Industrial heat source development activity monitoring method based on multi-source time sequence remote sensing image

Abstract

The invention relates to an industrial heat source development activity monitoring method based on a multi-source time sequence remote sensing image, and the method combines high-spatial resolution remote sensing image data, can clearly recognize small facilities in industrial land, can automatically and accurately complete the recognition, classification and production inversion of an industrial heat source, and improves the efficiency of industrial heat source development. Therefore, the method is used for large-scale industrial activity monitoring and trend analysis. In combination with the multi-source remote sensing image, the detailed industrial heat source list is quickly and effectively acquired, and the detailed industrial heat source list comprises attribute information such as spatial position distribution, quantity, scale, type, running tense and production capacity. Therefore, the method is of great significance to industrial resource mastering, ecological environment influence and comprehensive evaluation thereof, is beneficial to deepening understanding of the emission mode of greenhouse gas in industrial combustion, and is beneficial to development of energy conservation and emission reduction work.

Description

Technical Field

[0001] The present invention relates to a method for monitoring industrial heat source development activities based on multi-source time series remote sensing images, and belongs to the technical field of remote sensing geoscience applications. Technical Background

[0002] At present, the monitoring of terrestrial industrial heat source development activities at home and abroad mainly includes field survey methods and monitoring methods based on remote sensing image data. Among them, monitoring based on remote sensing image data can be divided into industrial monitoring based on low spatial resolution fire data products and industrial monitoring based on medium spatial resolution optical remote sensing images.

[0003] Chen Xinxin et al. used a sample plot survey to identify industrial locations. Wang Xueyuan et al. employed random interviews and questionnaires within the region. Field sampling is costly in terms of manpower, time, and funds, limiting the scope and number of surveys. Satellite remote sensing Earth observation technology, compared to traditional survey techniques, offers a wide range, short cycle time, high accuracy, and low cost, providing an opportunity for large-scale monitoring of industrial heat source development activities.

[0004] Industrial development and production activities are often accompanied by heat emissions from combustion. In industrial monitoring research using low-spatial-resolution fire data products, the data involved comes from sensors specifically designed to detect thermal anomalies in combustion. High-temperature combustion activity is detected based on elevated thermal radiation in the thermal infrared and mid-infrared bands. As early as 1973, Croft used night-light remote sensing to monitor oilfield combustion activity, demonstrating for the first time that night-light remote sensing could be used to identify spatiotemporal information related to oil and gas production. In 2006, Giglio et al. analyzed the MODIS Active Fire product to determine the global distribution of fire points. Since then, numerous studies have been conducted on oil and gas combustion characterization and detection using thermal infrared sensors or photomultiplier tube (PMT) visible sensor remote sensing products. For example, global natural gas combustion activity has been detected based on ATSR along-track radiometer data and DMSP-OLS night-light products. Using the Suomi-NPP VIIRS nighttime fire product, a multispectral high-temperature system was constructed to estimate natural gas flaring. Object-oriented extraction methods were used to report a global industrial heat source inventory. Franklin et al. also used this data to identify the location of oil and gas development in the Eagle Ford Shale from 2012 to 2016 using spatiotemporal hierarchical clustering and fitted a polynomial regression model to monthly flaring. Xu used the Himarawi-8 to estimate the radiative power generated by fires in Asia and Australia. The high temporal resolution and global coverage of these algorithmic data sources have enabled this data and its derivative products to be widely used to address scientific questions such as global ecological responses and greenhouse gas emissions, becoming a mainstream tool for monitoring global burning activity.

