Intelligent identification method and system for idle rural houses based on multi-source data fusion

Abstract

The present application relates to the field of image recognition, and more particularly to a kind of idle intelligent discrimination method and system of rural house based on multi-source data fusion.The method comprises the following steps: obtaining multi-source remote sensing satellite image, multi-time point fitting is carried out, and multi-temporal image set is constructed;Multi-temporal image set is carried out rural house target identification and fine profile segmentation, and multiple rural house image frame is marked;Based on multiple rural house image frame, time sequence tracking segmentation is carried out, and the multi-temporal image frame of each image frame is obtained;Based on the multi-temporal image frame, time sequence idle evolution mining is carried out, and the idle evolution track of each rural house is generated;According to the idle evolution track, idle state trend analysis is carried out, and intelligent discrimination is carried out, and idle house recognition result is obtained.The present application realizes the fast, accurate identification idle rural house, improves rural house resource management efficiency.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of image recognition, and in particular to a method and system for intelligent identification of idle rural houses based on multi-source data fusion. BACKGROUND

[0002] Compared with traditional rural house investigation methods, remote sensing image recognition technology has the advantages of high efficiency, wide range, non-contact, periodic updating, etc., and can realize automatic extraction and change analysis of rural house space state in a large scale. Especially under the background of continuous enrichment of multi-source remote sensing data, by fusing multi-type and multi-temporal image data such as optical satellite images, high-resolution images, multi-spectral images and historical time series data, not only can the rural house space form information be more comprehensively obtained, but also the time series change rule can be more deeply mined, thereby providing more reliable technical support for rural house use condition identification.

[0003] However, there are still many technical bottlenecks in the field of idle rural house identification. On the one hand, traditional methods mostly rely on manual field investigation or image classification based on a single time point, and the judgment result is easily affected by subjective experience and data timeliness, making it difficult to realize dynamic updating and large-scale promotion. On the other hand, existing image recognition models mostly only target detection or single temporal classification, lack the ability of time series modeling of rural house state evolution process, and are difficult to cope with the real characteristics of complex rural house use state, slow change and easy confusion. In addition, the problems of various rural house forms, complex roof materials and easily disturbed surrounding environment also bring great challenges to traditional image analysis methods. SUMMARY

[0004] To solve the above technical problems, the present application provides a method and system for intelligent identification of idle rural houses based on multi-source data fusion to solve at least one of the above technical problems.

[0005] To achieve the above purpose, the present application provides a method for intelligent identification of idle rural houses based on multi-source data fusion, comprising the following steps:

[0006] Step S1: acquiring multi-source remote sensing satellite images, performing multi-time point fitting, and constructing a multi-temporal image set;

[0007] Step S2: rural house target recognition and fine contour segmentation are performed on the multi-temporal image set, and a plurality of rural house image frames are marked;

[0008] Step S3: time series tracking segmentation is performed based on the plurality of rural house image frames, and a plurality of time series image frames of each image frame are obtained;

[0009] Step S4: time series idle evolution mining is performed based on the plurality of time series image frames, and an idle evolution trajectory of each rural house is generated;

[0010] Step S5: according to the idle evolution track, idle state trend analysis is carried out, and intelligent discrimination is carried out, and idle house recognition result is obtained.

[0011] In the present specification, an idle intelligent discrimination system for rural houses based on multi-source data fusion is provided, which is used to execute the idle intelligent discrimination method for rural houses based on multi-source data fusion as described above, comprising:

[0012] An image processing module is configured to acquire multi-source remote sensing satellite images, perform multi-time point fitting, and construct a multi-time image set.

[0013] A target recognition module is configured to perform rural house target recognition and fine contour segmentation on the multi-time image set, and mark a plurality of rural house image frames.

[0014] A template segmentation module is configured to perform time sequence tracking segmentation based on the plurality of rural house image frames, and obtain a plurality of time sequence image frames of each image frame.

[0015] An idle evolution analysis module is configured to perform time sequence idle evolution mining based on the plurality of time sequence image frames, and generate an idle evolution track of each rural house.

[0016] An intelligent discrimination module is configured to perform idle state trend analysis according to the idle evolution track, and perform intelligent discrimination, and obtain idle house recognition result.

[0017] The beneficial effects of the present application are as follows: by fusing high-resolution optical images, panchromatic images and multispectral images, not only the spatial resolution and the recognizable property of ground objects are improved, but also the expression ability of the physical properties of ground objects is enriched, which is helpful for fine extraction of rural house shape, roof material, use traces and other details. The multi-time image set formed by multi-time point fitting realizes time continuity modeling of the rural house state, so that the model has the ability to capture the "change trend" rather than "single point judgment", thereby significantly improving the timeliness and accuracy of the use state recognition of the rural house. By target detection and contour segmentation in the multi-time image, the system can automatically mark the position, size, shape, orientation and other information of each rural house, avoiding the subjective bias and low efficiency problems caused by manual drawing. The introduction of fine contour segmentation greatly improves the accuracy of house boundary recognition, solving the problem that the traditional rough frame selection cannot deal with complex scenes such as dense distribution of rural houses and irregular shape. This not only improves the positioning accuracy of individual houses, but also provides accurate spatial anchor points for subsequent change tracking and state evolution.

[0018] By tracking and segmenting the rural houses in each image frame over time, it can be ensured that the same house is represented in each time-series image frame, avoiding target mismatch, recognition drift and other problems, greatly improving the continuity and reliability of time-series analysis. In addition, the construction of time-series image frames provides each house with a complete chain of time information, allowing the system to accurately identify changes in the performance of the house in different years, seasons, and even imaging conditions, providing a basis for subsequent trajectory analysis. By analyzing key indicators such as brightness, texture, vegetation coverage, and structural changes in multi-temporal image frames, the system can extract the evolution trend of house usage in the time dimension. For example, a continuous decrease in roof reflectivity and an increase in surrounding vegetation over the years may indicate "gradual idleness"; if the roof color suddenly changes or the structure disappears, it may be "sudden demolition" or "complete abandonment". Such dynamic trajectories not only provide more comprehensive evidence for identifying the current state, but also reflect the impact of government policies, rural population flow, and land use adjustment on rural house usage.

[0019] The results of the entire data processing chain are converted into decision-making outputs, realizing an intelligent leap from "image information" to "behavior identification". Through trend analysis of the evolution trajectory of rural houses, the system can not only identify whether it is currently idle, but also determine the degree of idleness (such as temporary idleness, long-term idleness, and imminent idleness) and the idleness mode (such as intermittent use and structural damage). The intelligent identification part can use machine learning classifiers (such as decision trees and SVM) or more advanced time-series models (such as LSTM and Transformer) for state prediction, making the identification results interpretable and quantifiable. The identified idle houses are output in the form of spatial distribution maps and attribute tables, serving rural housing resource management, policy implementation, and grassroots governance. BRIEF DESCRIPTION OF DRAWINGS

[0020] Figure 1 is a step flowchart of the idle intelligent identification method for rural houses based on multi-source data fusion of the present application;

[0021] Figure 2 is a detailed implementation step flowchart of step S1;

[0022] Figure 3 is a detailed implementation step flowchart of step S2;

[0023] Figure 4 is a detailed implementation step flowchart of step S3. DETAILED DESCRIPTION

[0024] It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application.

