Enterprise data processing method and device based on satellite remote sensing, equipment and medium

Abstract

The present disclosure provides a satellite remote sensing-based enterprise data processing method and device, equipment and medium, relating to the technical field of multi-modal remote sensing driven enterprise-industry chain intelligent modeling method and the like. The specific implementation scheme is: based on a list of multiple enterprises, determining the factory area outline area of each enterprise in the remote sensing image; based on the factory area outline area, obtaining corresponding time series remote sensing data collected by multiple types of satellites; based on the factory area outline area and the time series remote sensing data, determining the number of pixels of each enterprise; based on the number of pixels, the factory area outline area and the time series remote sensing data, determining the remote sensing feature time series value of each enterprise; dividing the multiple enterprises into industry categories, and based on the remote sensing feature time series value, determining a remote sensing feature subset of an enterprise subset in each industry category; based on the remote sensing feature subset, predicting enterprise data of the enterprise subset to obtain enterprise prediction results.

Description

Technical Field

[0001] This disclosure belongs to the field of data processing technology, and particularly relates to image processing, deep learning, and other technical fields. Specifically, it discloses a method and apparatus for enterprise data processing based on satellite remote sensing, an electronic device, and a computer-readable storage medium. Background Technology

[0002] Currently, dynamic monitoring of business operations and macroeconomic indicators has significant application value in areas such as government decision-making, bank risk control, fund industry analysis, and stock market assessment. Traditional methods mainly utilize structured data sources such as government statistics, financial statements, and questionnaires. These methods generally suffer from problems such as long data acquisition cycles, low update frequency, strong lag, and coarse spatial resolution, making it difficult to achieve high-frequency dynamic perception of enterprise-level and regional-level economic activities.

[0003] In recent years, with the development of satellite remote sensing technology, "alternative data" based on satellite observation data has become an emerging source of information. Among these, remote sensing data such as nighttime light (NTL), land surface temperature (LST), and air pollutants (e.g., NO2, SO2, CO) have been shown to have significant correlations with human activities, energy consumption, industrial emissions, and economic activity. Therefore, some studies have attempted to use single remote sensing modes (such as nighttime light) to estimate urban economic activity, electricity consumption levels, or the intensity of social activities; however, their ability to identify micro-entities such as manufacturing enterprises remains limited.

[0004] Existing research has explored the use of remote sensing data to estimate macroeconomic indicators such as GDP, PMI, and industrial added value. For example, remote sensing imagery has been used to train regression models to fit county-level GDP or to make rough estimates of the output of a particular industry. However, these methods are mostly limited to city-level or industry-level macro-scales, lacking the ability to spatially locate, dynamically track, and model behavior at the enterprise level, and thus failing to achieve chain-like modeling from the smallest granular enterprise production and operation activities to macroeconomic indicators.

[0005] In addition, although there are studies on the fusion of remote sensing multi-source data (such as MODIS, Sentinel, VIIRS, TROPOMI, etc.), most of them focus on land cover classification, climate monitoring, environmental assessment, etc., and are still weak in the modeling of causal mechanisms between remote sensing modalities and corporate business behavior, industry adaptability matching, and time series feature mining. Summary of the Invention

[0006] This disclosure provides a data processing method and apparatus, an electronic device, and a computer-readable storage medium.

[0007] According to the first aspect, a method for processing enterprise data based on satellite remote sensing is provided. This method includes: determining the factory outline area of ​​each enterprise in a remote sensing image based on a list of multiple enterprises; acquiring corresponding time-series remote sensing data from various types of satellites based on the factory outline area; determining the number of pixels for each enterprise based on the factory outline area and the time-series remote sensing data; determining the time-series values ​​of remote sensing features for each enterprise based on the number of pixels, the factory outline area, and the time-series remote sensing data; classifying the multiple enterprises into industry categories, and determining the remote sensing feature subsets of enterprises within each industry category based on the time-series values ​​of remote sensing features; and predicting the enterprise data of the enterprise subsets based on the remote sensing feature subsets to obtain enterprise prediction results.

[0008] According to the second aspect, a satellite remote sensing-based enterprise data processing device is provided. The device includes: a region determination unit configured to determine the factory outline region of each enterprise in a remote sensing image based on a list of multiple enterprises; an acquisition unit configured to acquire corresponding time-series remote sensing data collected by various types of satellites based on the factory outline region; a quantity determination unit configured to determine the number of pixels for each enterprise based on the factory outline region and the time-series remote sensing data; a value determination unit configured to determine the time-series values ​​of remote sensing features for each enterprise based on the number of pixels, the factory outline region, and the time-series remote sensing data; a subset determination unit configured to classify the multiple enterprises into industry categories and determine the remote sensing feature subsets of enterprises within each industry category based on the time-series values ​​of the remote sensing features; and a prediction unit configured to predict the enterprise data of the enterprise subsets based on the remote sensing feature subsets to obtain enterprise prediction results.

[0009] According to a third aspect, an electronic device is provided, comprising: at least one processor; and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform a method as described in any implementation of the first aspect.

[0010] According to a fourth aspect, a non-transitory computer-readable storage medium is provided that stores computer instructions for causing a computer to perform the method described in any implementation of the first aspect.

[0011] The embodiments of this disclosure provide a satellite remote sensing-based enterprise data processing method and apparatus. First, based on a list of multiple enterprises, the factory outline areas of each enterprise in the remote sensing image are determined. Second, based on the factory outline areas, corresponding time-series remote sensing data collected by various types of satellites are acquired. Third, based on the factory outline areas and the time-series remote sensing data, the number of pixels for each enterprise is determined. Next, based on the number of pixels, the factory outline areas, and the time-series remote sensing data, the time-series values ​​of remote sensing features for each enterprise are determined. Then, the multiple enterprises are categorized by industry, and based on the time-series values ​​of remote sensing features, a subset of remote sensing features for each subset of enterprises within each industry category is determined. Finally, based on the subsets of remote sensing features, the enterprise data for the subsets of enterprises is predicted to obtain enterprise prediction results. Thus, by accurately matching the enterprise list with the remote sensing image, automatically extracting factory outlines, and fusing multi-source satellite time-series data, the production activities of each enterprise are transformed into computable pixel-level dynamic features. Subsequently, these features are clustered by industry to construct representative "remote sensing fingerprints," which are ultimately used to reverse-predict the production capacity, emissions, and even operational risks of any enterprise. Without requiring enterprise declaration, high-frequency, low-cost, comprehensive and objective digital twin archives can be generated across the entire region, enabling "zero-disturbance" monitoring and forward-looking insights into enterprise operations, and providing penetrating data support for precise government governance, financial risk control and supply chain optimization.

