Indoor processing method for ecological protection red line exploration data

Abstract

The invention relates to the technical field of data processing, in particular to an ecological protection red line exploration data interior processing method which comprises the following steps: collecting remote sensing images, landform and land utilization data, and eliminating error data and interference information to obtain cleaned basic data; and performing filtering processing on the cleaned basic data to generate data after comprehensive filtering. According to the method, continuous updating of the real-time remote sensing data is combined with optimization of the prediction model, so that the refined processing and dynamic management capability of the ecological protection red line geographical base map is improved. The geographic boundary is refined through morphological filtering, the accuracy of boundary recognition is enhanced, and the change of the ecological protection area is allowed to be monitored more accurately. The introduction of the dynamic spatio-temporal index ensures the real-time performance of data processing, and can quickly respond to environmental changes. And the adaptability and the perspectiveness of the prediction model are further enhanced through prediction of future geographical boundary changes by combining ARIMA and LSTM algorithms.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, and particularly relates to an ecological protection red line survey data office processing method. BACKGROUND

[0002] The ecological protection red line survey data office processing method is a set of systematic methods specifically designed for processing and analyzing data collected during the ecological protection area surveying process. It includes various types of environmental data such as topography, biodiversity indicators, and land use conditions. The purpose is to delineate and manage ecological protection red line areas through processing these data, ensuring effective protection and management of these areas, and preventing inappropriate human activities from causing damage to these sensitive areas.

[0003] However, the existing technology lacks the ability to update in real time and dynamically adapt to environmental changes in ecological protection red line survey data processing. This leads to decisions often based on outdated or incomplete information, affecting the timeliness and effectiveness of protection measures. In addition, the lack of effective data fusion technology makes it difficult to integrate geographic information collected from different sources, limiting the detail level and application value of the map. These technical limitations may lead to misjudgment of the boundaries of ecological protection areas in actual operation, increasing the risk of damage to the ecological system. SUMMARY

[0004] The purpose of the present application is to solve the shortcomings in the prior art and propose an ecological protection red line survey data office processing method.

[0005] In order to achieve the above-mentioned purpose, the present application adopts the following technical scheme, an ecological protection red line survey data office processing method, comprising the following steps:

[0006] Collect remote sensing images, topography and land use data, remove incorrect data and interference information, and obtain cleaned basic data; filter the cleaned basic data to generate comprehensive filtered data;

[0007] Based on the comprehensive filtered data, use morphological filtering to refine the geographical boundary, optimize the boundary recognition accuracy, and generate boundary optimization data; establish dynamic spatio-temporal index for the boundary optimization data, and obtain spatio-temporal index data;

[0008] Extract the current geographical features and land use conditions from the spatio-temporal index data, combine ARIMA and LSTM time series analysis algorithms to predict the geographical boundary changes in the future time period, form a preliminary prediction model, adjust and optimize the preliminary prediction model according to the real-time updated remote sensing data, and obtain the adaptive prediction result;

[0009] Based on the adaptive prediction result, geographical information is integrated to map a real-world terrain environment, multi-scale fusion of data is performed, and an ecological protection red line geographical base map is obtained.

[0010] Preferably, the obtaining step of the cleaned basic data is:

[0011] Collect remote sensing images, topography and land use data, exclude errors and external interference information, and obtain preliminary screening data;

[0012] Based on the preliminary screening data, the internal consistency and integrity of the data are analyzed, errors and inconsistencies are corrected, and optimized data are obtained.

[0013] Based on the optimized data, data integration is performed, data format and standards are unified, and cleaned basic data are obtained.

[0014] Preferably, the obtaining step of the comprehensive filtered data is:

[0015] The weighted least squares filter is applied to the cleaned basic data to obtain weighted least squares filter data.

[0016] Based on the weighted least squares filter data, Kalman filter is applied for processing to obtain comprehensive filtered data.

[0017] Preferably, the obtaining step of the boundary optimization data is:

[0018] Based on the comprehensive filtered data, an expansion operation is applied, and then a corrosion operation is performed to remove non-boundary information introduced by expansion to obtain morphological processing data.

[0019] Based on the morphological processing data, the overall quality index of the boundary optimization data is calculated, and the formula is improved as:

[0020]

[0021] Wherein, C i represents the contrast of the boundary point i, T i represents the terrain complexity of the point, W i represents the weight coefficient, D opt represents the overall quality index of the boundary optimization data, and N is the number of total boundary points.

[0022] Based on the overall quality index of the boundary optimization data, the boundary data points are adjusted to match the terrain features to obtain the boundary optimization data.

[0023] Preferably, the obtaining step of the spatio-temporal index data is:

[0024] Extract the timestamp and geographic coordinates of each data point from the boundary optimization data, apply a hash function to map the data points into a multi-dimensional data structure, build a dynamic spatio-temporal index, and obtain a preliminary spatio-temporal index framework;

[0025] Based on the preliminary spatio-temporal index framework, measure the efficiency of the spatio-temporal index, calculate the response rate, and the formula is:

[0026]

[0027] Where T new and T old represent the timestamps of adjacent data points, X, Y, and Z are the spatial coordinates of the data points, and R is the response rate.