[0025] The application example provides a kind of idle intelligent discrimination method and system of rural house based on multi-source data fusion.The execution subject of the idle intelligent discrimination method and system of rural house based on multi-source data fusion includes but is not limited to the following: mechanical equipment, data processing platform, cloud server node, network upload device and the like can be regarded as the general computing node of the application, the data processing platform includes but is not limited to: audio image management system, information management system, cloud data management system at least one.

[0026] Please refer to Figures 1 to 4 The application provides a method, comprising the following steps:

[0027] Step S1: obtain multi-source remote sensing satellite image, carry out multi-time point fitting, and construct multi-time image set;

[0028] Step S2: rural house target identification and fine contour segmentation are carried out to multi-time image set, and multiple rural house image frames are marked;

[0029] Step S3: time sequence tracking segmentation is carried out based on multiple rural house image frames, and the multi-time image frame of each image frame is obtained;

[0030] Step S4: time sequence idle evolution mining is carried out based on the multi-time image frame, and the idle evolution track of each rural house is generated;

[0031] Step S5: idle state trend analysis is carried out according to the idle evolution track, and intelligent discrimination is carried out, and idle house recognition result is obtained.

[0032] In the embodiment of the application, Figure 1 It is a kind of idle intelligent discrimination method of rural house based on multi-source data fusion of the application, and the steps of the idle intelligent discrimination method of rural house based on multi-source data fusion include:

[0033] Step S1: obtain multi-source remote sensing satellite image, carry out multi-time point fitting, and construct multi-time image set;

[0034] In this embodiment, multi-source remote sensing satellite image data is acquired for the study area (such as a target rural area) to ensure that the needs of spatial details, spectral information, and time coverage can be met for idle rural house discrimination. The multi-source data usually includes high-resolution optical images, multi-spectral images, and panchromatic images, among which the high-resolution optical images can clearly depict the shape and structure of the house, the multi-spectral images can provide the spectral reflection characteristics of vegetation, water, and other objects, and the panchromatic images have higher resolution in spatial details. When acquiring the data, it is necessary to ensure that the time span covers the target analysis period, for example, 2-3 years of continuous data acquisition every quarter or every month. In the data preprocessing stage, the images need to be radiometrically calibrated, atmospherically corrected, and geometrically precisely corrected to eliminate the brightness differences and geometric distortions between different time images. Subsequently, the images at each time point are aligned according to the spatial registration results, and a multi-time image set that is time-continuous and spatially consistent is generated through multi-time fitting methods (such as time series image fusion, interpolation reconstruction).

[0035] Step S2: rural house target recognition and fine contour segmentation are performed on the multi-time image set, and a plurality of rural house image frames are marked;

[0036] In this embodiment, after constructing the multi-time image set, it is necessary to identify the location of the rural house and accurately obtain its contour range. In specific implementation, a deep learning target detection algorithm (such as the YOLO series or Faster R-CNN) can be used to train a large number of labeled rural house samples, so that the model can automatically identify rural houses of different shapes and different roof materials. The output result of the detection is a rectangular boundary box (image frame) of the rural house, and each boundary box contains a coordinate position and a confidence score. To improve the spatial accuracy, fine contour segmentation needs to be performed on the detection result, and a semantic segmentation network (such as U-Net, DeepLab) can be used to perform pixel-level segmentation on the rural house area to obtain accurate house boundary lines. In the experiment, an optical image with a spatial resolution better than 1 meter can be selected as the input to ensure that the roof structure features are clearly visible. At the same time, the multiple detection results of the same rural house in different time images need to be processed to remove duplicates to avoid repeated marking. The final output result is a rural house target set containing spatial position, boundary shape, and confidence information, covering all identified rural houses in the study area.

[0037] Step S3: time series tracking segmentation is performed based on the plurality of rural house image frames to obtain multi-time image frames of each image frame;

[0038] In this embodiment, after obtaining the image frames of multiple farmhouses, the positions of these frames in the multi-temporal image set need to be corresponded and tracked, so as to extract the image frames of the same farmhouse at different times. The core of time sequence tracking segmentation is to establish the correspondence between spatial position and time sequence. Common methods include matching based on center point position, IoU (intersection over union) overlap detection, optical flow method or feature point matching, etc. In the experiment, the method combining the center point position and area stability can be used preferentially, that is, when the distance between the farmhouse detection frame at a certain time point and the center point at the previous moment is within 2 pixels and the area change is not more than 20%, it is determined as the same target. After time sequence tracking, different time image frames of the same farmhouse are cropped out to form a set of image frames in time sequence. This process also needs to handle the problems of occlusion and missing, for example, when the image is covered by clouds or the shadow affects the recognition effect, the sequence continuity can be maintained by interpolation or frame filling. Finally, each farmhouse will correspond to a multi-temporal image frame sequence arranged in time sequence, which provides direct input for subsequent state change analysis.

[0039] Step S4: time sequence idle evolution mining based on the multi-temporal image frames is performed to generate an idle evolution track of each farmhouse;

[0040] In this embodiment, after obtaining the multi-temporal image frames of each farmhouse, the next step is to analyze their time change characteristics to determine whether there are signs of long-term idleness. Time sequence idle evolution mining usually includes three aspects: brightness change analysis, texture feature change analysis and surrounding environment activity change analysis. Brightness change can reflect information such as roof material aging and illumination reflection change; texture change can capture changes in house surface structure or traces of human activities in the environment; and surrounding environment activity reflects changes in farming, vehicle access, etc. In the experiment, the average brightness, gray level co-occurrence matrix features and vegetation index of each temporal image frame can be calculated, and then the time sequence fitting and change trend calculation of these indexes are performed. If the change amplitude of a farmhouse is extremely low and lacks use traces for a long time, it may present idle characteristics. By concatenating the analysis results at each time point, an idle evolution track of each farmhouse can be generated, which is a curve or data sequence reflecting the change of house state over time.

[0041] Step S5: idle state trend analysis is performed according to the idle evolution track, and intelligent discrimination is performed to obtain an idle house recognition result.

[0042] In this embodiment, trend analysis is performed on the idle evolution track, and the intelligent discrimination model is combined to finally output the idle house recognition result. The goal of trend analysis is to extract long-term patterns from the evolution track and distinguish between short-term non-use and true long-term idle. In the experiment, the track can be analyzed by sliding by quarter or half a year, and the idle trend intensity of each time period is calculated. If a farm house maintains a high idle trend intensity in consecutive time periods and the trend does not decrease significantly, it can be determined as a high-probability idle house. In order to further improve the discrimination accuracy, a machine learning classifier (such as random forest, support vector machine or lightweight neural network) can be introduced, and the brightness change rate, texture stability, surrounding activity change and the like are input into the model as features, and the idle and non-idle classification results are output. The final result will be accompanied by a confidence score, combined with the preset time threshold to filter out the truly idle houses that meet the determination conditions. The output dataset will include house ID, spatial location, idle duration, idle trend and the like, which can be directly used for subsequent visualization and management.