[0005] In recent years, a small number of studies have utilized infrared bands from medium-resolution optical remote sensing imagery to detect high-temperature thermal anomalies on land. Murphy et al. primarily used the near-infrared, shortwave infrared 1, and shortwave infrared 2 bands of Landsat-8 OLI, as well as their pairwise ratios, as indicators for detecting burning activity. Schroeder et al. proposed a thermal anomaly detection algorithm for Landsat-7 ETM+ and adapted it for Landsat-8 OLI imagery. Zhang Wenlei et al. validated the feasibility of burn detection using clustering methods and area brightness temperature models based on Landsat-8 OLI data. Boschetti et al. used Landsat data to detect burned areas at a spatial resolution of 30 meters and to display their time series distribution. Schroeder et al. studied a fire point detection algorithm using Landsat-8 OLI data. Liu et al. analyzed the spectral characteristics of thermal anomalies in Sentinel-2 MSI imagery to monitor industrial burning activity. Roy et al. combined Landsat-8 and Sentinel-2 to monitor burning changes. Summary of the Invention

[0006] The technical problem to be solved by this invention is to overcome the shortcomings of existing technologies and propose a method for monitoring industrial heat source development activities based on multi-source time-series remote sensing imagery. This method uses high-spatial-resolution digital orthophotos, multi-source long-time series of medium-resolution optical remote sensing imagery, and VIIRS fire data products as its data foundation. By combining deep learning, cluster analysis, spatial filtering, histogram matching, minimum distance classification, and fitting inversion, this method automatically extracts an inventory of industrial heat sources (including their spatial location distribution, quantity, scale, type, operating status, production volume, and other attribute information) and constructs an integrated database for industrial heat sources encompassing "spatial location extraction, combustion activity detection, and production status analysis."

[0007] In order to solve the above technical problems, the present invention proposes a technical solution: a method for monitoring industrial heat source development activities based on multi-source time series remote sensing images, comprising the following steps:

[0008] The first step is to acquire and process multi-source remote sensing image data, including Landsat and Sentinel-2MSI optical images. High-resolution digital orthophotos of the study area are also acquired, along with long-term series of VIIRS Nightfire and VIIRS Active Fire data products covering the area, and pre-process the relevant data.

[0009] The second step is to use the Faster R-CNN deep learning model to detect industrial equipment targets in industrial sites using high-resolution imagery. The main steps are: training sample preparation, model training, and target detection. After obtaining preliminary detection results from the deep learning model, image processing techniques are used to remove noise to obtain the final, accurate spatial location identification results of the industrial sites.

[0010] The third step, industrial heat source detection, includes the following steps:

[0011] 3.1、Constructing high temperature thermal anomaly index TAI=(ρ far-SWIR -ρ near-SWIR ) / ρ NIR , where ρ far-SWIR Represents the reflectivity of the far- and short-wave infrared bands, ρ near-SWIR Represents the reflectivity in the near-shortwave infrared band, ρ NIR Represents the reflectance of the near-infrared band; for Sentinel-2MSI images, TAI = (B12-B11) / B8a; for Landsat-8OLI images, TAI = (B7-B6) / B5;

[0012] 3.2. After sampling, the TAI results are threshold segmented based on the image spectral characteristics. Combined with the industrial land locations extracted in the second step, the buffer zone analysis and the time persistence characteristics are used to extract the locations of long-term burning industrial heat sources.

[0013] Step 4: Industrial heat source classification and temperature characteristics analysis

[0014] 4.1. Generate a 500-meter buffer zone based on the center point of each heat source. Then, using VIIRS ActiveFire and VIIRS Nightfire data, count the fire points that fall within each industrial heat source buffer zone on an annual basis to obtain annual temperature data for each industrial heat source. Construct a temperature histogram for each industrial heat source and normalize it to a ratio.

[0015] 4.2. Using the minimum distance classification method, match the industrial heat source temperature histogram with the temperature template of the known industrial type. The category with the smallest distance is used as the output category of the object to complete the classification of the industrial heat source.

[0016] Step 5: Modeling and inversion of industrial production activities

[0017] 5.1. Fit the annual temperature data of various industrial heat sources with the industrial output of their corresponding categories to construct a regression model. Then, based on this model, invert the annual output of each industrial object to reflect the production activities and change trends of regional industrial heat sources.