[0028] Based on the response rate, adjust the key of the hash function or the storage framework of the index, and obtain the spatio-temporal index data.

[0029] Preferably, the preliminary prediction model is obtained by:

[0030] Based on the spatio-temporal index data, extract geographic features and land use data, including terrain type, land cover category and use frequency, and obtain current geographic and land use data.

[0031] Input the current geographic and land use data into the ARIMA and LSTM models, configure parameters and initialize models, train the ARIMA and LSTM models, and obtain a preliminary prediction model.

[0032] Preferably, the adaptive prediction result is obtained by:

[0033] Obtain remote sensing monitoring data through an API interface, including updated geographic coordinates, land cover types and change frequencies, and obtain a real-time updated remote sensing data set.

[0034] Adjust the preliminary prediction model using the real-time updated remote sensing data set, retrain the model, and obtain an adjusted prediction model.

[0035] Apply the adjusted prediction model to perform prediction testing, adjust model parameters according to the test results, and obtain an adaptive prediction result.

[0036] Preferably, the ecological protection red line geographic base map is obtained by:

[0037] Based on the adaptive prediction result, use GIS to standardize the data and obtain standardized geographic feature data.

[0038] Based on the standardized geographic feature data, multi-scale fusion and filtering are performed to integrate geographic information of different sources, edge enhancement is applied, and enhanced map data is obtained.

[0039] The enhanced map data is converted into an ecological protection red line geographic base map, the visual style and regional identification of the map are adjusted, the boundaries and geographic features of the ecological protection area are displayed, and the ecological protection red line geographic base map is obtained.

[0040] Compared with the prior art, the advantages and positive effects of the present application are:

[0041] The present application improves the refinement and dynamic management capabilities of the ecological protection red line geographic base map by combining continuous updating of real-time remote sensing data with optimization of the prediction model. The morphological filtering refines the geographic boundaries and enhances the accuracy of boundary recognition, allowing more accurate monitoring of changes in the ecological protection area. The introduction of dynamic spatiotemporal indexing ensures the real-time nature of data processing, enabling rapid response to environmental changes. The combination of ARIMA and LSTM algorithms for predicting future geographic boundary changes further enhances the adaptability and foresight of the prediction model. The application of multi-scale data fusion technology improves the information richness and visual effect of the map, making the supervision of the ecological protection red line more accurate based on precise geographic data, thereby effectively supporting ecological protection and resource management decisions. BRIEF DESCRIPTION OF DRAWINGS

[0042] Figure 1 The present application is a schematic diagram of the steps. DETAILED DESCRIPTION

[0043] In order to make the purpose, technical solution and advantages of the present application clearer and more understandable, the present application will be further described in detail below in combination with the drawings and examples. It should be understood that the specific examples described herein are only used to explain the present application and do not limit the present application.

[0044] Please refer to Figure 1 The present application provides a technical solution, a method for processing ecological protection red line survey data in the office, comprising the following steps:

[0045] Collect remote sensing images, topography and land use data, remove erroneous data and interference information, and obtain cleaned basic data; filter the cleaned basic data to generate comprehensive filtered data;

[0046] Based on the comprehensive filtered data, morphological filtering is used to refine the geographic boundaries, optimize the boundary recognition accuracy, and generate boundary optimization data; dynamic spatiotemporal indexing is established for the boundary optimization data to obtain spatiotemporal indexing data;

[0047] The current geographical features and land use are extracted from the spatio-temporal index data, combined with ARIMA and LSTM time series analysis algorithms, to predict the geographical boundary changes in the future time period, form a preliminary prediction model, and adjust and optimize the preliminary prediction model according to the real-time updated remote sensing data to obtain adaptive prediction results.

[0048] Based on the adaptive prediction results, the geographical information is integrated to map the real-world terrain environment, multi-scale fusion of data is performed, and the ecological protection red line geographical base map is obtained.

[0049] The acquisition steps of the cleaned basic data are:

[0050] Collect remote sensing images, topography and land use data, exclude errors and external interference information, and obtain preliminary screening data;

[0051] Based on the preliminary screening data, the internal consistency and integrity of the data are analyzed, errors and inconsistencies are corrected, and optimized data are obtained.

[0052] Based on the optimized data, data integration is performed, and the data format and standard are unified to obtain the cleaned basic data.

[0053] Specifically, after obtaining the existing remote sensing images, topography and land use data, each image record is compared with the pre-set pixel brightness value interval and coordinate coverage range, for example, the brightness value is compared between 10 and 250, and if it exceeds this range, it is marked as invalid record or external interference information, and the pixel coordinates involved in the data are marked as abnormal coordinates for subsequent comparison. If multiple time phase data at the same position are found to be abnormal during comparison, it is further determined that there is serious interference at this position. Then, by checking the cloud coverage or satellite signal jitter information frame by frame, obvious cloud shadows and noise points are identified. All data entries determined to be invalid are classified and summarized and removed from subsequent processing steps. During the process, the brightness value judgment range is also adjusted according to historical observation experience. This adjustment can be obtained by technicians after a large number of sample statistics in the early stage, such as adjusting the brightness value interval to 15-245 to more accurately adapt to the differences of multiple sensors. After removing all invalid data in this process, the preliminary screening data are obtained.