[0043] In this embodiment, refer to Figure 2 For a detailed implementation step flowchart of step S1, in this embodiment, the detailed implementation steps of step S1 include:

[0044] Obtain multi-source remote sensing satellite images of different periods of the target rural area, wherein the multi-source remote sensing satellite images include high-resolution optical images, multispectral images and panchromatic images.

[0045] Calculate the pixel gray value of the multi-source remote sensing satellite images to obtain a gray histogram.

[0046] Perform gray value distribution stretching processing on the gray histogram to obtain a stretched gray density function.

[0047] Perform global histogram equalization processing on the multi-source remote sensing satellite images according to the stretched gray density function to construct a brightness-enhanced image.

[0048] Perform multi-time point fitting on the brightness-enhanced image to construct a multi-time image set.

[0049] In this example, multi-source remote sensing satellite data with time span is acquired for the study area (e.g. a typical rural area in a province) to meet the demand for dynamic discrimination of idle rural housing. By "multi-source", it means using image data from different satellite platforms or sensors at the same time, such as high-resolution optical images (e.g. GF-2, WorldView-3 provides 0.5 m or higher resolution visible light data), multispectral images (e.g. Sentinel-2 MSI provides 1-3 bands, 10-60 m resolution multispectral information), and panchromatic images (e.g. 0.5 m level panchromatic data of Gaofen-1 or IKONOS). The time span is recommended to cover at least 2-3 years of multiple key agricultural time nodes (such as spring ploughing, summer harvest, and post-harvest), in order to observe seasonal and annual trends. When acquiring images, attention should be paid to cloud control (usually required to be less than 10%), and images of the same area in the same season or similar months should be selected as much as possible to reduce the interference of seasonal vegetation changes on the results. Image download sources can include China Resources Satellite Application Center (CRESDA), ESA Copernicus Data Center, and commercial data suppliers. After data acquisition, the projection coordinate system (such as WGS-84 / UTM 50N) needs to be unified and geometrically corrected to ensure that multi-temporal images are accurately aligned in space. This stage provides basic data guarantee for subsequent gray level calculation and brightness enhancement. The gray value distribution information of the pixels is extracted from the image. For optical and multispectral images, band selection and radiation calibration are needed to unify the original values obtained from different sources and different sensors, so that they can be compared within the same gray scale range. For example, for 8-bit images, the gray value range is 0 to 255; while 16-bit images need to be scaled or normalized to this range. Multispectral images usually calculate the gray value for each band independently, while panchromatic images directly use the brightness value of a single band. Then, by counting the number of pixels corresponding to each gray value in the image, a gray histogram can be generated. The horizontal axis of the histogram represents the gray level, and the vertical axis represents the number of pixels or frequency. This statistical process can be performed on the entire image, or it can be calculated in multiple regions or blocks, such as 256x256 pixels per unit, which can realize local brightness enhancement in subsequent processing. By observing the histogram, the brightness distribution of the image can be intuitively understood, for example, farmland and vegetation areas in rural areas are usually concentrated in the dark gray interval, while roofs, roads and other hard objects are usually concentrated in the bright gray interval.

[0050] The purpose of the gray value distribution stretching is to expand the brightness dynamic range of the image, so that the image with insufficient contrast is visually clearer and the different ground objects are more easily distinguished in calculation. First, analyze the gray histogram to find the two boundary values of the lower gray value and the higher gray value, raise the pixel value below the lower limit to the darkest end, compress the pixel value above the upper limit to the brightest end, and the pixel value in the middle part is uniformly distributed to the entire available gray range according to the proportion. For example, if a large number of pixels in the image are concentrated between gray values 40 and 210, and the parts below 40 and above 210 are very small, then 40 can be taken as the starting point of the dark part, 210 as the end point of the bright part, and the middle pixels are uniformly stretched to the entire brightness interval. In multispectral image processing, this operation is usually performed on each band separately to prevent color distortion; in the panchromatic image, the single-band brightness value is directly stretched. In the experiment, segmented stretching or nonlinear stretching can also be used to enhance the detail performance of the dark part or the bright part. After stretching, the brightness distribution of the image is more uniform, and the contrast between ground objects is more obvious, providing a more suitable input for the next step of histogram equalization. The core purpose of global histogram equalization is to enhance the overall contrast of the image, so that the number of pixels in different gray value ranges tends to be balanced, thereby enhancing the detail performance. According to the gray value distribution after stretching, the brightness value of the image is redistributed, so that the dark part is enhanced, the bright part is compressed, and the overall brightness transition is smoother. The specific method is to first calculate the frequency of each gray value in the image, then according to the cumulative situation of these frequencies, map the original brightness value to the new brightness level, so that the brightness distribution of the output image is closer to uniform distribution. In multispectral image processing, to avoid color deviation, the image can be converted to a color space (such as HSI or Lab) first, only the brightness component is equalized, and then combined with the original chroma component to return. Compared with local equalization, global equalization is suitable for large areas with uniform ground object distribution in rural areas, and can enhance the overall contrast without the phenomenon of local over-enhancement. After global equalization, the brightness difference between different ground objects such as house roofs, roads, farmland, and vegetation is clearer, and the boundaries are easier to identify, providing a more reliable visual and data basis for the idle discrimination of rural houses based on temporal changes.

[0051] The brightness-enhanced images at different time points are time-sequenced to form a multi-temporal image set for analyzing the change trend of the farmhouses over time. First, the spatial registration accuracy of the images is ensured to completely align the images at the pixel level, which usually requires a registration error of no more than one pixel. In the registration process, a feature point matching-based algorithm or a cross-correlation method based on image block similarity can be used. After completing the spatial registration, radiation normalization is also needed to eliminate the brightness difference between different time images, which can be achieved by selecting some targets with stable reflection characteristics (such as roads, cement ground) as a reference. Subsequently, a time series can be established for each pixel position to record the brightness change of the pixel at different time points, and the stable change trend can be extracted through smoothing or fitting methods (such as trend analysis, sliding window average). The final multi-temporal image set not only contains the enhanced images at each time point, but also generates a difference map reflecting the brightness change, which can be used to identify areas with small long-term brightness change, which may correspond to idle or long-term unused farmhouses. This result will be directly used as an important input data of the intelligent discrimination model for idle farmhouses.

[0052] In this embodiment, the steps for constructing the multi-temporal image set by fitting the brightness-enhanced images at multiple time points are as follows:

[0053] Atmospheric blur recognition is performed on the brightness-enhanced images to mark the atmospheric blur areas.

[0054] Dark target visual analysis is performed on the brightness-enhanced images to extract multiple image dark targets.

[0055] The digital pixel values of the image dark targets are calculated.

[0056] The digital pixel values are averaged and analyzed to obtain atmospheric scattering parameters.

[0057] Atmospheric scattering compensation is performed on the brightness-enhanced images according to the atmospheric scattering parameters to construct scattering compensation images.

[0058] Image geometric distortion recognition is performed on the scattering compensation images, and adaptive accuracy correction is performed to construct distortion correction images.

[0059] Multi-time point fitting is performed on the distortion correction images to construct a multi-temporal image set.