[0018] 5.2. Collect the total output value data of high-energy-consuming industries from the statistical yearbooks of the covered area for many years, and perform polynomial regression fitting on the annual temperature data of classified industrial heat sources and the output value of the corresponding industrial categories in the municipal administrative districts;

[0019] 5.3. Based on the regression model, invert the annual output of each industrial heat source and analyze the trend of production activities;

[0020] Step 6: Spatiotemporal Statistics and Database Construction

[0021] 6.1. Integrate the spatial location, classification, temperature, and output attributes of industrial heat sources to build an industrial heat source database covering the entire region;

[0022] 6.2. Calculate the temporal and spatial distribution of industrial heat sources by year and city or county, and assess the history and current status of industrial development.

[0023] The data sources used in this paper come from multiple sensor platforms (including high-resolution digital orthophotos, Landsat imagery, Sentinel-2MSI imagery, and VIIRS fire data products). The integration of these multiple data sources for industrial heat source detection and classification represents an innovation in data application. This paper constructs a batch processing framework centered around the Thermal Anomaly Index (TAI) and infrared band features. Combined with deep learning and minimum distance classification, this framework can automatically and accurately identify, classify, and invert production volumes of industrial heat sources at the provincial level. This allows for large-scale industrial activity monitoring and trend analysis, providing an important theoretical foundation for industrial energy consumption management and environmental assessment.

[0024] The relevant data extraction, model training, temperature statistics and yield inversion processes of the present invention are all implemented through the GEE cloud platform, the Faster R-CNN deep learning framework and the polynomial regression model, which significantly reduces manual participation and improves analysis efficiency.

[0025] In summary, the execution steps of the method of the present invention are highly systematic and effective in extracting the spatial distribution of industrial heat sources, classifying them, and analyzing production activities. Currently, there is limited data on the detailed production dynamics of high-energy-consuming industrial heat sources. This invention utilizes long-term series (2000–2023) of remote sensing data at multiple spatial scales to achieve high-precision classification and production estimation of industrial heat sources through matching thermal anomaly characteristics with temperature templates. This invention will help establish a dynamically updated industrial heat source database, supplementing the deficiencies of existing industrial statistical data, and providing a scientific basis for formulating industrial energy conservation and emission reduction policies and evaluating the impact of industrial heat sources on the environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] The present invention will be further described below with reference to the accompanying drawings.

[0027] Figure 1It is a flow chart of the monitoring method of industrial heat source development activities based on multi-source time series remote sensing images.

[0028] Figure 2 It is a flowchart for industrial land identification.

[0029] Figure 3 This is an example graph of the sample set.

[0030] Figure 4 This is the characteristic distribution map of industrial heat sources in Sentinel-2MSI images.

[0031] Figure 5 It is the TAI index image of different types of industrial heat sources in the Landsat series images.

[0032] Figure 6 This is the characteristic distribution diagram of industrial heat sources in the time series night fire dataset, (a) steel plant, (b) cement plant, (c) coal chemical plant, (d) oil and gas development; the third column is the clustering characteristics of thermal anomaly points with unchanged spatial position; the fourth column is the temperature characteristics of different industrial heat sources.

[0033] Figure 7 It is the location spatial distribution map of industrial sites.

[0034] Figure 8 This is a flowchart for analyzing industrial production activities, (a) technical flowchart; (b) conceptual diagram of the polynomial regression model inversion of industrial heat source objects. DETAILED DESCRIPTION

[0035] The present invention is described in detail below with reference to the accompanying drawings to make the technical route and operation steps of the present invention clearer. The example of the present invention uses the Texas area of ​​the United States as the study area. Based on the data of high spatial resolution (<1 meter) digital orthophotos, combined with image processing technologies such as deep learning and cluster analysis, the spatial location distribution of industrial land in the entire region is automatically extracted. And based on the spatial location extracted, multi-source long-term series of medium-resolution optical remote sensing image data (Landsat-5TM from 2000 to 2011, Landsat-7ETM+ from 2000 to 2013, Landsat-8OLI from 2013 to 2023, Sentinel-2MSI from 2015 to 2022) is used to construct a combustion detection model for industrial development activities, identify whether each industrial land is an industrial heat source, and analyze its spatiotemporal distribution from 2000 to 2023 (24 years). Building on these results, we further integrated long-term VIIRS fire data products (VIIRS Active Fire and VIIRS Nightfire) to subdivide industrial heat sources based on the different temperature characteristics of different industrial heat sources. We then fitted the annual temperature data with publicly available industrial enterprise production data to construct a polynomial regression model to estimate the combustion volume of industrial objects, thereby understanding the changing trends of industrial production.