[0054] On the basis of the preliminary screening data obtained in the foregoing, firstly, the timestamp and coordinate distribution of each data item are counted, and it is compared whether there are items outside the expected time period or geographical range, for example, the geographical coordinates are limited to the interval of longitude 70° to 140° and latitude 0° to 55°, if the coordinates of a record exceed the range, it is regarded as inconsistent data, then the coverage type information of each record is compared, and it is checked whether there is a contradiction with adjacent records, for example, the land cover type at the same location in the same period is marked as forest land in one record and bare land in another record, then it is determined as error or data conflict, for such items, the coverage of adjacent time phases can be checked or compared with the terrain complexity to make corrections, and the correction is made by referring to the previously defined comparison list, for example, for the area with a slope higher than 45° and an average albedo falling within a certain threshold range, it is inclined to consider that the land cover type is mountain forest land, and the error mark in a certain environment is updated to a more reasonable coverage type, finally, all the modified records are uniformly managed to obtain optimized data.

[0055] On the basis of the optimized data obtained in the foregoing, a global coordinate projection is selected as a unified standard, for example, EPSG: 4326, by comparing the original coordinate reference system of each record, the longitude and latitude units or the plane coordinate units of the selected projection are converted, the resolution difference is corrected one by one, if there is a record with image resolution lower than 10 meters, interpolation operation can be performed to make it reach 10 meters resolution, then the vector boundaries or raster blocks of all data are numbered and managed, according to the previously determined layer classification rules, the terrain and land use information are classified and summarized according to mountain, plain, water area, etc., and compared with the preset numerical range, for example, it is detected whether the records with slope in the interval of 0° to 30° and precipitation in the interval of 0 millimeter to 300 millimeters match the corresponding area, if it is found that the range is exceeded or data is missing, it is checked again and updated, after the completion of all data merging and coordinate unification, the cleaned basic data is obtained.

[0056] The obtaining step of the comprehensive filtered data is:

[0057] The weighted least squares filtering is applied to the cleaned basic data to obtain weighted least squares filtering data.

[0058] Based on the weighted least squares filtering data, Kalman filtering is applied for processing to obtain comprehensive filtered data.

[0059] Specifically, the weighted least squares filtering is applied to the cleaned basic data, firstly, each observation item is read from the cleaned basic data obtained in the foregoing, in the reading process, the confidence index of the observation item is counted, the confidence index is obtained from the noise level analysis of each observation item in the foregoing, the confidence index is converted into weight w k and the error estimation value of the observation item is recorded, then the target function of weighted residual sum of squares is set where k counts from 1 to N, indicating that there are N observation items in total, and the measured value of each observation item is y k , the prediction function is f(x k ), x k represents the input quantity corresponding to the observation item, and then the initial step size and the upper limit of the number of iterations are selected according to the characteristics of the measured data, and then the iteration operation is performed to minimize Q. In each iteration, the mean square error of the sum of the residuals of all observation items is calculated. If the absolute value of the residual of a certain observation item exceeds a certain empirical threshold, for example, 3 standard deviations, it is judged that the observation item may have an outlier phenomenon and its corresponding weight is adjusted down a notch in subsequent processing. The weight can be segmented according to the confidence level, for example, if the confidence level is higher than 0.9, w k = 0.8, if the confidence level is between 0.5 and 0.9, w k = 0.5, and if the confidence level is less than 0.5, w k = 0.2. After adjusting the weight, the target function Q is calculated again until the number of iterations reaches the pre-set upper limit or the change amplitude of the target function is less than a certain stop threshold estimated by experience. The final iteration result is taken as the output of the weighted least squares filtering, and the weighted least squares filtering data is obtained.

[0060] Based on the weighted least squares filtering data, first define the state vector and the observation vector. The state vector can include position and velocity elements. The observation vector reads the corresponding measured value from the weighted least squares filtering data obtained in the previous step at each time and constructs an observation matrix. Set the initial error covariance matrix and the prediction step. When performing Kalman filtering, first perform the prediction operation. According to the state vector of the last time, it is recursively propagated to the current time through the linear state equation and the error is estimated. Then, the Kalman gain is calculated according to the difference between the weighted least squares filtering data at the current time and the predicted value. If the absolute value of the difference is greater than a certain limit value obtained by statistical analysis of historical samples, for example, the limit value is set to be greater than 10 meters of position error or 2 meters per second of speed error, it is judged that there is a significant deviation between the observation and the prediction, and then the gain is increased to rely more on the measured data to correct the state estimation. Otherwise, maintain the original gain level and only make a small adjustment to the state. Repeat the prediction and update steps until the number of cycles reaches the pre-set upper limit or the state vector fluctuation is less than the convergence threshold determined by repeated calculation of historical tests. Finally, record the updated state vector to obtain the data after comprehensive filtering.