[0060] In this embodiment, after the image is subjected to brightness enhancement processing, it can still be affected by atmospheric conditions, especially in multi-source and multi-temporal data, different shooting times and weather conditions can cause blurring phenomena caused by water vapor and suspended particles in the air, which can reduce the sharpness and contrast of the image, thereby affecting the subsequent analysis accuracy. By identifying and marking the areas affected by atmospheric blurring, targeted compensation can be performed in subsequent processing. The identification process can first detect the blurring degree through spatial frequency analysis methods, such as calculating the local sharpness index (such as the sharpness score based on gradient amplitude) after high-pass filtering the image, and areas below the set threshold are determined as blurred areas. In addition, contrast analysis methods can also be combined to calculate the brightness variation range of the local area, and if the variation range is too small, it may be a smoothing effect caused by blurring. The threshold setting can be dynamically adjusted according to the statistical results of the reference clear image, for example, in a 0.5-meter resolution high-resolution image, if the local contrast is less than 70% of the normal value, it is marked as blurred. The marking method usually generates a binary mask image, with white representing blurred areas and black representing clear areas. This mask will be used as a reference weight in the subsequent atmospheric scattering compensation process, so that stronger compensation and correction can be performed in the blurred areas.

[0061] Atmospheric scattering compensation requires reference to "dark target" information in the image. Dark targets refer to ground objects with extremely low reflectivity and theoretical brightness close to zero, such as deep water areas, shadow areas, dense vegetation shadows, etc. In optical remote sensing images, these areas have very low brightness under normal circumstances, but due to atmospheric scattering, they still show a certain brightness, and this brightness offset is one of the characteristics of atmospheric scattering. Therefore, dark target analysis is needed for brightness-enhanced images. This includes a combination of manual experience and automatic detection, such as first using a brightness threshold method to select pixels with brightness values in the lowest 1% or 2% range as preliminary dark target candidate areas, and then removing small noise points through morphological filtering. To avoid confusion with shadows and artificial buildings, multi-spectral band information can be combined to exclude shadow artifacts of high-reflectivity materials. For rural area images with a resolution of 0.5 meters, at least tens of dark targets distributed in different locations are extracted from each image to ensure the spatial representativeness of the sample distribution. These dark targets will be used to estimate atmospheric scattering parameters in subsequent processing, so their reflectivity stability and location diversity need to be ensured.

[0062] After the extraction of dark target regions, the digital pixel values of these regions need to be measured accurately. The actual brightness offset values of these pixels, which should theoretically be close to zero brightness, are quantified. During the calculation process, the pixels of each dark target region are first extracted individually, and their gray scale values or radiant brightness values in the image are recorded. If it is a multispectral image, statistics need to be taken for different wavebands respectively, because the degree of influence of atmospheric scattering is different in different wavebands, for example, the scattering influence of short wave band is usually more significant than that of long wave band. To reduce the influence of individual abnormal pixels, a 3x3 or 5x5 sliding window smoothing calculation can be performed on the dark target regions first, and the median of the pixel values in the window is taken as the representative value. The pixel values of all dark targets will form a data set for the next step of average and distribution analysis.

[0063] After obtaining the dark target pixel value data set, statistical analysis needs to be performed on these values to estimate the strength and distribution characteristics of atmospheric scattering. First, the average brightness value of all dark targets is calculated, which represents the overall brightness offset introduced by atmospheric scattering in the image. Next, distribution analysis is performed to calculate indicators such as standard deviation and range of brightness values to determine the uniformity of scattering effects in different regions. If the standard deviation is large, it indicates that the influence of atmospheric scattering is not uniform within the image, and a spatially variable correction coefficient needs to be used in the compensation process; if the standard deviation is small, a globally uniform compensation coefficient can be used. The dark targets distributed in different directions of the image are usually analyzed separately to detect differences in local atmospheric conditions, for example, valleys and open plains may have different scattering characteristics. The atmospheric scattering parameters obtained in this way will be directly used to compensate the image for atmospheric scattering, thereby restoring brightness values closer to the true surface reflectance. The purpose of atmospheric scattering compensation is to remove the offset in the image brightness caused by the atmosphere, making the image closer to the true surface reflectance characteristics. According to the atmospheric scattering parameters obtained in the previous step, the brightness values of each pixel in the image can be corrected. The specific method is to subtract the atmospheric brightness offset reflected by the dark targets from all pixels in the image, and if the scattering parameters in different regions are significantly different, different compensation values are applied according to the spatial position. In multispectral images, the compensation values for different wavebands may be different, so compensation parameters need to be calculated and applied for each waveband respectively. After compensation, a comparison and verification will be performed immediately, by comparing with images taken under clear weather without clouds, to confirm whether the compensation effect is reasonable. The image after compensation processing will have a significant improvement in overall brightness and contrast, especially in long-distance or low-contrast areas, the boundaries of the surface features will be clearer, providing a higher quality data basis for geometric correction and subsequent analysis.

[0064] Geometric distortion is a common problem in remote sensing image processing, which may be caused by sensor imaging geometry, satellite attitude changes, and terrain undulations. After the completion of the scattering compensation, it is necessary to further check whether the image has geometric distortion, especially in multi-temporal contrast analysis. Any pixel-level spatial misplacement will cause the result deviation. The methods to identify geometric distortion include comparison with ground control points (GCP), pixel-level matching with high-precision reference images, and feature point matching algorithm to detect non-linear deformation. After detecting the distortion, the adaptive accuracy correction technology is adopted, and the appropriate geometric transformation model is selected according to the type and distribution of the distortion, such as polynomial transformation, affine transformation or local thin plate spline transformation. The registration accuracy after correction is required to be better than one pixel, and for 0.5-meter resolution images, the spatial error is not more than 0.5 meters. After geometric distortion correction, the image will have higher spatial accuracy and comparability with other temporal data. After atmospheric compensation and geometric correction, the images of different time phases need to be fitted and integrated to form a multi-temporal image set. This process requires the images to be highly consistent in spatial position and radiation characteristics to facilitate time series analysis. First, sort the images of each time phase according to the shooting time and ensure that all images are completely overlapped in space. In the time fitting process, the sliding time window method can be used to smooth the brightness changes of a pixel at different times, reducing the interference of random noise on the trend. For rural residential idle discrimination, the brightness change curve of the building area can be analyzed, and the area with no obvious change over a long period of time may correspond to the idle state. The final multi-temporal image set contains high-quality images of each time phase and also includes brightness change trend data, providing direct input support for intelligent analysis algorithms.

[0065] In this embodiment, referring to Figure 3 , the detailed implementation steps of step S2 include:

[0066] Pixel-level visual recognition is performed on the multi-temporal image set, and soil utilization classification is performed to obtain the soil type in the region;

[0067] The land types in the region include farmland, residential land, transportation road, and water body vegetation;

[0068] Residential land boundary detection is performed according to the soil type in the region, and a residential land boundary region is extracted;

[0069] Edge noise filtering is performed on the residential land boundary region to obtain a filtered and optimized land boundary region;

[0070] Farmhouse target recognition is performed on the filtered and optimized land boundary region, and fine contour segmentation is performed to mark multiple farmhouse image frames.