[0012] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this disclosure, nor is it intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0013] The accompanying drawings are provided to better understand this solution and do not constitute a limitation of this disclosure. Wherein:

[0014] Figure 1 This is a flowchart of an embodiment of an enterprise data processing method based on satellite remote sensing according to the present disclosure;

[0015] Figure 2a This is a schematic diagram of a resonant pixel calculated for large and medium-sized enterprises in this disclosure;

[0016] Figure 2b This is a schematic diagram of a factory buffer zone pixel constructed for small businesses in this disclosure;

[0017] Figure 3 This is a schematic diagram of a structure of an embodiment of an enterprise data processing apparatus based on satellite remote sensing according to the present disclosure;

[0018] Figure 4 This is a block diagram of an electronic device used to implement the enterprise data processing method based on satellite remote sensing according to the embodiments of this disclosure. Detailed Implementation

[0019] Unless otherwise expressly stated, throughout the specification and claims, the term "comprising" or its variations such as "including" or "comprises" shall be understood to include the stated elements or components without excluding other elements or other components.

[0020] The technical solutions of this disclosure are illustrated below through specific embodiments. It should be understood that one or more steps mentioned in this disclosure do not preclude the existence of other methods and steps before or after the combined steps, or that other methods and steps may be inserted between these explicitly mentioned steps. It should also be understood that these examples are for illustrative purposes only and are not intended to limit the scope of this disclosure. Unless otherwise stated, the numbering of each method step is only for the purpose of identifying each method step, and not to limit the order of each method or to limit the scope of implementation of this disclosure. Changes or adjustments to their relative relationships, without substantial changes to the technical content, can also be considered as within the scope of implementation of this disclosure.

[0021] The raw materials and instruments used in the examples are not subject to any specific restrictions on their source; they can be purchased from the market or prepared according to conventional methods known to those skilled in the art.

[0022] Using satellite remote sensing technology to predict enterprise revenue growth and industry activity levels currently faces several challenges:

[0023] 1. Remote sensing data types are limited and lack sufficient fusion depth.

[0024] Current research and practice have attempted to incorporate remote sensing data into macroeconomic analysis or enterprise-level production and operation monitoring. For example, nighttime light is primarily used to correlate with GDP (Gross Domestic Product) growth and population changes; surface temperature is often used to establish a link with steel companies, reflecting their production and operation activities. However, these applications involve limited data types and insufficient fusion depth. Most existing methods rely on nighttime light or thermal characteristics, failing to fully explore the complementarity and sensitivity of multimodal remote sensing information, such as pollutants, across different industries.

[0025] 2. Remote sensing data is large in scale and lacks a precise matching mechanism for remote sensing pixels of enterprises and factories.

[0026] Enterprise-level spatial positioning capabilities are limited. Current research focuses on city or regional scales and lacks a precise matching mechanism for remote sensing pixels of enterprises, especially manufacturing plants. Under the current resolution pixel scale of remote sensing data, it is difficult to effectively cover small enterprises, and it is difficult to identify the production activity areas of larger enterprises. Visual interpretation is the main method, and there is no effective and fast method for screening multiple enterprises.

[0027] 3. Remote sensing modeling generally ignores industry differences and lacks analysis and modeling methods for the relationship between remote sensing modes and industry structure.

[0028] Current remote sensing modeling methods generally neglect industry heterogeneity and lack differentiated modeling mechanisms for the modal response characteristics of enterprises in different industries. Different manufacturing industries exhibit significantly different modal response characteristics at the remote sensing observation level due to differences in production processes, energy structures, emission behaviors, and operating periods. Existing studies often use a uniform remote sensing feature set and a fixed modeling framework, failing to explore the fit between remote sensing variables and the industry to which the enterprise belongs. This easily leads to feature redundancy, information misinterpretation, and decreased model generalization ability.

[0029] To address the shortcomings of traditional technologies, this disclosure proposes a satellite remote sensing-based enterprise data processing method. By accurately matching the enterprise list with remote sensing images, the method automatically extracts the factory area outline and integrates multi-source satellite time-series data, transforming the production activities of each enterprise into computable pixel-level dynamic features. Figure 1 A flowchart 100 is shown as an embodiment of a satellite remote sensing-based enterprise data processing method according to the present disclosure, which includes the following steps:

[0030] Step 101: Based on the list of multiple enterprises, determine the outline area of ​​each enterprise's factory area in the remote sensing image.

[0031] In this embodiment, the factory area outline is the exclusive land boundary (AOI, Area of ​​Interest) of each enterprise, which is precisely delineated on satellite or aerial remote sensing images.

[0032] In this embodiment, the list of multiple enterprises is first matched with the publicly available business registration address database and historical land use vector map to obtain initial coordinates. Then, a deep learning semantic segmentation network is used to automatically extract the connected domains of buildings and impermeable surfaces on the sub-meter level remote sensing image. Combined with manual interactive correction, non-factory area pixels such as roads and green spaces are removed, and finally, the vector factory area outline region corresponding to each enterprise is generated for subsequent monitoring.

[0033] In this embodiment, step 101 further includes: determining the names of each enterprise in the list of multiple enterprises; extracting keywords from the names, calling the map API (Application Programming Interface) interface through the keywords, and automatically obtaining the latitude and longitude coordinates of the main factory area or registered production site of the manufacturing enterprise nationwide in combination with business information data; determining the factory location based on the latitude and longitude coordinates, superimposing high-resolution remote sensing images (such as Google Earth or Jilin-1, etc.) to determine the factory boundary, and identifying the spatial range of the main production area of ​​the enterprise in the GIS (Geographic Information System Platform) platform using manual drawing or image segmentation methods; and drawing the outline area of ​​the factory area on the map.

[0034] It should be noted that the lists and maps of multiple companies can be obtained through various public, legal and compliant means, such as from public datasets or from companies with their authorization.

[0035] Step 102: Based on the outline area of ​​the factory area, acquire the corresponding time-series remote sensing data collected by various types of satellites.

[0036] In this embodiment, the time-series remote sensing data is a collection of multiple satellite data from the same region arranged in chronological order, including optical imagery, radar imagery, and thermal infrared data. It can comprehensively utilize multimodal time-series remote sensing data such as nighttime light (NTL), daytime / nighttime surface temperature (LST_MOD / LST_MYD), and pollutant emissions (CO), overcoming the limitations of traditional methods that rely solely on single optical or thermal signals. This approach can more comprehensively reflect the multidimensional external manifestations of enterprise production activities, improving the explanatory power of remote sensing variables and the model's generalization ability.

[0037] In this embodiment, various types of satellites may include: communication satellites, meteorological satellites, navigation satellites, geodetic satellites, broadcasting satellites, etc., classified according to their purpose; and medium-Earth orbit satellites, low-Earth orbit satellites, sun-synchronous satellites, etc., classified according to their orbit. These are types of satellites that are classified from different perspectives, such as by purpose or orbit.