[0036] Figure 1 This is a flow chart of a method for monitoring industrial heat source development activities based on multi-source time series remote sensing images according to an embodiment of the present invention. The specific steps are as follows:

[0037] The first step is to acquire and process multi-source remote sensing data. This includes Landsat imagery (Landsat-5TM from 2001–2011, Landsat-7ETM+ from 2000–2013, and Landsat-8OLI from 2013–2023) and Sentinel-2MSI optical imagery (2015–2023). Digital orthophotos with a spatial resolution of less than 1 meter are also acquired within the study area. Long-term series of VIIRS Nightfire and VIIRS Active Fire data products are collected and preprocessed.

[0038] In this step, high-resolution imagery covering the entire region in 2023, acquired through production, will be used to identify the spatial locations of industrial land. Furthermore, Landsat and Sentinel-2 MSI imagery covering the Texas region from January 1, 2000, to December 31, 2023 (Landsat-5TM from 2000–2011, Landsat-7ETM+ from 2000–2013, Landsat-8OLI from 2013–2023, and Sentinel-2 MSI from 2015–2023) will be collected for industrial heat source detection and analysis. Furthermore, VIIRS Nightfire and VIIRS Active Fire data products will be collected from January 20, 2012, to December 31, 2024. These data will be normalized and preprocessed, including radiometric calibration, atmospheric correction, and cropping.

[0039] The second step is to automatically extract industrial land. Objects with industrial characteristics, that is, industrial equipment targets (circular fuel piles, oil tanks, chimneys, etc.) can be clearly presented in high-resolution digital orthophotos less than 1 meter, with the characteristics of geometric shape (circular), color more obvious than the surrounding background, stable area range, and equipment cluster distribution. At present, deep learning technology can make comprehensive use of these features to detect these small-area representative industrial equipment targets from a variety of complex backgrounds and numerous noise interferences. Taking into account the accuracy and speed issues, this embodiment applies the deep learning target detection framework Faster Regions with Convolutional Neural Network (Faster R-CNN) to the industrial land detection task of digital orthophotos. The main steps are: training sample production, model training, target detection ( Figure 2 ). Specifically, it includes the following aspects:

[0040] a. The project applies the existing research results of the project team to roughly calibrate the approximate location of industrial land, and then carefully search for relevant industrial equipment in and around these locations. The high-resolution images of the area in 2023 are cropped according to a fixed size (300×300 pixels) to obtain training sample images. In addition, LableImg is used to uniformly mark the target labels in Pascal VOC format to obtain the training sample library ( Figure 3 In addition, to enrich the sample set, data augmentation (translation, rotation, flipping, and distortion) was performed on the collected training sample set. Afterwards, the classic convolutional neural network model, Faster R-CNN, was used to train a model for detecting industrial land in the Texas region.

[0041] b. The digital orthophoto of the evenly divided area is converted into a 300×300 image to be detected. Target detection is then performed on each image based on the trained model. The detected targets are then mapped onto a map with geographic coordinates to obtain preliminary detection results. After obtaining these preliminary detection results through deep learning, image processing techniques such as cluster analysis and spatial filtering are used to remove noise, ultimately achieving accurate spatial location identification of industrial land.