[0071] In this embodiment, a pixel-level visual recognition method is used to combine spectral, texture, and temporal information for accurate differentiation of different ground objects. Machine learning algorithms based on supervised classification can be selected, such as support vector machine (SVM), random forest (RF), or deep learning convolutional neural network (CNN), especially 3D-CNN with multi-temporal feature input can effectively utilize time dimension information. The collection of training samples should cover all expected classification types, including farmland, residential land, transportation road, water body, and vegetation, etc. The number of samples for each class should be as balanced as possible, and distributed in different locations of the study area to ensure the generalization ability of classification. The classification input features include not only the multispectral reflectance values of single-period images, but also the derived features such as normalized difference vegetation index (NDVI), normalized water body index (NDWI), and image texture features (such as contrast and entropy of gray level co-occurrence matrix GLCM). The evaluation of classification accuracy can use confusion matrix and Kappa coefficient, and the overall accuracy should not be less than 85% to ensure the reliability of subsequent residential land extraction. After land use classification is completed, residential land categories can be selected from the classification results, and boundary detection can be performed to determine the exact range of residential land. The pixels marked as residential land in the classification map are extracted to form a binary mask, where residential land pixels are marked as 1 and non-residential land pixels are marked as 0. Since the classification results may contain noise or small areas of misclassified regions, morphological processing is needed, such as opening operation to remove small isolated pixel blocks and closing operation to fill small holes in residential land. Then, connected component analysis method is used to identify continuous residential land blocks, and small noise areas are filtered out according to the area threshold. For 0.5-meter resolution images, the minimum residential land area can be set to 50 square meters to avoid misidentifying single small structures or temporary sheds as residential land. Boundary detection can use algorithms based on contour extraction, such as Canny edge detection combined with contour tracking, to finally obtain the closed boundary polygon of each residential land.

[0072] Due to the influence of the complexity of ground features, image noise and lighting conditions in the image classification and boundary detection process, the extracted residential land boundary may have noise phenomena such as jaggedness, burr or local mispositioning. The boundary is smoothed and optimized to obtain a continuous, regular and more close-to-actual ground feature contour boundary line. The vertex simplification processing can be performed on the boundary polygon, for example, the Douglas-Peucker algorithm is used to reduce the redundant nodes while retaining the main shape features. Secondly, the median filter or Gaussian smoothing is used to smooth the boundary line to eliminate the jitter caused by high-frequency noise. To avoid over-smoothing and loss of details, the smoothing radius is usually controlled within 1 to 3 pixels. In addition, the boundary optimization method based on region growing can be used to adjust the boundary locally in combination with the brightness gradient information of the original image, so that the boundary is more close to the edge of the actual building or land. Further, the specific location and shape of the farm house are identified in the region. The object-oriented method can be used for farm house target recognition, that is, the image inside the residential land is first segmented into a plurality of image objects with similar spectral and texture features, and then classified based on shape, color and texture features. In the experiment, the farm house usually shows regular geometric shape, high brightness (strong roof reflection in panchromatic image) and low vegetation index value, which can be used as a basis for discrimination. If a deep learning method is used, a detection model such as Faster R-CNN, YOLOv5, etc. can be selected and trained with labeled farm house samples. After detecting the farm house target, fine contour segmentation is required, which can use the active contour model (Snake model) or Graph Cut method based on edge detection to make the segmentation result as close as possible to the actual boundary of the farm house. Finally, each farm house identified is marked with a rectangular or polygonal frame, and its position, area and contour information are recorded.

[0073] In this embodiment, reference is made to Figure 4 For a detailed implementation step flowchart of step S3, in this embodiment, the detailed implementation steps of step S3 include:

[0074] Based on the multiple farm house image frames, a repeated frame selection image frame is identified.

[0075] The repeated frame selection image frame is subjected to non-maximum suppression to obtain multiple high-confidence image frames.

[0076] The spatial position coordinates of the high-confidence image frames are calculated.

[0077] Based on the spatial position coordinates, a time sequence tracking process is performed on a multi-temporal image set, and a same position time sequence segmentation is performed to obtain multiple time sequence image frames of each image frame.

[0078] In this embodiment, after completing the identification of farmhouses, there may be a large number of repeated frame selection areas in the preliminary detection results, that is, for the same farmhouse target, the detection model may give multiple overlapping detection boxes. This repetition is more common when using deep learning detection algorithms (such as YOLO, Faster R-CNN), because the model may detect the same target at different feature layers and different anchor box sizes. The goal of this step is to identify those repeated frame selection image boxes from multiple candidate boxes for subsequent de-duplication optimization. The specific method is first to group all detection boxes by category and detection confidence in the same image, and only the detection boxes of the farmhouse category are judged for repetition. Then, calculate the overlap (intersection over union, IOU) of any two detection boxes, if the IOU is greater than the set repetition judgment threshold (0.5 to 0.7 in the experiment, adjust according to the accuracy requirements of the detection task), it is considered that they belong to the same farmhouse repeated frame selection result. In this process, in order to reduce false positives, the center point distance and area difference of the detection box can be combined to avoid misjudging two adjacent independent farmhouses as repeated frames. The non-maximum suppression (NMS) method is used to select the optimal frame from multiple overlapping detection boxes. Non-maximum suppression is a commonly used post-processing method for target detection, its basic idea is: among multiple candidate boxes of the same target, keep the one with the highest confidence, remove other boxes with lower confidence and high overlap. The specific method is to sort the repeated boxes by detection confidence from high to low, first select the box with the highest confidence as the reserved box, then calculate the overlap (IOU) with the remaining boxes, if the IOU exceeds the set suppression threshold (0.5 or 0.6 in the experiment), the box is removed from the candidate set, repeat this process until all boxes are processed. In order to avoid suppressing the actual small target, the suppression threshold can be appropriately relaxed for small area boxes.

[0079] The pixel position parameters of each detection frame in the image are obtained, including the pixel row and column numbers of the top-left corner and the bottom-right corner. Then, the pixel coordinates are accurately converted into geographic coordinates or projection coordinates (such as WGS-84 latitude and longitude or UTM coordinates) using the geographic reference information of the image (usually provided in the metadata or external registration file of remote sensing data, including spatial resolution, projection coordinate system, and geographic transformation parameters). For high-resolution images with a resolution of 0.5 meters, each pixel represents a real ground area of 0.5x0.5 meters, so the accuracy of coordinate conversion can reach sub-meter level. At the same time, it is recommended to record the center point coordinates and boundary polygon coordinates of the detection frame, which is not only beneficial for position matching, but also convenient for subsequent time series segmentation and area change analysis. With the accurate spatial coordinates of the rural house detection frame, position matching and tracking analysis can be performed in a multi-temporal image set. This step first extracts the corresponding image area in each image based on the spatial position coordinates. This process can be achieved through spatial query, i.e., finding the pixel range containing the location in different time images. To ensure temporal consistency, it is necessary to ensure that all images are geometrically fully registered, with a spatial error of no more than one pixel. During the extraction process, the boundaries can be appropriately expanded based on the original detection frame, such as increasing 1-2 pixels in each direction, to ensure complete coverage of the rural house and its surrounding area. The extracted image areas are arranged in chronological order to form a multi-temporal image frame set for the rural house. These image frames can be used to analyze the brightness, texture, and shape changes of the rural house at different times, such as by comparing the image frames at different times to determine whether the rural house has been in a long-term unchanged state, thereby inferring its idle state.