[0061] The acquisition steps of the boundary optimization data are:

[0062] Based on the data after comprehensive filtering, apply the dilation operation, and then perform the erosion operation to remove the non-boundary information introduced by the dilation to obtain the morphological processing data.

[0063] Based on the morphological processing data, the overall quality index of the boundary optimization data is calculated, and the formula is improved as:

[0064]

[0065] Wherein, C i represents the contrast of the boundary point i, T i represents the terrain complexity of the point, W i represents the weight coefficient, D opt represents the overall quality index of the boundary optimization data, and N is the number of total boundary points.

[0066] Based on the overall quality index of the boundary optimization data, the boundary data points are adjusted to match the terrain features, and the boundary optimization data is obtained.

[0067] Specifically, based on the comprehensive filtered data, after reading each pixel area, a size of a morphological kernel is selected, for example, under the condition that the image resolution is 10 meters, the morphological kernel is set to a square structure of 3*3, then the expansion operation is performed on all pixel areas in sequence, the newly added edge pixel area after expansion is recorded, if the gray value or color value of a certain pixel area is higher than the reference value 0.6 obtained by prior statistics, it is marked as a significant pixel, the significant pixel is accumulated one by one within the range of the morphological kernel, and the accumulation is stopped after 5 times. Then, the noise marked in the expansion stage is deleted by erosion operation. When eroding, the gray difference value of each significant pixel in the 3*3 range around the pixel is compared again, if the difference value is less than 0.2 based on the actual test summary, the point is regarded as an erosion object and is removed, if a certain area cannot be removed by continuous erosion for many times, the area is recorded as a special edge area, which can be matched with the terrain data or referenced by actual sampling detection. After one complete expansion and erosion, the remaining morphological edge distribution in the whole image is checked again, if the length of the boundary broken section detected is greater than the maximum continuous broken distance recorded during field surveying and mapping, for example, the distance is recorded as 15 pixel points after actual measurement and statistics, additional correction is performed, and after the final completion of the expansion and erosion of all areas, the morphological processing data is obtained.

[0068] The formula is beneficial in that the ratio of contrast C i and terrain complexity T i is introduced comprehensively and weighted W i for square accumulation, which takes into account both the surface details and the boundary strength in the same formula, so as to quantitatively measure the overall quality of the boundary optimization data.

[0069] C iThe acquisition step is: selecting the i-th boundary point from the morphological processing data, reading the difference between the average gray value in the 3x3 neighborhood around the point and the current pixel gray value, normalizing the difference to 0 to 1, for example, for a certain boundary point, if the average gray value in the neighborhood is 120, and the current pixel gray value is 180, then C i =(180-120) / 255≈0.235, which is measured point by point and recorded as an array.

[0070] T i The acquisition step is: in the 2x2 elevation measurement around the boundary point, the difference between the maximum slope and the minimum slope is counted, and the difference is normalized to 0 to 1, for example, if the maximum slope in the neighborhood of a certain boundary point is 35°, and the minimum slope is 10°, then T i =(35-10) / 90≈0.278, where 90° is the empirical upper limit of steep mountains, and the relevant slope values can be obtained through the terrain measurement results obtained before.

[0071] W i The acquisition step is: referring to the interaction between edge saliency and terrain change in morphology, and combining with the results of multiple field surveys, recording the stability factor of the same position in multiple temporal data, for example, observing the change range of the same boundary position in three surveys, if the change range is within 10 meters, then the stability is 0.9, then according to the stability multiplied by a segmented coefficient, if the stability is higher than 0.8, then W i =0.7, if between 0.5 and 0.8, then W i =0.5, if lower than 0.5, then W i =0.3.

[0072] The calculation process is: let N=3 select three boundary points for demonstration, set: C1=0.235, T1=0.278, W1=0.7, C2=0.4, T2=0.33, W2=0.5, C3=0.38, T3=0.27, W3=0.3;

[0073] Calculate

[0074]

[0075] Calculate

[0076]

[0077] Calculate

[0078]

[0079] Then, add the above three items:

[0080] 0.4998+0.735+0.594≈1.8288

[0081] Finally, calculate D opt :

[0082]

[0083] The results show that the ratio of the three boundary points selected at this time comprehensive contrast and terrain complexity is about 1.3523 after applying the set weight factor, the larger the value, the more prominent the corresponding boundary in the current observation range, if the value obtained by repeated calculation of a certain boundary area is greater than 2, it is recorded as a special significant boundary, the value between 1 and 2 can be regarded as an ordinary boundary, and the value less than 1 is regarded as a weak boundary area for subsequent investigation.

[0084] Based on the overall quality index of boundary optimization data, after reading the contrast and height difference between adjacent points of all boundary points, if the height difference is greater than the 5-meter limit value obtained by previous surveying and statistical analysis, it is determined that there is obvious terrain drop at this point. Mark such points on each boundary line and include them in the same paragraph with the surrounding boundary points. If the average terrain drop in the same paragraph exceeds the 8-meter threshold value refined based on historical observations, it is marked as a special paragraph during recording. The coordinate position of these special paragraphs is adjusted to match the terrain measurement data. After recording is completed, a new adjustment index is formed. Then further compare the contrast and slope information of all boundary points to determine whether there are multiple consecutive boundary points exceeding the 3-degree threshold value of angle difference. If it is satisfied, the paragraph is included in the terrain mutation area during output. After all boundary data points are processed, the boundary optimization data is obtained.