[0042] Step 3: Industrial heat source detection

[0043] Based on the spatial locations of industrial land extracted above, in order to determine whether these industrial lands are industrial heat sources (industrial heat sources are industrial locations that generate combustion heat emissions during industrial development activities, such as Figure 3 -(d2)(d4), Figure 4 -(a1)(b1)(c1)(d1)). This implementation case uses Landsat imagery series covering the region from 2000 to 2023 (Landsat-5TM from 2000–2011, Landsat-7ETM+ from 1999–2013, and Landsat-8OLI from 2013–2020) and Sentinel-2MSI imagery from 2015–2023. By utilizing the thermal sensitivity of the infrared bands of the images, we analyze and summarize the band difference characteristics between the sampled industrial heat sources and the background, construct an industrial heat source detection index, and complete the detection of industrial heat sources. Furthermore, by integrating various image processing techniques and leveraging the GEE (Google Earth Engine) cloud computing platform, we can perform a long-term analysis of the spatiotemporal changes of regional industrial heat sources.

[0044] The combustion of industrial heat sources occurs during industrial development or long-term production activities. The most intuitive manifestation in remote sensing images is that it occurs at the top of a vertical chimney ( Figure 3 -(d2)(d4)), or a torch in a pit on the ground ( Figure 4 -(b1)(d1)), it is obvious that the industrial heat source target has a small area but the characteristics of continuous high-temperature heat emission. The NIR, near-SWIR and far-SWIR of the Landsat series images and Sentinel-2MSI images are sensitive enough to high-temperature combustion heat anomalies. Based on the industrial land targets extracted above, a buffer zone is established to sample industrial heat sources and analyze and summarize their infrared band characteristics.

[0045] The experiment found that compared with other band combinations, the high-temperature emission points of industrial heat sources are most prominent in the false color image (FCI) of the NIR, near-SWIR and far-SWIR band combinations. Figure 4 As shown in the example, the combustion location of industrial heat sources is difficult to identify in True Color Image (TCI) ( Figure 4 First column: (a1)(b1)(c1)(d1)), but in the FCI of the combination of B8a, B11, and B12 of Sentnel-2MSI, a red-yellow color that is distinguishable from the background is shown ( Figure 4 The second column of the image is (a2), (b2), (c2), and (d2). Therefore, the reflectance values ​​of the NIR, near-SWIR, and far-SWIR bands corresponding to all sample object pixels are extracted, and the inter-band reflectance relationship between non-thermal anomaly pixels and thermal anomaly pixels in band 8a (B8a), band 11 (B11), and band 12 (B12) of the MSI image is determined, and the industrial heat source detection index is constructed.

[0046] Therefore, this step includes constructing the high temperature thermal anomaly index shown in the following formula, which is the industrial heat source detection index (TAI) obtained by the project team in advance. The TAI detection results are as follows Figure 4 The third column (a3)(b3)(c3)(d3) is shown.

[0047] TAI=(ρ far-SWIR -ρ near-SWIR ) / ρ NIR

[0048] Among them, ρ far-SWIR Represents the reflectivity of the far- and short-wave infrared bands, ρ near-SWIR Represents the reflectivity in the near-shortwave infrared band, ρ NIR Represents the reflectivity in the near-infrared band.

[0049] For Sentinel-2MSI images, TAI = (B12-B11) / B8a;

[0050] Among them, B12 is the far-shortwave infrared band of Sentinel-2MSI image, B11 is the near-shortwave infrared band of Sentinel-2MSI image, and B8a is the near-infrared band of Sentinel-2MSI image.

[0051] For Landsat-8 OLI images, TAI = (B7 - B6) / B5.

[0052] Among them, B7 is the far-shortwave infrared band of Landsat-8OLI image, B6 is the near-shortwave infrared band of Landsat-8OLI image, and B5 is the near-infrared band of Landsat-8OLI image.

[0053] When constructing the TAI index, cloud noise and oversaturated reflectivity pixels need to be removed. For Landsat optical images, the cloud band is QA, and for Sentinel-2MSI optical images, the cloud band is QA60. The cloud band is used for filtering. The filtering condition for oversaturated reflectivity pixels is: ρ far-SWIR ≥1 and ρ near-SWIR ≥1, that is, satisfying ρ far-SWIR ≥1 and ρ near-SWIR Pixels with a reflectivity ≥ 1 are oversaturated pixels.