[0080] In this embodiment, step S4 includes the following steps:

[0081] Based on the multi-temporal image frames, the spectral changes of the drying area, the footprints of the courtyard soil, and the degree of road wear in front of the door are analyzed to obtain the life trace features of each rural house;

[0082] The multi-frame calculation of the aging rate of the roof material is performed on the multi-temporal image frames to obtain the aging rate;

[0083] Based on the aging rate, the house maintenance frequency is calculated, and the resident attention level is evaluated to obtain the resident attention level evaluation value;

[0084] Based on the multi-temporal image frames, the green state around the rural house is generated to generate a vegetation time series index curve;

[0085] Based on the vegetation time series index curve, the resident attention level evaluation value, and the life trace features, the time series idle evolution is mined to generate the idle evolution trajectory of each rural house.

[0086] In this embodiment, in multi-temporal image frame analysis, life traces are an important basis for judging the use activity of farmhouses. The possible drying area in the image is located, which usually appears in the courtyard, roof or front yard. Using the reflectivity characteristics of visible light and near-infrared bands of multispectral images, the spectral change characteristics of drying objects (such as food, crops, clothes) in different time periods can be detected, for example, the reflectivity decreases in blue and red light bands, and the near-infrared reflectivity slightly rises. Then, the texture analysis and color change detection of the courtyard soil area are carried out, the texture feature extraction method based on gray level co-occurrence matrix is adopted, and the multi-temporal color index change comparison is adopted to identify the soil compaction, color darkening or texture smoothing phenomenon caused by frequent trampling. For the detection of road wear degree in front of the door, the brightness stability and roughness change of the road area are used to judge the situation where the vehicle or pedestrian flow frequency is high. In the experiment, a multi-temporal sequence with a time interval of 1 month can be set for detection, and at least one year of change data can be combined for stability analysis. Finally, the drying activity frequency, the significance of courtyard trampling traces, and the road wear change amplitude are integrated to form the life trace feature vector of the farmhouse, providing basic information for idle discrimination.

[0087] The aging rate of the roof material is an important indicator for evaluating the use and maintenance of the house. In multi-temporal image frames, by analyzing the spectral reflectivity change, color shift trend and texture degradation of the roof area, the aging speed of the roof can be calculated. First, the roof area boundary is extracted using high-resolution optical images, and the color mean, color saturation and brightness distribution characteristics of each frame of image are calculated. Material aging usually shows color fading, brightness decrease and surface roughness increase, such as tile darkening, metal roof losing reflectivity. The near-infrared and short-wave infrared bands in multispectral images can more sensitively reflect the structural changes of the material surface. In the experiment, the roof features at each time point are differentially calculated with the initial state, and the change curve is fitted according to the time sequence, so as to obtain the aging rate of the material (such as the annual brightness decrease ratio, color shift amount). If the aging rate is obviously high and lacks periodic recovery signs, it may mean that the house lacks regular maintenance. This result will be directly used for subsequent maintenance frequency calculation and inhabitant attention evaluation.

[0088] After obtaining the roof aging rate, the maintenance frequency of the house can be calculated. Specifically, the occurrence time of the roof maintenance event is identified by the mutation point of the material state in the multi-temporal sequence (such as significant improvement of brightness, color recovery, and surface texture smoothing). For example, in the continuous 12-month image, if the roof color is obviously recovered in the 7th month, and the spectral reflectance characteristics are close to the performance of new materials, it can be determined as a maintenance behavior. The number of maintenance events in the entire analysis period is counted, and the maintenance frequency index is obtained. Then, the maintenance frequency is combined with the aging rate to evaluate the attention degree of the household to the house: high maintenance frequency and low aging rate mean high attention degree of the household; otherwise, the attention degree is low. In the experiment, the attention degree can be divided into high (>0.8), medium (0.5~0.8), and low (<0.5) three levels, and the score is based on the normalized maintenance frequency and the reciprocal of the aging rate. The evaluation value of the attention degree of the household is not only an important reference for the idle judgment of the rural house, but also can be combined with the life trace features and the vegetation state to improve the discrimination accuracy.

[0089] The green state around the rural house can reflect the management and use of the environment around the house. Through multi-spectral images, the difference in vegetation reflection in the red and near-infrared bands can be used to analyze the green state. The specific method is as follows: first, extract the buffer zone of a certain range (such as a radius of 10 meters) around the rural house, and identify the vegetation area therein. Then, in each frame of multi-temporal image, the spectral index of the vegetation area is calculated, such as red reflectance and near-infrared reflectance, and a time series curve reflecting the vegetation vigor is generated. In the experiment, it is recommended to select data for more than 1 year to cover the complete growth period of the vegetation. For a house with someone living in it, the green around the house will usually show periodic pruning, reseeding, and other management signs, making the vegetation vigor curve relatively smooth and the peak cycle regular. For a long-term idle house, the vegetation around the house may show irregular rapid growth, wilting, or even being covered by weeds, with large fluctuations in the curve and lack of signs of human intervention. The generated vegetation time series index curve will be one of the important inputs for idle evolution trajectory analysis.

[0090] The vegetation time series index curve, the household attention evaluation value and the life trace feature are fused to carry out time series idle evolution mining. Specifically, the system will map the three types of features to the time axis, with month or quarter as the time unit, to construct a multi-dimensional time series feature matrix. Then, through trend analysis methods (such as sliding window average, time series clustering, change point detection), the change mode of the rural house state is identified. For example, if the life trace activity frequency decreases, the vegetation curve shows signs of loss of management, and the household attention is low for a long time, then the idle trend of the rural house is significantly increased. In the experiment, the threshold of idle trend intensity can be set, for example, when the comprehensive score is lower than 0.4 for 6 consecutive months, it is marked as a potential idle house. The final output of the idle evolution track will reflect the use-idle change process of each rural house in the entire analysis period, providing data support for subsequent idle discrimination and spatial distribution analysis.

[0091] In this embodiment, the specific steps of step S5 are:

[0092] The idle evolution track is decomposed into multiple time window track segments;

[0093] Idle state trend analysis is performed on the multiple time window track segments to generate an idle state trend;

[0094] Progressive idle trend enhancement analysis is performed on the idle state trend to mark potential idle houses;

[0095] Idle duration calculation is performed on the potential idle houses to obtain idle duration;

[0096] Comparative analysis is performed on the idle duration based on the preset idle house time threshold to obtain an idle house recognition result;

[0097] Based on the idle house recognition result, high-light visual rendering is performed to construct an idle distribution visualization view to complete the idle discrimination of the rural house.

