GEE-based ecological space management and control area development identification early warning method and system
By using the GEE platform's random forest model in the ecological space control area for remote sensing image processing and classification, the problems of low identification activities in the existing technology are solved, and high-precision land use and land cover type change detection is achieved. Support near-real-time early warning and long-term change analysis.
Patent Information
- Application Number
- CN202411078198.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-07
- Publication Date
- 2024-12-27
AI Technical Summary
The prior art has misjudgment and misjudgment in the identification of development and construction activities in ecological space control areas, and lacks understanding of the spectral response and timing characteristics of the remote sensing image reacts to the geometry state, resulting in low monitoring efficiency and poor accuracy.
The GEE-based ecological space control area development early warning method is adopted to obtain remote sensing image data for image preprocessing, and use a random forest model for training and classification, combining different spectral characteristics and attributes to realize the change detection of land use and land cover types. , supports autonomous upload of high-precision data, and uses machine learning and cross-platform communication to achieve near-real-time early warning.
The accuracy and efficiency of the identification activities are improved, the average overall accuracy reaches 79%, and the Kappa coefficient reaches 0.67, which assists in manual interpretation, reduces the workload, and visualizes the activity strength and weakness through heat maps, supporting long-term change detection and early warning.
Smart Images
Figure CN119202919A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of ecological space control area identification and early warning, and specifically to a GEE-based ecological space control area development identification and early warning method and a GEE-based ecological space control area development identification and early warning system. Background Art
[0002] As for the identification of development and construction activities in ecological space control areas, the commonly used means in the industry is to manually identify development and construction activities by visually interpreting high-resolution satellite images of the target area before and after. Or the machine vision method uses computer semantic segmentation and change detection technology to achieve comprehensive monitoring and control of illegal and irregular development and construction activities in ecological space protection areas.
[0003] The existing methods have certain shortcomings:
[0004] Manual verification methods rely on professional interpreters, which are labor-intensive, time-consuming, and prone to missed judgments. Machine vision methods mainly use computer technology to automatically identify and judge remote sensing images. There is a lack of understanding of physical factors such as the spectral response of remote sensing images to reflect the state of objects, and there is little consideration of factors affecting the temporal characteristics of objects. There is a certain gap between technology and reality. Summary of the invention
[0005] Based on this, it is necessary to provide an identification and early warning method and system for the development of ecological space control areas based on GEE to address the problems of misjudgment, missed judgment or large errors in existing identification and early warning methods.
[0006] To achieve the above object, the present invention adopts the following technical solutions:
[0007] The identification and early warning method for ecological space control areas based on GEE includes the following steps:
[0008] S1. Obtain remote sensing image data of the study area and annotate them by category, perform image preprocessing on the annotated remote sensing images, input the preprocessed images into a pre-built random forest model for training, and obtain a trained random forest model;
[0009] S2. Collect remote sensing image data of the target area according to a pre-set collection period, input it into the trained random forest model, obtain the probability distribution of each category of the target area within the collection period and store it;
[0010] S3. Compare the probability distribution of each category with the probability distribution of each category in the target area in the previous collection cycle, determine whether the land use and land cover type of the target area has changed based on the difference value, and make the following decisions:
[0011] (1) If the difference value is within the preset error range, it is determined that the land use and land cover type of the target area has not changed and is marked as unchanged;
[0012] (2) If the difference value is not within the preset error range, the land use and land cover type changes in the target area are determined and marked;
[0013] S4. Calculate and visualize the category, number of pixels, and area of the target area with reference to the marking information. Use a heat map to visualize the category change matrix of the target area in different acquisition cycles, and use different colors to visualize the number of pixels corresponding to the target area images in different acquisition cycles to obtain pixel-level change detection results for early warning.
[0014] Furthermore, in step S1, the remote sensing image is preprocessed, including using different cloud masking techniques to process image clouds and shadows and normalizing the reflectivity of the image.
[0015] Furthermore, in step S1, the remote sensing image data of the study area includes at least 1000 samples obtained from open source data.
[0016] Furthermore, in step S2, the probability distribution data of each category of the target area is stored in a cloud database in a time series, and the updated probability distribution data is transmitted to the local database at regular intervals.
[0017] The present invention also relates to an ecological space control area development identification and early warning system based on GEE, which mainly includes a model training module, a classification calculation module, a change detection module and a visualization module.