[0054] In this step, a buffer zone is created for the industrial land locations extracted in the second step. Heat sources within the buffer zone are considered industrial heat sources. The buffer zone has a radius of 500 meters.

[0055] After proposing a TAI method for detecting thermal anomalies based on Sentinel-2 MSI images, the present invention applied this method to Landsat series images, considering that the wavelength ranges of the near-infrared, near-shortwave infrared, and far-shortwave infrared bands of Landsat-5TM, Landsat-7ETM+, and Landsat-8OLI are similar to those of the MSI images. For four types of industrial heat sources (cement plants, steel plants, coking plants, and oil and gas combustion), TAI was calculated in Landsat-5TM, Landsat-7ETM+, and Landsat-8OLI images, and it was found that high-temperature facilities showed high TAI index characteristics similar to those in MSI images ( Figure 5 ).

[0056] In addition to the high-temperature combustion characteristics, industrial heat sources also have temporal persistence in the medium-resolution multispectral image. Compared with sporadic thermal anomalies such as biomass combustion, industrial heat sources are thermal anomalies that exist at the same location for a long time. Taking advantage of this characteristic, the present invention uses long-term series images to perform threshold segmentation (TAI ≥ 0.45) on the TAI detection results to obtain binary results (0 / 1), and then accumulates the TAI binary results of the long-term series to obtain the TAI cumulative frequency graph ( Figure 4 The fourth column: (a4)(b4)(c4)(d4)) obtains the spatial location of the long-term burning industrial heat source. In this embodiment, the pixels with a frequency greater than or equal to 3 (the frequency threshold is 3) are determined to be long-term burning industrial heat sources. The embodiment of the present invention uses the GEE (Google Earth Engine) platform and a long-term series of multi-source median-resolution remote sensing images to calculate the TAI index, perform threshold segmentation, filter, denoise, and verify to obtain the spatiotemporal distribution of industrial heat sources in the entire region every year from 2000 to 2023.

[0057] Step 4: Classification of industrial heat sources and analysis of temperature characteristics.

[0058] According to the spatial location of the global industrial heat sources extracted above, based on the fire data products from 2012 to 2024 (VIIRS Active Fire and VIIRS Nightfire), the temperature characteristics of different industrial types are analyzed to complete the industrial classification of industrial heat sources (petroleum / coking processing industry, chemical raw material manufacturing industry, metal smelting, electricity / heat industry). In the time series fire data products, industrial heat sources are significantly different from other thermal anomaly phenomena (biomass combustion, volcanic activity, etc.), and have the characteristics of "unchanged position-continuous time-high temperature combustion". The main manifestations are: in the spatial dimension, due to the unchanged combustion position, the emission thermal anomaly points generated by industrial heat sources are detected frequently and are densely distributed, while the thermal anomaly points generated by biomass combustion are relatively dispersed and have a low density; in the temporal dimension, industrial heat sources are usually in continuous operation and can be detected multiple times over a long period of time (such as a year), while naturally occurring or artificially induced biomass combustion has obvious seasonality; in the temperature dimension, the thermal anomaly temperature distribution characteristics of similar industrial heat sources are similar, while the thermal anomaly temperature distribution of biomass combustion is relatively random, as shown in the following example. Figure 6 shown.

[0059] According to the above characteristics, this embodiment uses the spatial position of the regional industrial heat source extracted above ( Figure 7 ), a 500-meter buffer zone is generated based on the center point of each heat source object. Then, using VIIRS Active Fire and VIIRS Nightfire data from 2012 to 2024, the fire point data falling within each industrial heat source buffer zone are counted on an annual basis to obtain the annual temperature data of each industrial heat source object. The temperature histogram of the industrial heat source object is then normalized to a ratio ( Figure 6 ) to describe the temperature characteristics of the industrial heat source object.