[0085] The steps for obtaining the space-time index data are as follows:

[0086] Extract the timestamp and geographic coordinates of each data point from the boundary optimization data, apply a hash function to map the data points to a multi-dimensional data structure, construct a dynamic space-time index, and obtain a preliminary space-time index framework;

[0087] Based on the preliminary space-time index framework, measure the efficiency of the space-time index, and calculate the response rate. The calculation formula is:

[0088]

[0089] Where, T new and T old represent the timestamps of adjacent data points, X, Y, and Z are the spatial coordinates of the data points. R is the response rate;

[0090] Based on the response rate, adjust the storage framework of the key or index of the hash function, and obtain the spatiotemporal index data.

[0091] Specifically, the timestamp and geographic coordinates of each data point are extracted from the boundary optimization data. First, all boundary points are read one by one, and the timestamp format of each record is verified during reading. For example, refer to the structure of year-month-day-hour-minute and check if it is complete. After confirming that the timestamp is correct, parse the corresponding year, month, day, and hour information and convert it to a unified hour count value. Then, the coordinates of each record are compared. If the coordinates are outside the previously defined longitude interval of 75° to 135° or the latitude interval of 10° to 50°, they are considered abnormal records and are not included in the subsequent construction. The normal records that pass the test are then hashed in memory. During hashing, the longitude and latitude values in the coordinates and the timestamp value are first combined into a multi-dimensional key sequence, such as [hour timestamp, longitude, latitude]. The multi-dimensional key sequence is then discretized, and a hash operation is performed to assign data points with the same or similar time and space location to adjacent storage areas. After construction, the hash mapping table is scanned one by one. If data points within the same time range (such as within 24 hours) and with a geographical location difference less than 5 kilometers based on previous test summaries are found, these points are further aggregated into an index paragraph. In this way, several segments are generated in memory or a database for subsequent quick search and dynamic update. After all records are completed, these index segments are summarized to obtain a preliminary spatiotemporal index framework.

[0092] The formula has the benefit of considering the product effect of time interval and three-dimensional space distance at the same time. By introducing time difference and space coordinate difference in the same expression, the response degree of adjacent data points can be reflected in a comprehensive index.

[0093] T new The acquisition step is: selecting the latest data point in the preliminary spatiotemporal index framework obtained in the previous paragraph, parsing and converting it to an hour count value from its record. For example, the timestamp record of a certain data point is 2024-08-1012:00. The year, month, day, and hour information are summarized as total hours. If the time is recorded from January 1, 2024, 00:00 as the origin, this time may correspond to about 5496 hours (only an example, the specific value needs to be calculated according to the accurate date), and this value is assigned to T new .

[0094] T old The acquisition step is: similarly, select the earlier data point adjacent to T new and parse the timestamp to convert it to an hour count value. If the data point time is 2024-08-0912:00, its corresponding hour count is about 5472 hours, and this value is assigned to Told ;

[0095] The acquisition of X, Y, Z is to read the geographical position of the data point in the recorded three-dimensional coordinate system, for example, to convert the longitude and latitude height information into X=200, Y=300, Z=20 in meters by using the geodetic coordinate system conversion, and the specific conversion needs to be combined with the used ellipsoid parameters and projection method, and obtained from the previous geographical measurement link.

[0096] Calculation process:

[0097] Calculate max(|T new -T old |):

[0098] |5496-5472|=24

[0099] Calculate the spatial distance

[0100]

[0101] Calculate the exponential term

[0102] exp(-361.18)≈1.62×10 -157

[0103] Bring the above results into the formula R:

[0104]

[0105] The result shows that for the two records that differ by 24 hours in time and are about 361.18 meters apart in three-dimensional space, the response rate is only 6.75×10 -159 , the smaller the value, the farther apart the data points are in time and space, when the value is higher than 0.01, it can be marked as high degree of adjacency, and when it is lower than 10 -5 , it can be marked as extremely low degree of adjacency.

[0106] Based on the response rate, compare each data point when scanning all index paragraphs, when finding that the response rate value of a plurality of records is less than the threshold value of 0.02 calculated based on historical observation, the hash value corresponding to the records is remapped and marked as different groups respectively, and the space allocation strategy of the current storage framework is adjusted in the updating process, for example, it is determined that it should increase a mapping conflict check inside the index, if a hash slot appears data overflow more than 10 records, a new slot is added, and the original mapping section is reserved to store other scattered data, after completing a round of update, the response rate value distribution is summarized again to confirm whether there are more than 30 records concentrated in the same slot, if there are, the slot is split into two or more slots, and the timestamp and geographic coordinate information corresponding to the slot are migrated, each migration calls the geographic coordinate information and adjacent timestamp difference obtained in the previous text for final check, and after all slot adjustments are completed, the new spatio-temporal index data is obtained.

