity differences, strengthening local feature expression in areas rich in detail and enhancing global information perception in structurally dominant areas, thereby effectively improving the accuracy and stability of land space feature modeling.

[0055] In summary, this invention provides an artificial intelligence-based land space survey data acquisition and monitoring system. It obtains different regional units based on the color distribution characteristics of monitored images; obtains the decomposition necessity based on the differences in texture intensity distribution within the regional units; obtains an adaptive step size based on the difference in decomposition necessity between the initial downsampled image and the corresponding downsampled object; it iteratively downsamples the initial downsampled image using the adaptive step size to obtain sub-images of different scales; obtains the size evaluation feature value of the convolution kernel based on the downsampling order characteristics, decomposition necessity, and pixel count characteristics of the sub-images; and adaptively adjusts the preset convolution kernel size based on the size evaluation feature value. This invention improves the accuracy of land space modeling and monitoring by acquiring sub-image features using a convolutional neural network and adaptive size.

[0056] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0057] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. A land spatial survey data acquisition and monitoring system based on artificial intelligence, characterized in that, The system includes the following modules: The image acquisition module is used to acquire monitoring images of the national land space; The image analysis module is used to obtain different region units based on the color distribution characteristics of the monitored image; to obtain the decomposition necessity of the region units based on the difference characteristics of the texture intensity distribution in the region units; to determine whether to downsample based on the decomposition necessity; and to downsample the region units according to a preset minimum step size to obtain an initial downsampled image. The step size adjustment coefficient is obtained based on the difference in decomposition necessity between the initial downsampled image and the corresponding downsampled object; the adaptive step size is obtained based on the step size adjustment coefficient. The initial downsampled image is iteratively downsampled again based on the adaptive step size to obtain sub-images of different scales; The feature processing module is used to obtain the size evaluation feature value of the convolution kernel based on the downsampling order feature of the sub-image, the decomposition necessity and the number of pixels. The preset convolutional kernel size is adaptively adjusted based on the size evaluation feature value to obtain an adaptive size; The spatial modeling module is used to obtain the features of the sub-image based on the convolutional neural network and the adaptive size, and to perform feature modeling of the national land space. The step of obtaining the decomposition necessity of a region unit based on the difference in texture intensity distribution within the region unit includes: In the formula, S represents the decomposition necessity degree of the region unit. This indicates the preset weight, and N represents the number of pixels within the region unit. Linear normalization is represented; the maximum color difference between a pixel in the region unit and other pixels within a preset neighborhood is calculated to obtain the local color difference feature value of the pixel; This represents the local color difference feature value of the nth pixel. This represents the average value of the color difference feature of pixels within a region unit. This represents the upper limit of the color difference feature value; The step of obtaining the convolution kernel size evaluation feature value based on the downsampling order feature of the sub-image, the decomposition necessity feature, and the pixel count feature includes: In the formula This represents the size evaluation feature value corresponding to the q-th sub-image. This indicates the preset first weighting factor. This indicates a preset second weighting factor. This indicates a preset third weighting factor. This indicates the downsampling order corresponding to the q-th sub-image, and T represents the number of sub-images. Let represent the decomposition necessity of the q-th sub-image, and N represent the number of pixels in the region unit. This represents the number of pixels in the q-th sub-image; The step of adaptively adjusting the preset convolutional kernel size based on the size evaluation feature value to obtain the adaptive size includes: In the formula This represents the adaptive size of the convolution kernel corresponding to the q-th sub-image. This represents the function that takes the nearest odd number. This indicates the minimum size of the preset convolution kernel. This indicates the maximum preset kernel size. This represents the size evaluation feature value of the q-th sub-image.

2. The artificial intelligence-based land space survey data acquisition and monitoring system according to claim 1, characterized in that, The step of obtaining different region units based on the color distribution characteristics of the monitored image includes: The monitoring image is smoothed by mean filtering to obtain a smoothed image; a feature vector is constructed based on the coordinates and color of each pixel in the smoothed image; the pixels are clustered based on the differences in the feature vectors to obtain different clusters; the range of pixels in the same cluster in the monitoring image is taken as a region unit.

3. The artificial intelligence-based land space survey data acquisition and monitoring system according to claim 1, characterized in that, The step of determining whether to downsample based on the decomposition necessity includes: The region units whose decomposition necessity exceeds a preset threshold are downsampled.

4. The artificial intelligence-based land space survey data acquisition and monitoring system according to claim 1, characterized in that, The step of obtaining the step size adjustment coefficient based on the difference in decomposition necessity between the initial downsampled image and the corresponding downsampled object includes: In the formula, W represents the step size adjustment coefficient of the initial downsampled image. This indicates the decomposition necessity of the initial downsampled image. This indicates the decomposition necessity of the downsampled image corresponding to the initial downsampled image.

5. The artificial intelligence-based land space survey data acquisition and monitoring system according to claim 1, characterized in that, The step of obtaining the adaptive step size based on the step size adjustment coefficient includes: In the formula, L represents the adaptive step size for the next downsampling. This indicates the preset minimum step size. This indicates the preset maximum step size, and W represents the step size adjustment coefficient. This indicates rounding down to the nearest integer.

6. The artificial intelligence-based land space survey data acquisition and monitoring system according to claim 1, characterized in that, The step of iteratively downsampling the initial downsampled image according to the adaptive step size to obtain sub-images of different scales includes: The initial downsampled image is downsampled according to the adaptive step size to obtain a new downsampled image; the adaptive step size of the new downsampled image is obtained cyclically and downsampling continues until the decomposition necessity of the finally obtained downsampled image does not exceed a preset threshold; all images obtained by the region unit during the downsampling process are used as sub-images of different scales.

Images