[0018] The model training module is used to obtain remote sensing image data of the study area and mark the categories, perform image preprocessing on the remote sensing images with category marking, input the preprocessed images into the pre-built classification model for training, and obtain a trained classification model;
[0019] The classification calculation module is used to collect remote sensing image data of the target area according to a preset collection period, input it into the trained classification model, obtain the probability distribution of each category of the target area within the collection period and store it;
[0020] The change detection module is used to compare the probability distribution of each category with the probability distribution of each category in the target area in the previous acquisition cycle, determine whether the land use and land cover type of the target area has changed based on the difference value, and make corresponding marks based on the judgment result;
[0021] The visualization module is used to calculate and visualize the category, number of pixels, and area of the target area with reference to the marking information. A heat map is used to visualize the category change matrix of the target area in different acquisition cycles, and different colors are used to visualize the number of pixels corresponding to the target area images in different acquisition cycles to obtain pixel-level change detection results for early warning.
[0022] Furthermore, the GEE-based ecological space control zone development identification and early warning system also includes a data upload module for users to upload target area data by themselves. The formats uploaded by users include shp, zip, dbf, prj, shx, cpg, fix, qix, sbn, shp or xml.
[0023] Furthermore, the user obtains local disk downloading and sharing permissions based on the client ID and key.
[0024] Furthermore, the pixel-level change detection result downloaded by the user is a tif image or a shp file.
[0025] Furthermore, the period of the collection cycle ranges from 3 to 7 days, depending on the latitude of the target area.
[0026] Furthermore, the classification model includes a random forest model, a decision tree model or a neural network model.
[0027] Compared with the prior art, the beneficial effects of the present invention include:
[0028] 1. The present invention calculates and analyzes the remote sensing images of the target area in near real time by classifying land use and land cover, combining different spectral characteristics and attributes, with an average overall accuracy of 79% and a Kappa coefficient of 0.67. In addition, it also supports autonomous uploading of higher-precision data. Based on the existing platform and data, the present invention fully explores the relationship between technology and physical reality, and uses machine learning, cross-platform communication, near-real-time early warning and other methods to realize ecological environment monitoring, assist manual interpretation, reduce workload and improve work efficiency.
[0029] 2. The present invention uses the long-term changes in land use and land cover types to build in early warning judgment criteria based on changes in land feature types in different periods. It can distinguish development and construction activities based on change detection results, and use different color heat maps to distinguish the intensity of the activities, which are then displayed on the early warning interface. BRIEF DESCRIPTION OF THE DRAWINGS
[0030] The disclosure of the present invention is described with reference to the accompanying drawings. It should be understood that the drawings are only for illustrative purposes and are not intended to limit the scope of protection of the present invention. Among them:
[0031] Figure 1This is a flow chart of the GEE-based ecological space control area development identification and early warning method introduced in the present invention. DETAILED DESCRIPTION
[0032] It is easy to understand that according to the technical solution of the present invention, without changing the essential spirit of the present invention, a person skilled in the art can propose a variety of interchangeable structural modes and implementation modes. Therefore, the following specific implementation modes and drawings are only exemplary descriptions of the technical solution of the present invention, and should not be regarded as the whole of the present invention or as a limitation or restriction to the technical solution of the present invention.
[0033] Example 1
[0034] See also Figure 1 This embodiment introduces a GEE-based ecological space control area development identification and early warning method, including the following steps:
[0035] S1. Obtain remote sensing image data of the study area and label them by category, perform image preprocessing on the labeled remote sensing images, input the preprocessed images into a pre-built random forest model for training, and obtain a trained random forest model.
[0036] In step S1, the random forest model is based on the GEE (Google Earth Engine) platform. The dataset for training the random forest model includes nearly 1,000 samples of data with manual visual interpretation and identification annotation; the platform is based on the open source LULC datasets (WorldCover 10m v100 / v200 and DynamicWorldV1) based on Sentinel-1 and Sentinel-2.