[0038] In this embodiment, step 102 includes: acquiring time-series remote sensing data of the corresponding factory area outline from data collected from various types of satellites, such as time-series remote sensing data including nighttime light (NTL) raster data of "VNP46A2 product"; daytime surface temperature (LST_MOD) and nighttime surface temperature (LST_MYD) raster data of "MOD11A2 product" and "MYD11A2 product"; and atmospheric pollutant concentration products of TROPOMI sensor, namely nitrogen dioxide (NO2), sulfur dioxide (SO2), and carbon monoxide (CO) raster data. The time-series raster data of NTL, LST_MOD, LST_MYD, NO2, SO2, and CO are resampled to 500 meters according to the NTL (500 meters) with the smallest spatial resolution, and unified into the WGS84 geographic coordinate system to obtain time-series remote sensing data.

[0039] In this embodiment, step 102 may further include: uploading the factory area outline vector file to a cloud geospatial data analysis platform, first using filterBounds to limit the ROI, and then batch filtering data from various types of satellites according to satellite, cloud cover threshold, and date range; subsequently synthesizing cloudless images and automatically synchronizing them to the local machine, ultimately obtaining a time-series remote sensing data of the factory area containing different satellites, yearly or monthly data, with a spatial resolution unified to a preset spatial resolution.

[0040] Step 103: Based on the factory area outline and time-series remote sensing data, determine the number of pixels for each enterprise.

[0041] In this embodiment, a pixel is the smallest resolvable unit of a digital image. A pixel is the ground area represented by each small square after the image is gridded and its corresponding data. The number of pixels is the number of grid units (pixels) falling within the factory area outline (enterprise vector boundary). When the spatial resolution of the digital image is 10m, one pixel corresponds to a ground area of ​​10m×10m.

[0042] In this embodiment, the temporal remote sensing data in the factory area outline region is first rasterized into a mask larger than the spatial resolution, with a mask value of 1 representing a pixel within the factory area. Then, the temporal remote sensing data is superimposed on the enterprise mask scene by scene and band by band. The number of times the effective observation (cloud mask = 0) pixels appear in all time phases within each enterprise mask is counted, and finally, an "Enterprise-Pixel Count" table is obtained, where each row gives the total number of cumulative pixels observed for an enterprise, which is used for subsequent production estimation or land use change monitoring.

[0043] Optionally, step 103 above includes: based on a GIS platform, spatially overlaying the factory area outline region and time-series remote sensing data, and counting the number of pixels covered by the factory area range vector corresponding to the factory area outline region.

[0044] Step 104: Based on the number of pixels, the outline area of ​​the factory area, and the time-series remote sensing data, determine the time-series values ​​of the remote sensing features of each enterprise.

[0045] In this embodiment, the time-series remote sensing data of multiple enterprises are first precisely registered with the outline areas of their respective factory areas to determine the time-series remote sensing data of each enterprise. Based on the number of pixels, the pixel values ​​falling within the outline area of ​​each enterprise are statistically analyzed periodically to form the time-series values ​​of remote sensing features, that is, the feature index sequence of each enterprise at different time points, which is used for subsequent change detection or operational status analysis of each enterprise.

[0046] Step 105: Divide the multiple enterprises into industry categories, and determine the remote sensing feature subsets of the enterprise subsets in each industry category based on the remote sensing feature time series values.

[0047] In this embodiment, all listed manufacturing companies can be classified according to industry standards (such as the National Industrial Classification of Economic Activities GB / T 4754) into... There are several different industry categories, each of which Includes several companies Each enterprise possesses fused multimodal remote sensing feature time-series numerical values: .

[0048] In this embodiment, the remote sensing feature time series values ​​are a one-dimensional sequence of indicators such as multi-band reflectance, vegetation index, nighttime light, surface temperature, and building density that are continuously extracted from satellite / UAV images and change over time.

[0049] In this embodiment, the industry category classification includes using an unsupervised clustering algorithm to first cluster multiple enterprises into several clusters based on remote sensing feature time series values, and then mapping the clusters to industry labels such as "manufacturing, logistics and warehousing, science and technology park, and agricultural processing" by referring to business registration or POI data.

[0050] In this embodiment, the remote sensing feature subset of the enterprise subset refers to a feature selection in each industry (such as random forest or LASSO regression based on Gini impurity) that retains only the features with the strongest discriminative power for that industry (e.g., retaining nighttime light and building density for manufacturing, and NDVI and surface temperature for agricultural processing).

[0051] Step 105 above includes: for each industry A subset of enterprises is used to calculate the time-series values ​​of each remote sensing feature and the enterprise's revenue growth label. The mutual information between them is shown in Equation (1):

[0052] (1)

[0053] In equation (1), This represents the k-th term of the time series numerical representation of remote sensing modal characteristics. For joint probability distribution, , These represent marginal probabilities. All remote sensing feature time-series data are scored and ranked, and the Top-K remote sensing feature time-series values ​​are selected to form a feature subset for this industry. ,Right now:

[0054] (2)

[0055] Optionally, step 105 above includes: first, performing NLP parsing on the enterprise registration information to obtain industry tags; then, using time window sliding statistics to form a high-dimensional vector of the remote sensing features mean, variance, and trend slope of each enterprise; and finally, performing clustering within the same industry to select the most representative subset of remote sensing features (such as the mean of heat extraction infrared values ​​in the steel industry + the slope of nighttime lights), thereby reducing dimensionality and highlighting industry commonalities.

[0056] Step 106: Based on the remote sensing feature subset, predict the enterprise data of the enterprise subset to obtain the enterprise prediction results.

[0057] In this embodiment, the enterprise forecast result is the result obtained after forecasting the enterprise data. The enterprise data is data related to the enterprise's operation and management status. For example, if the enterprise data can be the GDP indicator, then the enterprise forecast result is the enterprise's GDP indicator value for a future period of time.

[0058] In this embodiment, a small number of key indicators (such as nighttime light intensity, changes in factory area, and stockpile material volume) that are highly relevant to the target industry are extracted from satellite or aerial imagery. The selected sample of companies to be evaluated are then used to predict risk or production capacity through a pre-trained machine learning model, thereby outputting results such as credit rating, output or default probability for each company in the future.

[0059] To address the problems of traditional technologies, this disclosure proposes a satellite remote sensing-based enterprise data processing method. This method integrates multi-source remote sensing data, such as nighttime light, surface temperature (day and night temperature), and atmospheric emissions, and combines enterprise geographic location with remote sensing image raster pixels to construct an enterprise-level remote sensing feature library. By introducing a factory area buffer mechanism, multimodal resonant pixel extraction, and outlier removal algorithms, the method enhances the representation and stability of remote sensing features. Simultaneously, by combining industry coding and mutual information scoring methods, it establishes an adaptation relationship between industries and remote sensing modalities, enabling structured selection and classification prediction of remote sensing features. This method significantly improves the accuracy of enterprise operational status perception, the timeliness of prediction, and the reliability of macroeconomic indicator estimation, and has broad application prospects in scenarios such as economic forecasting, financial risk control, and industry analysis.