[0098] In this embodiment, after the multi-temporal rural house image frame extraction is completed, each rural house corresponds to a "idle evolution track" that changes over time, which records the state characteristics of the rural house at different time nodes, such as brightness change, texture change, and use trace index. Since the use state change of the rural house may not occur uniformly, it is necessary to divide the long time sequence into multiple short time periods for analysis. This segmentation method is called multi-time window decomposition. Specifically, the time window length is set according to the time scale of the research task and the sampling frequency of the data, for example, every 3 months as a time window, or divided by quarter, half year, etc. When decomposing, it is necessary to ensure that each time window is independent and comparable, so a sliding time window method can be used, that is, there is a certain overlap (such as 1 month) between adjacent time windows, in order to capture the state change across time periods. For multi-temporal image data of 2-3 years, a common decomposition strategy is to set a 6-month time window and slide it forward by 12 months, which can cover seasonal changes and retain medium-term trend characteristics. After multi-time window decomposition, the state track of each rural house is divided into several track segments, and each track segment can be analyzed separately to capture the phased changes of idle and non-idle states in more detail. After obtaining multiple time window track segments, idle state trend analysis needs to be performed on each track segment to determine whether the rural house presents idle characteristics in this period. During the analysis process, multiple feature indicators can be integrated, such as the fluctuation amplitude of image brightness, the change of roof reflection characteristics, the activity of the surrounding environment (such as vegetation change, temporary cover appearance), etc. For typical characteristics of idle state, it is often characterized by small amplitude of brightness change, long-term stable texture characteristics, and lack of significant use trace change. The change rate of these characteristics can be calculated for each track segment, and then compared with the set idle state discrimination threshold, for example, when the brightness change is lower than a certain percentage and the texture change is lower than a certain fixed value, it is determined as idle trend. In order to improve the accuracy, consistency verification can be combined with the results of multiple time windows, that is, the same rural house presents idle trend in two or more consecutive time window track segments, then its idle state trend credibility is higher. Finally, each rural house will generate an idle state trend curve to describe its state evolution law in the entire research period.

[0099] In the gradual idle trend enhancement analysis, the continuity and enhancement of the state trend of the rural house in multiple time windows are tracked to filter out the rural houses that have a long-term idle tendency. Specifically, the idle trend intensity (i.e., the idle feature significance of each time window) of the rural house is cumulatively analyzed. If the idle trend intensity gradually increases on the time axis, the rural house is determined to be a potential idle house. An enhancement threshold can be set, for example, if the idle trend index increases or remains high and stable in three consecutive time windows, the rural house is marked as a potential idle house. This gradual analysis helps to distinguish between seasonal non-use and long-term idle, reducing misjudgment. After filtering through this step, the rural house will enter the next stage of idle time calculation.

[0100] The total idle time is obtained by adding up the lengths of the trajectory segments determined to have an idle trend. If there are discontinuities between the trajectory segments, these discontinuities should be removed or separately counted. To ensure calculation accuracy, the time unit should be consistent with the time window length, for example, if the time window is half a year, the idle time is calculated in half-year units; if a sliding time window is used, the overlapping part needs to be considered for time deduction. For a 3-year research period, rural houses with idle time exceeding 1 year and distributed in multiple time windows are more likely to be long-term idle objects. The calculation result of the idle time will be one of the important bases for the final idle house identification, and will be compared and analyzed with the time threshold. After calculating the idle time of each potential idle house, it needs to be compared with the preset time threshold to determine the final idle house identification result. This time threshold can be set according to the research target and policy requirements, for example, set to 12 months or 18 months, indicating that only the rural house with idle time exceeding the threshold is identified as an idle house. In the comparative analysis, all potential idle houses can be divided into "exceeding the threshold" and "not exceeding the threshold" two categories, and an identification label is generated for each rural house. To improve the robustness of the result, a confidence score can be introduced, which is calculated by combining the idle time and the idle trend intensity. Only the rural house with a score exceeding the double threshold is finally identified as idle. An idle house list containing information such as spatial location, idle time, and discrimination confidence will be generated, which can be directly used for decision-making and subsequent management measures by the management department. This step marks the completion of the entire idle intelligent identification process of rural houses based on multi-source remote sensing data.

[0101] In this embodiment, the specific steps for highlighting visualization rendering based on the idle house identification result and constructing idle distribution visualization view to complete the idle identification of rural houses are as follows:

[0102] Calculate the spatial coordinates of each idle house based on the idle house identification result.

[0103] Perform idle distribution analysis based on the spatial coordinates to obtain a regional idle distribution map.

[0104] The regional idle distribution map is highlighted and visualized, and an idle distribution visualization view is constructed to complete the idle discrimination of rural houses.

[0105] In this embodiment, after completing the idle house identification, a unique identification tag is established for each rural house determined to be idle. The pixel position of these rural houses in the image is converted into accurate geographic coordinates. Read the boundary box information of each idle house in the final detection image, including the pixel row and column numbers of the top left and bottom right corners, and the center point position of the detection box. Then, the pixel coordinates are converted into projected coordinates or geographic coordinates using the geographic registration information of remote sensing images (usually including spatial resolution, projection coordinate system, geographic transformation parameters, etc. metadata). For example, for high-resolution optical images with a spatial resolution of 0.5 meters, the center point pixel coordinates can be accurately converted to the actual position within 0.5 meters after geographic conversion. To improve accuracy, ortho-corrected image data can be used first, and small-range geometric error correction can be performed during the calculation process to ensure spatial consistency between different time images. Finally, each idle rural house will obtain a unique spatial coordinate (center point coordinate) and corresponding spatial range (polygon boundary), providing basic spatial data for the next step of idle distribution analysis. All coordinate points are imported into the geographic information system (GIS) platform or spatial analysis software, combined with the boundary data of the study area for spatial statistical analysis. Common methods include nearest neighbor distance analysis (judging the aggregation of distribution), kernel density estimation (generating continuous spatial density surface), spatial autocorrelation analysis (such as Moran's I coefficient, judging the spatial correlation of distribution), etc. If the study area is large and the number of houses is large, the kernel density estimation method can be used, with a search radius of 200 meters or 500 meters, to reflect the idle concentration degree of rural houses in the village or area. After analysis, the generated results can be formed into a regional idle distribution map through spatial interpolation or density mapping methods.

[0106] In the GIS platform or remote sensing visualization software, the results of kernel density analysis or distribution analysis are rendered by color grading according to the numerical interval, for example, the area with high idle density is represented by bright red or orange, the medium density is represented by yellow, and the low density is represented by light green, so as to highlight the hotspot area. Five rendering levels can be set, and the area with the highest density value of 20% is highlighted as the first level, and the actual rural house position point is superimposed on the map, so that the corresponding relationship between the specific rural house and the density hotspot can be seen at the same time. In order to further enhance the readability, the background elements such as village boundary, road and water body can be added, and zoom-in viewing of local details is supported. The finally generated visualization view can not only be interactively browsed on the computer side, but also can be exported into a static map or an online map service, which can be directly used in relevant departments, township planning personnel or scientific research reports, so as to complete the visualization presentation of the idle discrimination of rural houses.

[0107] In the embodiment, an idle intelligent discrimination system of rural houses based on multi-source data fusion is provided, which is used to execute the idle intelligent discrimination method of rural houses based on multi-source data fusion as described above, and includes:

[0108] An image processing module is configured to acquire multi-source remote sensing satellite images, perform multi-time point fitting, and construct a multi-time image set.

[0109] A target recognition module is configured to perform rural house target recognition and fine contour segmentation on the multi-time image set, and mark a plurality of rural house image frames.

[0110] A template segmentation module is configured to perform time sequence tracking segmentation based on the plurality of rural house image frames, and obtain a plurality of time sequence image frames of each image frame.