[0037] Specifically, the Sentinel-2 remote sensing images in the study area can be retained, and the images can be preprocessed, including using different cloud masking techniques to process clouds and shadows, and normalizing and adjusting the reflectance values of the images to improve the balance and processing effect of the data. The blue light, green light, red light, near infrared light, shortwave infrared 1 and shortwave infrared 2 bands are obtained, and normalized indices such as NDVI (normalized vegetation index), NDSI (normalized snow cover index), NDBI (normalized building index) and NDWI (normalized difference water index) are added, and the NBR (normalized burn index), BSI (bare ground index), Wetness (humidity), EVI (enhanced vegetation index) and Brightness (brightness) index are calculated.
[0038] S2. Collect remote sensing image data of the target area according to a pre-set collection period, input it into the trained random forest model, obtain the probability distribution of each category of the target area within the collection period and store it.
[0039] In the actual training process, we can try to introduce other weakly supervised learning methods to improve the performance and generalization ability of the model. Use the GEE platform to process remote sensing images in the target area and classify land use and land cover (LULC) in near real time. Near real time is preferably 5 days / time.
[0040] The probability distribution data of each category in the target area is stored in the cloud database in time series, and the updated probability distribution data is transmitted to the local database regularly.
[0041] S3. Compare the probability distribution of each category with the probability distribution of each category in the target area in the previous collection cycle, determine whether the land use and land cover type of the target area has changed based on the difference value, and make the following decisions:
[0042] (1) If the difference value is within the preset error range, it is determined that the land use and land cover type of the target area has not changed and is marked as unchanged;
[0043] (2) If the difference value is not within the preset error range, the land use and land cover type changes in the target area are determined and marked;
[0044] S4. Calculate and visualize the category, number of pixels, and area of the target area with reference to the marking information. Use a heat map to visualize the category change matrix of the target area in different acquisition cycles, and use different colors to visualize the number of pixels corresponding to the target area images in different acquisition cycles to obtain pixel-level change detection results for early warning.
[0045] This embodiment calculates and analyzes the remote sensing images of the target area in near real time through the classification of land use and land cover, combined with different spectral characteristics and attributes, with an average overall accuracy of 79% and a Kappa coefficient of 0.67. In addition, it also supports autonomous uploading of higher-precision data. Based on the existing platform and data, it fully explores the relationship between technology and physical reality, and uses machine learning, cross-platform communication, near-real-time early warning and other methods to realize ecological environment monitoring, assist manual interpretation, reduce workload, and improve work efficiency.
[0046] Example 2
[0047] This embodiment introduces the development of an identification and early warning system for ecological space control areas based on GEE, which mainly includes a model training module, a classification calculation module, a change detection module, a visualization module, and a data upload module. All of these modules are based on the GEE platform.
[0048] The model training module is used to obtain remote sensing image data of the study area and mark the categories, perform image preprocessing on the remote sensing images with category marking, input the preprocessed images into the pre-built classification model for training, and obtain a trained classification model;
[0049] The data upload module is used for users to upload target area data by themselves. The formats uploaded by users include shp, zip, dbf, prj, shx, cpg, fix, qix, sbn, shp or xml.
[0050] The classification calculation module is used to collect remote sensing image data of the target area according to a preset collection period, input it into the trained classification model, obtain the probability distribution of each category of the target area within the collection period and store it;
[0051] The change detection module is used to compare the probability distribution of each category with the probability distribution of each category in the target area in the previous acquisition cycle, determine whether the land use and land cover type of the target area has changed based on the difference value, and make corresponding marks based on the judgment result;
[0052] The visualization module is used to calculate and visualize the category, number of pixels, and area of the target area with reference to the marking information. A heat map is used to visualize the category change matrix of the target area in different acquisition cycles, and different colors are used to visualize the number of pixels corresponding to the target area images in different acquisition cycles to obtain pixel-level change detection results for early warning.
[0053] In practical application, this system can realize cross-platform data analysis and transmission, namely:
[0054] (1) Script-driven near-real-time computing: The local script periodically drives the GEE code to calculate and output the latest classification results to Google Drive.
[0055] (2) Scheduled data transmission: The local script periodically fetches the latest calculation results from Google Drive to the local disk, taking timing control issues into consideration.
[0056] It can also realize the change detection and early warning of long-term LULC classification results, that is, the judgment logic of change detection of LULC classification results is built-in. Qualitative and quantitative judgment rules are preset for the changes before each land use and land cover type. Early warning of changes in long-term LULC classification results. Combining short-term changes with long-term changes, determine the real traces of human activities in the control area.