[0060] This method has been empirically validated on a large scale using a sample of listed manufacturing companies. The revenue prediction model, built based on multi-source remote sensing data including nighttime light, thermal radiation, and industrial gas emissions, achieves a prediction accuracy of 81.9% and an F1 score of 90.3% when employing the optimal data fusion strategy, significantly outperforming single-data-source models (e.g., using only thermal radiation data yields an accuracy of approximately 59% and an F1 score of approximately 69%). Among 1594 manufacturing companies listed for more than 8 years in the market, as many as 1346 companies achieved stable prediction accuracy exceeding 60% in the model, with nearly 1000 companies achieving accuracy exceeding 90%.

[0061] The enterprise data processing method based on satellite remote sensing provided in this disclosure firstly determines the factory outline area of ​​each enterprise in the remote sensing image based on a list of multiple enterprises; secondly, based on the factory outline area, it acquires corresponding time-series remote sensing data collected by various types of satellites; thirdly, based on the factory outline area and the time-series remote sensing data, it determines the number of pixels for each enterprise; fourthly, based on the number of pixels, the factory outline area, and the time-series remote sensing data, it determines the time-series values ​​of remote sensing features for each enterprise; then, it divides the multiple enterprises into industry categories, and based on the time-series values ​​of remote sensing features, it determines the remote sensing feature subsets of the enterprise subsets in each industry category; finally, based on the remote sensing feature subsets, it predicts the enterprise data of the enterprise subsets to obtain the enterprise prediction results. Therefore, by accurately matching multi-source satellite time-series remote sensing data with the factory area outline, the pixel-level dynamic information of enterprises is transformed into industry-comparable remote sensing feature time-series values, which significantly improves the real-time perception and prediction capabilities of cross-industry and cross-scale enterprise activities, and realizes early identification and early warning of industrial operation status, providing a high-confidence decision-making basis for precise government governance, financial risk control and supply chain optimization.

[0062] In some optional implementations of this disclosure, the above-mentioned determination of the factory area outline region of each enterprise in the remote sensing image based on the list of multiple enterprises includes: determining the geographical location and remote sensing image of each enterprise among the multiple enterprises based on the list of multiple enterprises; and obtaining the factory area outline region of each enterprise in the remote sensing image based on the geographical location and the remote sensing image.

[0063] In this optional implementation, based on a list of multiple enterprises, the geographical locations of the enterprises in the list are automatically determined nationwide by calling a map application interface and combining business registration information data; based on the geographical location, remote sensing images are overlaid to determine the factory area boundary and the spatial range of the enterprise's production area; the spatial range is saved as a vector format to obtain the factory area outline.

[0064] In this optional implementation, the execution entity running on the satellite remote sensing-based enterprise data processing method first performs semantic matching between the enterprise list and authoritative geographic databases (such as Gaode / Baidu POI, National Land Use Status Database) to obtain the coordinates of the "seed point" for each enterprise; then, it calls the remote sensing image of the area corresponding to the coordinates (such as through time-series retrieval via Google EarthEngine API), and uses the "object-oriented segmentation + deep edge detection" model (such as DeepLab-v3+ with fine-tuning on the NIR-RG band combination) to segment the image into plots; finally, using the seed point as the initial area, it performs morphological optimization based on the enterprise's industry priors (such as chemical plants usually include cooling towers, and steel plants must have blast furnaces), extracts the high-confidence factory area outline, and realizes end-to-end automation of "list in - vector out".

[0065] The optional implementation provides a method for obtaining the factory area outline region. By using the geographical location of the enterprise and remote sensing imagery, the outline region of each enterprise's factory area in the remote sensing imagery is obtained, which improves the reliability and accuracy of obtaining the factory area outline region.

[0066] In some optional implementations of this disclosure, determining the number of pixels for each enterprise based on the factory area outline and time-series remote sensing data includes: spatially overlaying the time-series remote sensing data with the factory area outline to obtain an overlaid image; and determining the number of pixels for each enterprise that cover the time-series remote sensing data based on the overlaid image.

[0067] In this optional implementation, the remote sensing images of each period are first spatially overlaid with the factory boundary, which is equivalent to "covering" the vector boundary on the raster image. All pixels falling within the boundary of a certain enterprise are counted as the number of pixels of that enterprise. By repeating this operation on each period of images, the change in the number of pixels of each enterprise at different time points can be obtained, providing a quantitative basis for subsequent analysis of production capacity, land expansion, etc.

[0068] The optional implementation provides a method for determining the number of pixels by spatially overlaying time-series remote sensing data with the outline area of ​​the factory area to obtain an overlaid image; based on the overlaid image, the number of pixels covered by the time-series remote sensing data for each enterprise is determined, providing a reliable method for obtaining the number of pixels.

[0069] In some optional implementations of this disclosure, the determination of the remote sensing feature time-series values ​​of each enterprise based on the number of pixels, the factory area outline, and time-series remote sensing data includes: dividing multiple enterprises into small enterprises and large and medium-sized enterprises based on the number of pixels; calculating a resonance index based on time-series observations extracted from time-series remote sensing data; clustering the factory area outline based on the resonance index to determine primary and secondary activity zones; using a set of pixels within a defined range centered on small enterprises as a buffer zone, and determining spatial weights within the buffer zone based on the distance of each pixel to the center of the buffer zone; calculating the remote sensing feature time-series values ​​of small enterprises based on the spatial weights and the time-series remote sensing data of small enterprises; extracting the remote sensing feature time-series values ​​of large and medium-sized enterprises from the time-series remote sensing data based on the primary and secondary activity zones; and using the remote sensing feature time-series values ​​of small enterprises and large and medium-sized enterprises as the remote sensing feature time-series values ​​of each enterprise.

[0070] In this optional implementation, the number of pixels can be used as a reference. And remote sensing observation capabilities are used to categorize small enterprises into large and medium-sized enterprises; if Enterprises are classified into large and medium-sized enterprises; if The company is classified as a small business.

[0071] In this optional implementation, the Remote Sensing Resonance Index (RSRI) is a spectral index used to quantify the severity of plant diseases (such as rice panicle diseases). Constructed based on reflectance in specific spectral bands, it can sensitively reflect spectral changes in plants when they are affected by disease. In the field of satellite remote sensing, the larger the absolute value of the RSRI, the more the performance of that pixel in that mode deviates from the overall average level at that time phase; it can be understood as a quantification of the degree of "resonance" or "anomaly" with the population.

[0072] In this optional implementation, various clustering algorithms, such as the K-Means algorithm, can be used based on the remote sensing resonance index (RSRI) to cluster the factory area outline and determine primary and secondary activity zones. These zones can include: high-activity zones (core areas), medium-activity zones (secondary activity zones), and inactive zones (background interference). For large and medium-sized enterprises, primary and secondary activity zones are defined, representing areas of active or inactive production under remote sensing observation. High-activity zones are extracted through clustering, and multimodal remote sensing feature values ​​are extracted using the vector range of these zones, thus enhancing the feature extraction process. Specifically, K-Means clustering (K=3) is performed on the RSRI activity map. Spatial post-processing opening and closing operations and connectivity analysis are used to fill gaps, forming closed activity zone outlines. Based on cluster labels, these areas are divided into high-activity zones (core areas), medium-activity zones (secondary activity zones), and inactive zones (background interference). Figure 2aAs shown, the unselected areas are the resonant pixels calculated for large enterprises, while the selected bright pixels are the medium-to-high activity areas extracted after clustering. The remote sensing feature time series values ​​of multiple modalities will be extracted from this factory area.