[0107] The obtaining step of the preliminary prediction model is:

[0108] Based on the spatio-temporal index data, extract geographic features and land use data, including terrain type, land cover category and use frequency, to obtain current geographic and land use data;

[0109] Input the current geographic and land use data into the ARIMA and LSTM models, configure parameters and initialize the models, train the ARIMA and LSTM models, and obtain the preliminary prediction model.

[0110] Specifically, based on the spatio-temporal index data, first traverse each point's geographic features and land use information according to the previously obtained spatio-temporal index data, and extract its corresponding terrain type, land cover category and use frequency, etc. If the terrain type is marked as mountain, record the height and slope, if the cover category is marked as farmland, record the annual plowing frequency and compare the planting species information, and the use frequency in all records is disassembled into monthly or quarterly cumulative statistics and compared, for example, in a year of time span, if the farmland use frequency is found to be more than 5 times and the water demand is more than 500 cubic meters per mu, it is marked as high-frequency use plot, for forest land, read its tree species, density and maturity, if the tree density exceeds the 300 trees per hectare statistically obtained in the previous investigation, it is marked as higher coverage forest land, after completing the disassembly and analysis of all point data, the current geographic and land use data is obtained.

[0111] The current geographic and land use data is input into the ARIMA and LSTM models. After reading the data, the ARIMA model is set to a daily interval of 30 days for the time series length, and the autoregressive order and moving average order are configured according to the data obtained earlier. For example, in an area with flat cultivated land, after analyzing the data recorded continuously for 1 year, p is set to 2 and q is set to 1. Initial estimates of the parameters are made during model initialization, and then multiple iterations are performed. In each iteration, the current parameters are corrected by the least squares error principle and the new estimated values are recorded. If the number of iterations exceeds the upper limit of 1000 determined by a large number of calculations on historical data in advance, the training is stopped and the latest converged parameter value is used as the final output. At the same time, the same data set is provided to the LSTM model for sequential processing. In LSTM, the number of neurons is determined to be 64 and the time series batch size is set to 7 days as a time step. The use frequency, coverage category and other information obtained earlier are used as input dimensions. After each forward propagation calculation, the error is backpropagated and the weights are updated. The error threshold can be 0.001 after multiple comparisons during actual data testing. When the error is lower than the threshold or the number of training iterations exceeds 500, the training is stopped and the model weights at that time are used as the final output. The preliminary prediction model is finally trained.

[0112] The adaptive prediction result acquisition step is:

[0113] Remote sensing monitoring data is obtained through an API interface, including updated geographic coordinates, land cover types and change frequencies, to obtain a real-time updated remote sensing data set.

[0114] The real-time updated remote sensing data set is used to adjust the preliminary prediction model, retrain the model, and obtain an adjusted prediction model.

[0115] The adjusted prediction model is applied for prediction testing, and the model parameters are adjusted according to the test results to obtain an adaptive prediction result.

[0116] Specifically, the remote sensing monitoring data is obtained through the API interface, and a query instruction is sent to the designated data source in sequence to extract updated geographic coordinates, land cover types, and change frequency, etc. when each request is made. When reading the returned geographic coordinates, the format is decoded and it is confirmed whether the coordinates fall within the range of 65° to 140° in longitude and 0° to 60° in latitude determined during previous surveying. If the coordinates of a certain record exceed this range, it is marked as an abnormal record and not included in subsequent processing. Then, according to the land cover type information returned, the cover type list established previously is compared, such as grassland, forest land, farmland, and construction land, etc. If it is found that a certain record is identified as grassland but has a high construction frequency, it is recorded as a mixed cover category. When reading the change frequency, the number of cover changes within a month is compared. If it exceeds the threshold of 3 times obtained from previous sample statistics, it is assigned to the high-frequency change type and additional remarks are added. In this process, the returned monitoring time stamp is also compared in time sequence. If the time difference between the two adjacent times is less than 12 hours and the cover type changes, it is determined as a short-period change point and is classified and stored. After the analysis of all records is completed, the real-time updated remote sensing data set is obtained.

[0117] The real-time updated remote sensing data set is used to adjust the preliminary prediction model. First, the geographic coordinates and cover type changes are read from the real-time updated remote sensing data set. Then, they are compared with the existing parameter index in the preliminary prediction model obtained previously. If it is found that the current location has been labeled as farmland in the front but is recorded as construction land in the new data set and the frequency statistics show that the location has more than 2 cover changes within the past two months, it is determined that the change amplitude is large. In this case, the farmland weight of the model can be adjusted downward according to the actual observation sample statistics, and the frequency factor of the construction land is correspondingly improved. Then, all updated entries are combined to generate a new training sample. When training is performed, a time sequence data set is arranged in time sequence. The ARIMA part is windowed with a 30-day window, and the LSTM part is extracted every 7 days as a time step. After each iteration, the current prediction error is compared. If the mean square error of 5 consecutive iterations is less than the threshold of 0.002 measured in the previous period, the training is ended and the weight coefficient of the iteration is retained when recording. Otherwise, the iteration continues until the upper limit of 200 training is reached. After the above process is completed, the adjusted prediction model is obtained.