[0111] An idle evolution analysis module is configured to perform time sequence idle evolution mining based on the plurality of time sequence image frames, and generate an idle evolution track of each rural house.

[0112] An intelligent discrimination module is configured to perform idle state trend analysis and intelligent discrimination according to the idle evolution track, and obtain an idle house recognition result.

[0113] Therefore, no matter from which point, the embodiments should be regarded as exemplary and non-limiting, the scope of the present application is defined by the appended claims rather than the above description, and therefore all changes falling within the meaning and scope of the equivalent elements of the application file are intended to be included in the present application.

[0114] The foregoing merely illustrates the principles of the application and application of its principles. Various modifications and alterations to this implementation will occur to those skilled in the art. The scope of the application should be determined, however, by the following claims rather than by the embodiments shown.

Claims

1. A method for intelligently identifying idle rural houses based on multi-source data fusion, characterized in that, The method comprises the following steps: Step S1: acquiring multi-source remote sensing satellite images, performing multi-time point fitting, and constructing a multi-time image set; Step S2: performing rural house target recognition and fine contour segmentation on the multi-time image set, and marking a plurality of rural house image frames; Step S3: performing time sequence tracking segmentation based on the plurality of rural house image frames to obtain a plurality of time sequence image frames of each image frame; Step S4: performing time sequence idle evolution mining based on the plurality of time sequence image frames to generate an idle evolution track of each rural house; Step S5: performing idle state trend analysis according to the idle evolution track, and performing intelligent discrimination to obtain an idle house recognition result; The specific steps of step S4 are as follows: Based on the plurality of time sequence image frames, the spectral change of the drying area, the yard soil trampling trace, and the road wear degree in front of the door are analyzed to obtain the life trace features of each rural house; The aging rate of the roof material is calculated based on the plurality of time sequence image frames to obtain the aging rate; The house maintenance frequency is calculated according to the aging rate, and the attention degree of the residents is evaluated to obtain an attention degree evaluation value of the residents; Based on the plurality of time sequence image frames, the green state around the rural house is generated to generate a vegetation time sequence index curve; Based on the vegetation time sequence index curve, the attention degree evaluation value of the residents, and the life trace features, time sequence idle evolution mining is performed to generate an idle evolution track of each rural house; The specific steps of step S5 are as follows: The idle evolution track is decomposed into a plurality of time window track segments; Idle state trend analysis is performed on the plurality of time window track segments to generate an idle state trend; Progressive idle trend enhancement analysis is performed on the idle state trend to mark potential idle houses; Idle time length calculation is performed on the potential idle houses to obtain an idle time length; Based on a preset idle house time threshold, the idle time length is compared and analyzed to obtain an idle house recognition result; Based on the idle house recognition result, a highlight visual rendering is performed to construct an idle distribution visual view, thereby completing the idle discrimination of the rural house. 2.The idle intelligent discrimination method for rural houses based on multi-source data fusion according to claim 1, characterized in that, The specific steps of step S1 are as follows: Multi-source remote sensing satellite images of different time periods of a target rural area are acquired, and the multi-source remote sensing satellite images include high-resolution optical images, multi-spectral images, and panchromatic images; The pixel gray value of the multi-source remote sensing satellite images is calculated to obtain a gray histogram; The gray histogram is subjected to gray value distribution stretching processing to obtain a stretched gray density function; Global histogram equalization processing is performed on the multi-source remote sensing satellite images based on the stretched gray density function to construct a brightness enhanced image; Multi-time point fitting is performed on the brightness enhanced image to construct a multi-time image set. 3.The idle intelligent discrimination method of rural houses based on multi-source data fusion according to claim 2, characterized in that, The specific steps of performing multi-time point fitting on the brightness enhanced image to construct a multi-time image set are as follows: Atmospheric blur recognition is performed on the brightness enhanced image to mark an atmospheric blur area; Dark target visual analysis is performed on the brightness enhanced image to extract a plurality of image dark targets; The digital pixel value of the image dark target is calculated; The digital pixel value is subjected to average calculation and distribution analysis to obtain an atmospheric scattering parameter; Atmospheric scattering compensation is performed on the brightness enhanced image based on the atmospheric scattering parameter to construct a scattering compensation image; The scattered compensation image is subjected to image geometric distortion identification and adaptive precision correction to construct a distortion correction image; The distortion correction image is subjected to multi-time point fitting to construct a multi-time image set. 4.The idle intelligent discrimination method for rural houses based on multi-source data fusion according to claim 1, characterized in that, The specific steps of step S2 are as follows: The multi-time image set is subjected to pixel-level visual identification and soil body utilization classification to obtain the soil body types in the region; The soil types in the region include farmland, residential land, traffic road and water body vegetation; The residential land boundary is detected according to the soil body types in the region to extract a residential land boundary region; The residential land boundary region is subjected to edge noise filtering to obtain a filtered and optimized land boundary region; The filtered and optimized land boundary region is subjected to farmhouse target identification and fine contour segmentation to mark multiple farmhouse image frames. 5.The idle intelligent discrimination method for rural houses based on multi-source data fusion according to claim 1, characterized in that, The specific steps of step S3 are as follows: The multiple farmhouse image frames are subjected to repeated frame selection and image frame recognition; The repeated frame selection image frame is subjected to non-maximum suppression to obtain multiple high-confidence image frames; The spatial position coordinates of the high-confidence image frames are calculated; The multi-time image set is subjected to time sequence tracking processing based on the spatial position coordinates and the same position time sequence segmentation to obtain multiple time sequence image frames for each image frame. 6.The idle intelligent discrimination method for rural houses based on multi-source data fusion according to claim 1, characterized in that, The specific steps for completing the farmhouse idle discrimination job based on the idle house identification result are as follows: The spatial coordinates of each idle house are calculated based on the idle house identification result; The idle distribution analysis is performed based on the spatial coordinates to obtain a regional idle distribution map; The idle distribution visualization view is constructed by highlighting and visualizing the regional idle distribution map to complete the farmhouse idle discrimination job.

7. A rural house idle intelligent discrimination system based on multi-source data fusion, characterized in that, The specific steps for completing the farmhouse idle discrimination job based on the idle house identification result are as follows: The spatial coordinates of each idle house are calculated based on the idle house identification result; The idle distribution analysis is performed based on the spatial coordinates to obtain a regional idle distribution map; The idle distribution visualization view is constructed by highlighting and visualizing the regional idle distribution map to complete the farmhouse idle discrimination job. The specific steps for completing the farmhouse idle discrimination job based on the idle house identification result are as follows: An image processing module is configured to acquire multi-source remote sensing satellite images, perform multi-time point fitting, and construct a multi-time image set; A target identification module is configured to perform farmhouse target identification and fine contour segmentation on the multi-time image set and mark multiple farmhouse image frames; A template segmentation module is configured to perform time sequence tracking segmentation based on the multiple farmhouse image frames to obtain multiple time sequence image frames for each image frame; An idle evolution analysis module is configured to perform time sequence idle evolution mining based on the multiple time sequence image frames to generate an idle evolution trajectory for each farmhouse; An intelligent discrimination module is configured to perform idle state trend analysis based on the idle evolution trajectory and obtain an idle house identification result.

Images