[0073] In this optional implementation, a set of pixels (buffer zone) within a certain range centered on the enterprise's factory area is defined. Spatial weights are assigned to each pixel based on its distance from the factory center. Then, the remote sensing observations of these pixels are weighted and averaged to obtain an enhanced representation of the remote sensing characteristics of small enterprises—a time-series numerical representation of the remote sensing features. Figure 2b As shown, a buffer zone of pixels is constructed for a small enterprise. The points in the image are buffer points, and the selected areas represent stable remote sensing feature time series values. Due to the small size of the factory area, the random error of pixel values ​​increases. Therefore, time series values ​​of multiple modal remote sensing data are extracted by combining buffer point extraction with factory point extraction, and then the mean is calculated using spatial weighted averaging to obtain the remote sensing feature time series values. Specifically:

[0074] As shown in equation (3), let the coordinates of the factory's centroid be... Any pixel The coordinates are If the buffer radius is defined as R, then the set of cells Bi within the buffer is:

[0075] (3)

[0076] The weighted average of remote sensing observations is performed using the method shown in Equation (4). In Equation (4), for modal variable X, the spatial weight w of each buffer cell q is... q for:

[0077] (4)

[0078] The enhanced remote sensing feature representation of small enterprises – the time series value of remote sensing features – is the weighted average result shown in equation (5):

[0079] (5)

[0080] In this optional implementation, the above-mentioned extraction of remote sensing feature time series values ​​of large and medium-sized enterprises from time series remote sensing data based on primary and secondary activity areas includes: extracting multimodal NTL, LST_MOD, LST_MYD, and CO remote sensing data time series values ​​in the high-activity area and medium-activity area, and weighting the average to obtain remote sensing feature time series values ​​that can reflect the scope of production and operation activities of large enterprises.

[0081] This optional implementation provides a method for determining the time-series data values ​​of remote sensing features. By mapping enterprise POIs, factory boundaries, and remote sensing raster, different pixel enhancement methods are adopted for enterprises of different sizes. A "resonance pixel extraction + spatial buffer enhancement" method is proposed. For large enterprises, resonance pixel enhancement is used to obtain the production activity area; for small enterprises with fewer remote sensing pixels, a spatial distance weighting strategy for constructing buffer points is used to enhance the remote sensing signal, improving the adaptability of remote sensing observation to the micro-activities of enterprises. Different methods are used for small enterprises and large and medium-sized enterprises to obtain the time-series values ​​of remote sensing features, thereby extracting different values ​​for different enterprises and improving the accuracy and reliability of the obtained time-series values ​​of remote sensing features. By proposing a factory area buffer enhancement mechanism and a multimodal resonance pixel extraction algorithm, even for enterprises with small factory areas and sparse pixel numbers, a representative and robust remote sensing feature set can be constructed, significantly improving the expressive power and predictive stability of remote sensing features, and solving the problem of missing remote sensing information at the micro-scale in traditional methods.

[0082] Optionally, the above method for determining the time-series values ​​of remote sensing features for each enterprise based on pixel count, factory area outline, and time-series remote sensing data includes: a method to more accurately calculate the "brightness time series" of remote sensing images for enterprises of different sizes. First, enterprises are divided into "small enterprises (point-like)" and "large and medium-sized enterprises (area-like)" based on pixel count. Then, a "resonance index" curve is calculated using observations from all time phases throughout the year to cluster the most active primary and secondary activity areas (factory buildings / storage yards vs. open spaces) within the factory area. For small enterprises, a circular buffer zone is drawn centered on the enterprise, and each pixel within it is weighted according to its distance from the center. The weighted pixel values ​​are then combined to form a remote sensing feature time series belonging only to that small enterprise. For large and medium-sized enterprises, a time series is directly extracted from each of the pre-divided primary and secondary activity areas and then spliced ​​together to form its feature. Finally, the sequences of the two types of enterprises are combined to obtain the time-series values ​​of the remote sensing features for each enterprise.

[0083] In some optional implementations of this disclosure, the above calculation of the resonance index based on the temporal observations extracted from the temporal remote sensing data includes: extracting the temporal observations from the temporal remote sensing data using a value extraction raster method; and, for each pixel in each modality, standardizing the temporal observations of that pixel in the temporal observations to obtain the resonance index of that pixel.

[0084] In this optional implementation, time-series observations of multiple remote sensing modalities (NTL, LST_MOD, LST_MYD, CO) are extracted using the GIS platform's value extraction raster method. After standardization, their resonance index (RSRI) is constructed to represent the intensity of the co-variation of the pixel in the multimodal data. The calculation formula is shown in equation (6):

[0085] (6)

[0086] In equation (6), For the first Each pixel in mode The value of μ ij Let σ be the mean of the i-th pixel across multiple modes (M types, M>1), and let σ be the standard deviation of the i-th pixel across multiple modes. A higher resonance index indicates that the pixel is an "activity hotspot" of multi-source remote sensing modes at the current moment, and is more likely to reflect real economic behavior.

[0087] The method for obtaining the resonance index of a pixel provided by this optional implementation uses a value extraction raster method to extract time-series observations from time-series remote sensing data; for pixels under each mode, the time-series observations of that pixel are standardized, thereby improving the reliability of obtaining the resonance index of the pixel.

[0088] In some optional implementations of this disclosure, the above-mentioned prediction of enterprise data based on a subset of remote sensing features to obtain enterprise prediction results includes: inputting the subset of remote sensing features into a pre-trained enterprise data model to obtain the enterprise prediction results output by the enterprise data model.

[0089] In this optional implementation, the remote sensing feature subset refers to the values ​​of variables (such as nighttime light intensity, changes in yard area, transport vehicle density, and factory expansion rate) that best reflect the company's operating status after being filtered from the time series values ​​of remote sensing features.

[0090] Pre-trained enterprise data models: These are generally machine learning models (such as XGBoost, LSTM, or Transformer) trained using multi-source data, including historical remote sensing, financial, and business data.

[0091] In this optional implementation, a subset of remote sensing features extracted from satellite imagery (e.g., nighttime light intensity, factory area, number of chimneys, etc. in an industrial park) is fed into a pre-trained enterprise data model using a large number of enterprise samples. The enterprise data model maps these remote sensing features into quantitative predictions of each enterprise's production capacity, emissions, or risks, ultimately directly outputting "enterprise prediction results" (e.g., an estimated output or carbon emissions for a chemical plant next month). The entire process requires no additional data reporting from enterprises; it relies entirely on remote sensing imagery and AI for automated monitoring, improving the reliability of the enterprise prediction results.