[0118] The adjusted prediction model is applied to call the model for prediction test on subsequent time series data. First, a continuous 15-day observation entry is selected from the real-time updated remote sensing data set as a test set. The ARIMA part is iteratively predicted according to the previously set autoregressive and moving average order. The difference between the predicted value and the measured value is summarized and compared day by day. If the difference value appears continuously more than the 5% threshold quantified from the historical data, the parameter adjustment process is triggered. After triggering the adjustment, the learning rate in the LSTM part is fine-tuned according to the cumulative deviation degree. For example, if the difference rate is in the interval of 5% to 10% for 3 consecutive days, the learning rate is increased by 0.0005. If it exceeds 10%, the learning rate is increased by 0.001 again. A number of rounds of supplementary training are immediately performed. After the supplementary training is completed, the prediction test is performed again. If the average difference rate of the prediction result within the next 5 days is reduced to below 5%, the adjustment process is stopped and the current model parameters are recorded. Finally, the adaptive prediction result is obtained.

[0119] The acquisition steps of the ecological protection red line geographic base map are:

[0120] Based on the adaptive prediction result, the data is standardized by GIS to obtain standardized geographic feature data;

[0121] Based on the standardized geographic feature data, multi-scale fusion and filtering are performed to integrate different source geographic information, and edge enhancement is applied to obtain enhanced map data;

[0122] The enhanced map data is converted into an ecological protection red line geographic base map, the visual style and regional identification of the map are adjusted, the boundary and geographic features of the ecological protection area are displayed, and the ecological protection red line geographic base map is obtained.

[0123] Specifically, based on the adaptive prediction result, first, all geographic numerical data are compared according to a unified coordinate projection, and the longitude and latitude are respectively segmented and counted, for example, the longitude is divided into two parts of 70° to 110° and 110° to 140° for checking, and at the same time, whether the latitude is within the range of 0° to 50° is compared, if there is a record exceeding the range, it is marked as an abnormal point, then the height value of each record is re-quantized, for example, the height is compared with the interval of 0 meters to 5000 meters, if it is outside this interval, it is marked again as a record to be checked, for the land use type, a reference list containing farmland, forest land, water area and other types can be set up, and the cover type in the record is checked piece by piece whether it conforms to the list, if missing or mismatching is found, the previous field observation data is used for supplement or classification, and the use frequency is normalized according to the time sequence observation result obtained in the previous step, for example, the use frequency of each month is graded according to the interval of 0 to 12 times, and more than 12 times is regarded as abnormal, if the data appears more than three times, manual checking is called, after the checking is completed, all data are converted into unified decimal or integer form, the geographic coordinates, cover type, use frequency and other fields are integrated into the same standard format, the resolution is uniformly distinguished when distinguishing between vector and raster types, if it is found that the resolution of some raster is lower than 10 meters, interpolation upsampling is performed to ensure the accuracy, and after all the items are uniformly completed, the results are output as standardized geographic feature data at one time.

[0124] Based on the standardized geographic feature data, multiple scale ranges are selected for fusion, for example, the raster data is first accumulated in the scale of 10 meters resolution for gray scale or albedo value, then it is locally subdivided in a higher resolution such as 5 meters to observe whether the gray scale difference of adjacent pixels exceeds the 30 threshold value obtained by previous sample statistics, if it exceeds the threshold value, the pixel is listed as a high-contrast area and edge recognition is performed, when filtering, the gray scale variance in the neighborhood of each pixel is compared, if the variance is greater than the 200 empirical value obtained by historical analysis, it is judged as a noise pixel and mean or median operation is performed, when the multi-scale fusion is completed, the layers are output in proportion and the boundaries are enhanced, in the enhancement step, a 3x3 convolution kernel can be set first to perform weighted processing on the boundaries with gradient change exceeding the set reference value 0.3, if the continuous length of a certain edge exceeds 10 pixel points, the segment is highlighted, after all the enhancement processing is completed, the vector and raster information are integrated with the confirmed geographic identification to obtain enhanced map data.

[0125] The enhanced map data is converted into the ecological protection red line geographic base map, and the enhanced map data is read by layering according to administrative partition or natural partition, if the range of a partition contains two types of mountain and water area, the number of boundary points is checked, if the number of boundary points is greater than 500 points obtained before topographic survey, the partition is divided into several sub-areas, the color scale scheme determined above is applied to each sub-area, and the area label is added, for the mountain area, the visible light band is appropriately displayed in dark color, and the thick line is added to the vector boundary, for the water area, the light blue color is presented, and the adjacent slope information is marked, if the average slope of a region is higher than 30 degrees, the region is marked as steep region and is marked more prominently, finally, the layers are superimposed into the same projection coordinate system, the legend and the scale are uniformly laid out, and after one-time inspection, the ecological protection red line geographic base map is obtained.

[0126] The above is only the preferred embodiment of the present application, and does not limit the present application in other forms, any skilled person in the art can use the disclosed technical content to make changes or modifications into equivalent embodiments applied to other fields, but any simple modification, equivalent change and modification made according to the technical essence of the present application to the above embodiments without departing from the technical solution content of the present application still belongs to the protection scope of the technical solution of the present application.