[0060] Then, based on the temperature templates of different industrial types (cement plants, steel mills, coal chemical plants, refineries / terminals, and onshore oil and gas combustion) that the team had already sampled, the minimum distance classification method was used to calculate the Euclidean distance between the temperature histogram of the industrial heat source object to be classified and each temperature template. The category with the smallest distance was used as the output category of the object, completing the classification of the entire industrial heat source.

[0061] Step 5: Modeling and inversion of industrial production activities

[0062] After obtaining the classification of industrial heat sources, the annual temperature data of various industrial heat sources are fitted with the industrial output of their corresponding categories to construct a polynomial regression model. Then, the annual output of each industrial object is inverted based on this model, thereby reflecting the production activities and changing trends of regional industrial heat sources.

[0063] Statistical yearbooks for districts and counties from 2012 to 2024 were compiled and summarized, and the annual "output of major industrial products above designated size" for each city was summarized. The total output value of energy-intensive heavy industries (petroleum / coking processing, chemical raw materials manufacturing, metal smelting, and electricity / heating) was extracted. Furthermore, the temperature data for different types of industrial heat sources obtained from the VIIRS fire product data analysis were combined, with the city-level administrative district as the statistical unit, to sum the annual temperatures of different types of industrial heat sources for each city.

[0064] At this point, we have obtained the annual total output value of various high-energy-consuming industries in each city, as well as the annual temperature data of the corresponding industrial heat sources. We fit the two together to construct a quadratic polynomial regression model. Then, based on this model and the annual temperature values ​​of each industrial heat source object, we invert the annual output of each industrial heat source object, thereby obtaining the production activities of industrial heat sources at the provincial scale from 2012 to 2024, so as to understand the production activity of industrial heat sources at the provincial scale. The specific process is as follows: Figure 8 shown.

[0065] Step 6: Spatiotemporal Statistics and Database Construction

[0066] 6.1. Integrate the spatial location, classification, temperature, and output attributes of industrial heat sources to construct an industrial heat source database covering the entire region; extract the regional industrial heat source database with location, quantity, scale, operating time, annual heat exhaust temperature, and annual production attributes.

[0067] 6.2. The temporal and spatial distribution of industrial heat sources in the database is statistically analyzed by year and city and county, and the historical and current status of the industry is analyzed.

[0068] In addition to the above embodiments, the present invention may also have other implementations. Any technical solution formed by equivalent replacement or equivalent transformation falls within the scope of protection required by the present invention.

Claims

1. A method for monitoring industrial heat source development activities based on multi-source time series remote sensing images, comprising the following steps: The first step is to acquire and process multi-source remote sensing imagery data for the study area. The multi-source remote sensing optical imagery includes Landsat imagery, Sentinel-2MSI optical imagery, and high-spatial-resolution digital orthophotos. VIIRS Nightfire and VIIRS Active Fire data products are collected over a long time series covering the area. The second step is to use the Faster R-CNN deep learning model to extract industrial equipment targets from the high-spatial-resolution digital orthophotos. The main steps are: training sample preparation, model training, and target detection. After the deep learning model obtains preliminary detection results, noise is removed to obtain the final accurate spatial location identification results of the industrial land. The third step, industrial heat source detection, includes the following steps: 3.1、Constructing high temperature thermal anomaly index TAI=(ρ far-SWIR -ρ near-SWIR ) / ρ NIR , where ρ far-SWIR Represents the reflectivity of the far- and short-wave infrared bands, ρ near-SWIR Represents the reflectivity in the near-shortwave infrared band, ρ NIR Represents the reflectance of the near-infrared band; for Sentinel-2MSI images, TAI = (B12-B11) / B8a, where B12 is the far- and short-wave infrared bands of Sentinel-2MSI images, B11 is the near- and short-wave infrared bands of Sentinel-2MSI images, and B8a is the near-infrared band of Sentinel-2MSI images; for Landsat-8OLI images, TAI = (B7-B6) / B5, where B7 is the far- and short-wave infrared bands of Landsat-8OLI images, B6 is the near- and short-wave infrared bands of Landsat-8OLI images, and B5 is the near-infrared band of Landsat-8OLI images; 3.