[0092] In some optional implementations of this disclosure, the aforementioned enterprise forecasting results include: macroeconomic indicators. The enterprise data model is trained through the following steps: based on the acquired remote sensing feature set, a training set and a test set are determined; based on the training set, a revenue forecasting model predicting the growth or decline direction of the future annual operating revenue of each enterprise is trained to obtain a trained revenue forecasting model; the trained revenue forecasting model is used to forecast the test set to obtain forecasting results; based on the forecasting results, satellite-observable enterprises among multiple enterprises are identified; a set of macroeconomic feature vectors of satellite-observable enterprises is extracted from the remote sensing feature subset; based on the set of macroeconomic feature vectors, a pre-constructed economic trend forecasting model is trained to obtain a trained economic trend forecasting model, which is used to forecast macroeconomic indicators; the revenue forecasting model and the economic trend forecasting model are used as the enterprise data model.

[0093] Based on the core position of listed manufacturing companies in the national economic system, this paper further constructs a mapping model from micro-level remote sensing features to macroeconomic variables. Specifically, companies whose revenue forecasting model achieves a prediction accuracy exceeding a preset threshold (e.g., 60%) across various industries are defined as satellite-observable companies. Because these companies possess high remote sensing representation capabilities and consistency with financial responses, their remote sensing features are representative of macroeconomic changes. Therefore, the remote sensing feature variables extracted from this subset of companies are weighted and aggregated according to dimensions such as industry, time period, and geographical region to form a set of macroeconomic feature vectors.

[0094] In this optional implementation, the prediction accuracy primarily derives from the cross-validation results of the revenue prediction model. The specific steps are as follows: A training set is constructed using historical remote sensing enhanced features and corresponding labels indicating actual revenue increases or decreases; samples are cross-validated by time or enterprise (e.g., hold-out method, k-fold method); the accuracy of the model in predicting the "direction of revenue increase or decrease" on the validation set is calculated; enterprises with accuracy exceeding a preset threshold (e.g., 60%) in each industry are defined as "remote sensing observable enterprises." This accuracy metric is used to select enterprise samples with high modeling quality and strong consistency between remote sensing features and financial performance, thereby enhancing the representativeness and stability of the macro-modeling stage.

[0095] Furthermore, using the aforementioned set of macroeconomic feature vectors as input variables, a classification and regression model for macroeconomic indicator forecasting is constructed. The selected macroeconomic forecasting indicators include, but are not limited to, whether the Purchasing Managers' Index (PMI) is above the expansion / contraction threshold and whether the quarterly GDP growth rate is higher than last year. In the specific implementation, the random forest algorithm is preferred for the economic trend forecasting model; alternatively, XGBoost, logistic regression, and other models can be used as baselines. Through model training and validation, a forecasting system is established that starts with enhanced remote sensing features of enterprises, aggregates them across industries, and ultimately maps them to the macroeconomic state, constructing a four-level linked chain modeling path of "remote sensing pixel-enterprise-industry-macroeconomic."

[0096] In this optional implementation, a "revenue forecasting model"—a supervised classifier that takes remote sensing features (nighttime lights, factory area, number of vehicles in parking lots, and other satellite-observable variables) as input and outputs the rise and fall of the company's future annual operating revenue—is first trained on the training set and validated on the test set to screen out "satellite-observable enterprises" whose remote sensing signals are highly correlated with their business performance and can therefore be continuously monitored by satellites. Then, the remote sensing features of these enterprises are extracted into macro-feature vectors (regional-industry aggregated indicators), and these vectors are used to train a second economic trend forecasting model, enabling it to output the trends of macroeconomic indicators such as GDP and industrial added value. Finally, these two connected models are packaged together and collectively referred to as the "enterprise data model," realizing a one-stop closed loop from space imagery to macroeconomic forecasting.

[0097] The optional implementation method for training enterprise data models first uses a revenue forecasting model to predict the test set; based on the prediction results, satellite-observable enterprises are identified; a set of macroeconomic feature vectors is extracted, and an economic trend prediction model is trained using the set of macroeconomic feature vectors. The trained economic trend prediction model is then used as the enterprise data model, realizing a hierarchical mapping and prediction from enterprise-level business behavior to macroeconomic indicators. This invention is the first to realize a multi-level modeling path of "remote sensing pixel - enterprise - industry - macroeconomy". By utilizing the remote sensing features and prediction results of "satellite-observable enterprises", a macroeconomic prediction model for PMI and GDP is constructed, forming a structured remote sensing modeling system that reasones step by step from the micro-enterprise level to national-level economic indicators, which has strong systematicity and scalability.

[0098] In some optional implementations of this disclosure, the above-mentioned revenue prediction model based on the training set, which predicts the growth or decline direction of the future annual operating revenue of each enterprise, to obtain the trained revenue prediction model includes: encoding each industry category to obtain an industry encoding vector; concatenating the industry encoding vector with the remote sensing features of the corresponding industry in the training set to obtain a new training set; and using the new training set to train the revenue prediction model based on the growth or decline direction of the future annual operating revenue of each enterprise, to obtain the trained revenue prediction model.

[0099] In this optional implementation, a subset of remote sensing features is matched with listed companies in various industries. To predict the direction of growth or decline in a company's future annual operating revenue, and to train revenue forecasting models related to the company's operating revenue. , specifically as formula (7):

[0100] (7)

[0101] In equation (7), Indicates enterprise Feature vectors on a subset of remote sensing features Classification models such as Ridge Regression, Logistic Regression, and Random Forest can be selected.

[0102] Alternatively, it can be modeled in combination, incorporating industry category one-hot encoded vectors. This data is then concatenated with remote sensing features and used as input to a unified model, i.e., one-hot encoding is used to obtain the industry-specific encoding vector.

[0103] (8)

[0104] In equation (8), the j-th term is 1, indicating that the enterprise belongs to the industry. There are m industries in total.

[0105] In this optional implementation, the industry to which a company belongs is first represented by a string of numbers (industry coding vector). This string of numbers is then combined with the company's features extracted from satellite images to form an "enhanced" training sample. These enhanced samples are then used to train a binary classification model, enabling it to determine whether a company's revenue will increase or decrease next year. Once trained, this model can be directly used to predict the revenue direction of any company, improving the reliability of the revenue prediction model training. An industry classification coding and mutual information scoring mechanism is introduced to identify differences in the response of different industries to remote sensing modalities, automatically selecting the feature combination with the highest suitability for modeling. This mechanism overcomes the industry non-discrimination problem of "one-size-fits-all remote sensing models," improving the model's applicability and modeling accuracy across different manufacturing sub-sectors.

[0106] Further reference Figure 3 As an implementation of the methods shown in the above figures, this disclosure provides an embodiment of an enterprise data processing device based on satellite remote sensing, which is similar to... Figure 1 Corresponding to the method embodiments shown, this device can be specifically applied to various electronic devices.