Claims

1. A method for internal processing of ecological protection red line demarcation data, characterized in that: The following steps are involved: Collect remote sensing images, topography, and land use data, remove erroneous data and interference information, and obtain cleaned basic data; Processing the cleaned basic data by filtering to generate comprehensive filtered data; Based on the comprehensive filtered data, morphological filtering is used to refine the geographic boundaries, optimize the boundary recognition accuracy, and generate boundary optimization data; a dynamic spatiotemporal index is established for the boundary optimization data to obtain spatiotemporal index data; Extracting current geographic features and land use from the spatiotemporal index data, combining ARIMA and LSTM time series analysis algorithms to predict geographic boundary changes in future time periods, forming a preliminary prediction model, and adjusting and optimizing the preliminary prediction model based on real-time updated remote sensing data to obtain an adaptive prediction result; Based on the adaptive prediction results, geographic information is integrated to map the real-world terrain environment, and multi-scale fusion of data is performed to obtain a geographic base map of the ecological protection red line.

2. The method for internal processing of ecological protection red line demarcation data according to claim 1 is characterized in that: The steps for obtaining the cleaned basic data are: Collect remote sensing images, topography and land use data, eliminate errors and external interference information, and obtain preliminary screening data; Based on the preliminary screening data, analyzing the internal consistency and completeness of the data, correcting errors and inconsistencies, and obtaining optimized data; Based on the optimized data, data integration is performed to unify data formats and standards to obtain cleaned basic data.

3. The method for internal processing of ecological protection red line demarcation data according to claim 1 is characterized in that: The steps for obtaining the data after comprehensive filtering are as follows: Applying weighted least squares filtering to the cleaned basic data to obtain weighted least squares filtered data; Based on the weighted least squares filtered data, Kalman filtering is applied to perform processing to obtain comprehensive filtered data.

4. The method for internal processing of ecological protection red line demarcation data according to claim 1 is characterized in that: The steps for obtaining the boundary optimization data are as follows: Applying a dilation operation to the comprehensive filtered data, and then performing an erosion operation to remove non-boundary information introduced by the dilation, to obtain morphologically processed data; Based on the morphologically processed data, the overall quality index of the boundary optimization data is calculated, and the improved formula is: Among them, C i represents the contrast of boundary point i, T i Represents the terrain complexity of the point, W i Represents the weight coefficient, D opt Represents the overall quality index of the boundary optimization data, where N is the number of total boundary points; Boundary optimized data is obtained by adjusting boundary data points to match terrain features based on the overall quality index of the boundary optimized data.

5. The method for internal processing of ecological protection red line demarcation data according to claim 1 is characterized in that: The steps for obtaining the spatiotemporal index data are as follows: Extracting the timestamp and geographic coordinates of each data point from the boundary optimization data, applying a hash function to map the data point into a multidimensional data structure, constructing a dynamic spatiotemporal index, and obtaining a preliminary spatiotemporal index framework; Based on the preliminary spatiotemporal indexing framework, the efficiency of spatiotemporal indexing is measured and the response rate is calculated using the following formula: Among them, T new and T old represents the timestamps of adjacent data points, X, Y, Z are the spatial coordinates of the data points, and R is the response rate; Based on the response rate, the key of the hash function or the storage framework of the index is adjusted to obtain spatiotemporal index data.

6. The method for internal processing of ecological protection red line demarcation data according to claim 1 is characterized in that: The steps for obtaining the preliminary prediction model are: Extracting geographic features and land use data based on the spatiotemporal index data, including terrain type, land cover category, and use frequency, to obtain current geographic and land use data; The current geographic and land use data are input into the ARIMA and LSTM models, parameter configuration and model initialization are performed, the ARIMA and LSTM models are trained, and a preliminary prediction model is obtained through training.

7. The method for internal processing of ecological protection red line demarcation data according to claim 1 is characterized in that: The steps for obtaining the adaptive prediction result are: Obtain remote sensing monitoring data through the API interface, including updated geographic coordinates, land cover types and change frequency, and obtain real-time updated remote sensing datasets; Using the real-time updated remote sensing data set to adjust the preliminary prediction model, retrain the model, and obtain an adjusted prediction model; The adjusted prediction model is applied to perform prediction tests, and the model parameters are adjusted according to the test results to obtain adaptive prediction results.

8. The method for internal processing of ecological protection red line demarcation data according to claim 1 is characterized in that: The steps for obtaining the ecological protection red line geographic base map are as follows: Based on the adaptive prediction results, the data is standardized using GIS to obtain standardized geographic feature data; Based on the standardized geographic feature data, multi-scale fusion and filtering are performed to integrate geographic information from different sources, and edge enhancement is applied to obtain enhanced map data; The enhanced map data is converted into an ecological protection red line geographic base map, the visual style and regional identification of the map are adjusted, the boundaries and geographical features of the ecological protection area are displayed, and the ecological protection red line geographic base map is obtained.

Images