2. For Sentinel-2MSI and Landsat-8OLI images, threshold segmentation is performed using the TAI value of each pixel. Combined with the location and temporal persistence characteristics of industrial land extracted in the second step, long-term burning industrial heat sources are extracted. Step 4: Industrial heat source classification and temperature characteristics analysis 4.

1. Generate a buffer zone based on the center point of each industrial heat source object. Then, using VIIRS Active Fire and VIIRS Nightfire data, count the fire point data falling within each industrial heat source buffer zone on an annual basis to obtain the annual temperature data of each industrial heat source object. Build a temperature histogram for each industrial heat source object and normalize it to a ratio. 4.

2. Using the minimum distance classification method, match the industrial heat source temperature histogram with the temperature template of the known industrial type. The category with the smallest distance is used as the output category of the object to complete the classification of the industrial heat source. Step 5: Modeling and inversion of industrial production activities 5.

1. Fit the annual temperature data of various industrial heat sources to their corresponding industrial output to construct a regression model. Then, based on this model, invert the annual output of each industrial object to reflect the production activities and changing trends of regional industrial heat sources. 5.

2. Collect the total output value data of high-energy-consuming industries from the statistical yearbooks of the covered area for many years, and perform polynomial regression fitting on the annual temperature data of classified industrial heat sources and the output value of the corresponding industrial categories in the municipal administrative districts; 5.

3. Based on the regression model, the annual output of each industrial heat source is inverted and the trend of production activities is analyzed.

2. The method for monitoring industrial heat source development activities based on multi-source time series remote sensing images according to claim 1, characterized in that: In step 3.2, a buffer zone is established based on the industrial land locations extracted in the second step, and the heat sources within the buffer zone are industrial heat sources.

3. The method for monitoring industrial heat source development activities based on multi-source time series remote sensing images according to claim 1, characterized in that: In step 3.2, after threshold segmentation is performed on the TAI detection results to obtain binary results, the TAI binary results of the long time series are accumulated to obtain the TAI cumulative frequency map. The spatial location of the long-term burning industrial heat source is obtained based on the frequency threshold. The condition for threshold segmentation is TAI ≥ 0.45, that is, the TAI index is greater than or equal to 0.45 and is an industrial heat source.

4. The method for monitoring industrial heat source development activities based on multi-source time series remote sensing images according to claim 1, characterized in that: In the fourth step, when classifying the VIIRS temperature data, the buffer radius was 500 m, and the Euclidean distance was used for the minimum distance classification.

5. The method for monitoring industrial heat source development activities based on multi-source time series remote sensing images according to claim 1, characterized in that: In step 5.1, the industrial production inversion model uses quadratic polynomial regression to fit the relationship between temperature and output value.

6. The method for monitoring industrial heat source development activities based on multi-source time series remote sensing images according to claim 1, characterized in that: In the second step, for industrial land extraction, Faster R-CNN training samples are expanded through data augmentation.

7. The method for monitoring industrial heat source development activities based on multi-source time series remote sensing images according to claim 1, characterized in that: In step 3.1, the TAI index is constructed by first removing cloud noise and then removing pixels with oversaturated reflectivity. The filtering condition for pixels with oversaturated reflectivity is: ρ far-SWIR ≥1 and ρ near-SWIR ≥1.

8. The method for monitoring industrial heat source development activities based on multi-source time series remote sensing images according to claim 1, characterized in that: It also includes the sixth step, spatiotemporal statistics and database construction, which specifically includes: 6.

1. Integrate the spatial location, classification, temperature, and output attributes of industrial heat sources to build an industrial heat source database covering the entire region; 6.

2. Calculate the temporal and spatial distribution of industrial heat sources by year and city or county, and assess the history and current status of industrial development.

Images