[0107] like Figure 3 As shown, the enterprise data processing device 300 based on satellite remote sensing provided in this embodiment includes: a region determination unit 301, an acquisition unit 302, a quantity determination unit 303, a value determination unit 304, a subset determination unit 305, and a prediction unit 306. The region determination unit 301 can be configured to determine the factory outline area of ​​each enterprise in the remote sensing image based on a list of multiple enterprises. The acquisition unit 302 can be configured to acquire corresponding time-series remote sensing data collected by various types of satellites based on the factory outline area. The quantity determination unit 303 can be configured to determine the number of pixels of each enterprise based on the factory outline area and the time-series remote sensing data. The value determination unit 304 can be configured to determine the time-series value of remote sensing features of each enterprise based on the number of pixels, the factory outline area, and the time-series remote sensing data. The subset determination unit 305 can be configured to classify multiple enterprises into industry categories and determine the remote sensing feature subset of the enterprise subset in each industry category based on the time-series value of the remote sensing features. The aforementioned prediction unit 306 can be configured to predict enterprise data of a subset of enterprises based on a subset of remote sensing features, and obtain enterprise prediction results.

[0108] In this embodiment, the specific processing and technical effects of the following components in the enterprise data processing device 300 based on satellite remote sensing—namely, the region determination unit 301, the acquisition unit 302, the quantity determination unit 303, the value determination unit 304, the subset determination unit 305, and the prediction unit 306—can be found by referring to [reference needed]. Figure 1 The relevant descriptions of steps 101, 102, 103, 104, 105, and 106 in the corresponding embodiments will not be repeated here.

[0109] In some embodiments of this disclosure, the aforementioned region determination unit 301 is configured to: determine the geographical location and remote sensing image of each of the multiple enterprises based on a list of multiple enterprises; and obtain the factory area outline region of each enterprise in the remote sensing image based on the geographical location and the remote sensing image.

[0110] In some embodiments of this disclosure, the quantity determination unit 303 is configured to: spatially overlay time-series remote sensing data with the factory area outline to obtain an overlaid image; and determine the number of pixels covered by the time-series remote sensing data for each enterprise based on the overlaid image.

[0111] In some embodiments of this disclosure, the numerical determination unit 304 is configured to: divide multiple enterprises into small enterprises and large and medium-sized enterprises based on the number of pixels; calculate a resonance index based on time-series observations extracted from time-series remote sensing data; cluster the factory area outline region based on the resonance index to determine primary and secondary activity zones; use a set of pixels within a set range centered on small enterprises as a buffer zone, and determine spatial weights in the buffer zone based on the distance of each pixel to the center of the buffer zone; calculate the time-series remote sensing feature values ​​of small enterprises based on the spatial weights and the time-series remote sensing data of small enterprises; extract the time-series remote sensing feature values ​​of large and medium-sized enterprises from the time-series remote sensing data based on the primary and secondary activity zones; and use the time-series remote sensing feature values ​​of small enterprises and large and medium-sized enterprises as the time-series remote sensing feature values ​​of each enterprise.

[0112] In some embodiments of this disclosure, the numerical determination unit 304 is further configured to: extract temporal observations from temporal remote sensing data using a value extraction raster method; and, for each pixel in each modality, standardize the temporal observations of that pixel in the temporal observations to obtain the resonance index of that pixel.

[0113] In some embodiments of this disclosure, the prediction unit 306 is configured to input a subset of remote sensing features into a pre-trained enterprise data model to obtain the enterprise prediction result output by the enterprise data model.

[0114] In some embodiments of this disclosure, the prediction unit 306 is further configured to: determine a training set and a test set based on the acquired remote sensing feature set; train a revenue prediction model based on the training set to predict the growth or decline direction of the future annual operating revenue of each enterprise, thereby obtaining a trained revenue prediction model; use the trained revenue prediction model to predict the test set, thereby obtaining a prediction result; based on the prediction result, identify satellite-observable enterprises among multiple enterprises; extract a set of macroeconomic feature vectors of satellite-observable enterprises from the remote sensing feature subset; train a pre-constructed economic trend prediction model based on the set of macroeconomic feature vectors, thereby obtaining a trained economic trend prediction model, which is used to predict macroeconomic indicators; and use the revenue prediction model and the economic trend prediction model as enterprise data models.

[0115] In some embodiments of this disclosure, the prediction unit 306 is further configured to: encode each industry category to obtain an industry coding vector; concatenate the industry coding vector with the remote sensing features of the corresponding industry in the training set to obtain a new training set; and use the new training set to train a revenue prediction model that predicts the growth or decline direction of the future annual operating revenue of each enterprise, thereby obtaining a trained revenue prediction model.

[0116] The enterprise data processing apparatus based on satellite remote sensing provided in the embodiments of this disclosure firstly involves a region determination unit 301 determining the factory outline area of ​​each enterprise in a remote sensing image based on a list of multiple enterprises; secondly, an acquisition unit 302 acquiring corresponding time-series remote sensing data collected by various types of satellites based on the factory outline area; thirdly, a quantity determination unit 303 determining the number of pixels for each enterprise based on the factory outline area and the time-series remote sensing data; fourthly, a value determination unit 304 determining the time-series value of remote sensing features for each enterprise based on the number of pixels, the factory outline area, and the time-series remote sensing data; then, a subset determination unit 305 classifying the multiple enterprises into industry categories and determining the remote sensing feature subset of the enterprise subset in each industry category based on the time-series value of remote sensing features; finally, a prediction unit 306 predicting the enterprise data of the enterprise subset based on the remote sensing feature subset to obtain the enterprise prediction result. Therefore, by integrating multi-source satellite time-series remote sensing data, precise positioning of factory area outlines, and industry-specific remote sensing feature extraction, a progressively refined analysis chain of "region-pixel-feature-industry" is constructed. This not only significantly improves the dynamic perception capability and data prediction accuracy of the actual production status of enterprises, but also enables low-cost and automated supervision of cross-industry, large-scale enterprise groups, providing governments or investment institutions with highly timely and reliable macro-decision-making basis.

[0117] The collection, storage, use, processing, transmission, provision, and disclosure of enterprise information or user personal information involved in the technical solution disclosed herein comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0118] According to embodiments of this disclosure, this disclosure also provides an electronic device, a readable storage medium, and a computer program product.

[0119] Figure 4 A schematic block diagram of an example electronic device 400 that can be used to implement embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their patterns are merely examples and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0120] like Figure 4As shown, device 400 includes a computing unit 401, which can perform various appropriate actions and processes based on a computer program stored in read-only memory (ROM) 402 or a computer program loaded from storage unit 408 into random access memory (RAM) 403. RAM 403 may also store various programs and data required for the operation of device 400. The computing unit 401, ROM 402, and RAM 403 are interconnected via bus 404. Input / output (I / O) interface 405 is also connected to bus 404.

[0121] Multiple components in device 400 are connected to I / O interface 405, including: input unit 406, such as keyboard, mouse, etc.; output unit 407, such as various types of monitors, speakers, etc.; storage unit 408, such as disk, optical disk, etc.; and communication unit 409, such as network card, modem, wireless transceiver, etc. Communication unit 409 allows device 400 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0122] The computing unit 401 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 401 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 401 performs the various methods and processes described above, such as a satellite remote sensing-based enterprise data processing method. For example, in some embodiments, the satellite remote sensing-based enterprise data processing method can be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 408. In some embodiments, part or all of the computer program can be loaded and / or installed on device 400 via ROM 402 and / or communication unit 409. When the computer program is loaded into RAM 403 and executed by the computing unit 401, one or more steps of the satellite remote sensing-based enterprise data processing method described above can be performed. Alternatively, in other embodiments, computing unit 401 may be configured by any other suitable means (e.g., by means of firmware) to perform enterprise data processing methods based on satellite remote sensing.

[0123] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0124] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable satellite remote sensing-based enterprise data processing device, such that when executed by the processor or controller, the patterns / operations specified in the flowcharts and / or block diagrams are implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0125] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0126] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0127] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0128] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.

[0129] The foregoing description of specific exemplary embodiments of this disclosure is for illustrative and explanatory purposes. These descriptions are not intended to limit this disclosure to the precise forms disclosed, and it will be apparent that many changes and variations can be made in accordance with the foregoing teachings. The exemplary embodiments were chosen and described in order to explain the specific principles of this disclosure and their practical application, thereby enabling those skilled in the art to implement and utilize various different exemplary embodiments of this disclosure, as well as various different choices and variations. The scope of this disclosure is intended to be defined by the claims and their equivalents.

Claims

1. A method for processing enterprise data based on satellite remote sensing, characterized in that, The method includes: Based on a list of multiple companies, the outline areas of each company's factory area in the remote sensing image were determined. Based on the outline area of ​​the factory area, corresponding time-series remote sensing data collected by various types of satellites are acquired; the time-series remote sensing data includes: nighttime light, daytime surface temperature, nighttime surface temperature, nitrogen dioxide, sulfur dioxide, and carbon monoxide raster data. Determining the number of pixels for each enterprise based on the factory area outline and the time-series remote sensing data includes: spatially overlaying the time-series remote sensing data with the factory area outline to obtain an overlaid image; and determining the number of pixels for each enterprise that cover the time-series remote sensing data based on the overlaid image. Based on the number of pixels, the factory area outline, and the time-series remote sensing data, determining the time-series values ​​of remote sensing features for each enterprise includes: classifying the multiple enterprises into small enterprises and large / medium-sized enterprises based on the number of pixels; calculating a resonance index based on time-series observations extracted from the time-series remote sensing data; clustering the factory area outline based on the resonance index to determine primary and secondary activity zones; using a set of pixels within a defined range centered on the small enterprises as a buffer zone, determining spatial weights within the buffer zone based on the distance of each pixel to the center of the buffer zone; and determining the spatial weights and the time-series remote sensing data of the small enterprises. The remote sensing feature time series values ​​of the small enterprises are calculated; based on the primary and secondary activity areas, the remote sensing feature time series values ​​of the large and medium-sized enterprises are extracted from the time series remote sensing data; the remote sensing feature time series values ​​of the small enterprises and the large and medium-sized enterprises are used as the remote sensing feature time series values ​​of each enterprise; the calculation of the resonance index based on the time series observation values ​​extracted from the time series remote sensing data includes: extracting time series observation values ​​from the time series remote sensing data using a value extraction raster method; for each pixel under each mode, the time series observation values ​​of that pixel are standardized to obtain the resonance index of that pixel; The multiple enterprises are classified into industry categories, and based on the remote sensing feature time series values, the remote sensing feature subsets of the enterprise subsets in each industry category are determined. Based on the remote sensing feature subset, predictions are made on the enterprise data of the enterprise subset to obtain enterprise prediction results, which include carbon emission estimates.

2. The method according to claim 1, characterized in that, The process of determining the factory outline areas of each enterprise in the remote sensing image based on a list of multiple enterprises includes: Based on a list of multiple companies, the geographical location and remote sensing image of each company within the list were determined. Based on the geographical location and the remote sensing image, the outline area of ​​the factory area of ​​each enterprise in the remote sensing image is obtained.

3. The method according to claim 1, characterized in that, The prediction of enterprise data based on the remote sensing feature subset, to obtain enterprise prediction results, includes: The remote sensing feature subset is input into a pre-trained enterprise data model to obtain the enterprise prediction results output by the enterprise data model.

4. A satellite remote sensing-based enterprise data processing device, characterized in that, The device includes: The region determination unit is configured to determine the outline area of ​​each enterprise's factory area in the remote sensing image based on a list of multiple enterprises. The acquisition unit is configured to acquire corresponding time-series remote sensing data collected by various types of satellites based on the outline area of ​​the factory area; the time-series remote sensing data includes: nighttime light, daytime surface temperature, nighttime surface temperature, nitrogen dioxide, sulfur dioxide, and carbon monoxide raster data. The quantity determination unit is configured to determine the number of pixels for each enterprise based on the factory area outline and the time-series remote sensing data; the quantity determination unit is configured to: spatially overlay the time-series remote sensing data with the factory area outline to obtain an overlaid image; and determine the number of pixels for each enterprise that cover the time-series remote sensing data based on the overlaid image. A numerical determination unit is configured to determine the time-series values ​​of remote sensing features for each enterprise based on the number of pixels, the factory area outline, and the time-series remote sensing data. The numerical determination unit is further configured to: classify the multiple enterprises into small enterprises and large / medium-sized enterprises based on the number of pixels; calculate a resonance index based on time-series observations extracted from the time-series remote sensing data; cluster the factory area outline based on the resonance index to determine primary and secondary activity zones; use a set of pixels within a defined range centered on the small enterprises as a buffer zone, and determine spatial weights within the buffer zone based on the distance of each pixel to the center of the buffer zone; and... Based on the spatial weights and the time-series remote sensing data of the small enterprises, the time-series values ​​of the remote sensing features of the small enterprises are calculated; based on the primary and secondary activity areas, the time-series values ​​of the remote sensing features of the large and medium-sized enterprises are extracted from the time-series remote sensing data; the time-series values ​​of the remote sensing features of the small enterprises and the large and medium-sized enterprises are used as the time-series values ​​of the remote sensing features of each enterprise; the value determination unit is further configured to: extract time-series observations from the time-series remote sensing data using a value extraction raster method; for pixels under each mode, the time-series observations of the pixel in the time-series observations are standardized to obtain the resonance index of the pixel; The subset determination unit is configured to classify the plurality of enterprises into industry categories and, based on the remote sensing feature time series values, determine the remote sensing feature subsets of the enterprise subsets in each industry category. The prediction unit is configured to predict enterprise data of the enterprise subset based on the remote sensing feature subset to obtain enterprise prediction results, which include carbon emission estimates.

5. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 3.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 3.

Images