[0057] Judgment logic For example, if there are land use types of "water bodies", "crops", "trees" and "shrubs" converted to "buildings" in the control area, there is a high probability that there are obvious human activities, and it will be marked as "strong traces of human activities" (generally represented by red in the heat map). If there are "woodland" types converted to "shrubs" and "grassland" in the control area, there may be human logging activities, and it will be marked as "relatively strong traces of human activities" (generally represented by orange in the heat map).
[0058] The local data visualization of this system is as follows:
[0059] a. Calculate and visualize the category, number of pixels, and area of each scene.
[0060] b. Use heat maps to visualize the category change matrix of the two-scene classification images in the previous and next time series. For example, the vertical axis of the image is the classification result type of the previous time series, and the horizontal axis is the classification result type of the subsequent time series. The numbers in the heat map are the changed pixel values. Red represents an increase in the value, and blue represents a decrease in the value.
[0061] c. Use different colors to visualize the specific pixels of the two classified images in the previous and next time series, with cyan representing the unchanged parts and colors representing the changed pixels. Based on the different colors, you can know which type has changed to which type.
[0062] For the output of visualization results, the changing results can be output according to the LULC classification results in different periods. The results are tif graphs and can also be converted into shp files.
[0063] The specific implementation cases are as follows:
[0064] (1) Platform calculation script
[0065] The user logs in to Google Earth Engine and runs the following script in the Code Editor interface to obtain the land use and classification results of the target area at the current time (within 3-5 days; Sentinel-2 generally images high-latitude areas once every 3 days and low-latitude areas once every 5 days).
[0066] The "target area" is the redline area shp file that has been uploaded to Assets. If you need to change a different redline area, you only need to upload a new area shp file (including shp, zip, dbf, prj, shx, cpg, fix, qix, sbnor shp.xml.), import the shp area as a new variable table in the calculation script, and rerun it.
[0067] (2) Automatically run GEE platform calculation scripts locally
[0068] Running drive_gee.py locally can drive the GEE script and output the latest classification results to the ESA folder in Google Drive.
[0069] Need to be in bash ahead of time:
[0070] Add execution permission, chomd+x drive_gee.py
[0071] Set up a scheduled task:
[0072] crontab -e
[0073] Scheduled to run at 14:00 every Friday afternoon.
[0074] 014**5 / usr / bin / python3 / path / drive_gee.py
[0075] The above times can be modified.
[0076] (3) Storage of classification results
[0077] Run download_from_drive.py locally to download the latest classification results to a specified local folder. The downloaded pixel-level change detection results are tif images, which can also be converted into shp files.
[0078] You need to open the Google Drive API in advance and export the credentials.json file containing the client ID and key, or you can directly open the ESA file sharing permissions to synchronize data directly.
[0079] Need to be in bash ahead of time:
[0080] Add execution permission, chomd+x download_from_drive.py
[0081] Set up a scheduled task:
[0082] crontab -e
[0083] Scheduled to run every Saturday at 8 am,
[0084] 08**6 / usr / bin / python3 / path / download_from_drive.py
[0085] The above times can be modified.
[0086] A classification map of the target area will be automatically calculated every week and downloaded to a folder in the specified local path (.. / data).
[0087] In addition, local data analysis and interaction programs were developed based on PySimpleGUI.
[0088] On the corresponding login interface, you can enter your username and password to log in. You can register for the first login. After entering the data interface, you can upload and display your own data. Through the visualization interface, you can view the visualization results of different dimensions of the classification results.
[0089] For the GEE platform calculation part, the model scheme of random forest is used to classify the land use and land cover types in the target area. However, other models can also be used as alternatives to the invention to realize the functions of the system, such as the support vector machine (SVM) model, which is a supervised learning method, which solves the optimization problem in the high-dimensional feature space by introducing the concept of kernel function, and then finds the optimal classification hyperplane to solve the complex data classification problem. There is also a decision tree (Decision Tree) model, which is a classification method of inductive reasoning, by defining rules for information such as image spectrum, color, space, etc., starting from the central node, comparing various information values of the image, deriving new branches, and obtaining new decision trees by updating rules, until the classification requirements are met, and the final node is the classification result. Even some model methods of deep learning, convolutional neural network (CNN), recurrent neural network (RNN), generative adversarial network (GAN), etc., can also realize the classification problem of LULC.
[0090] The local data analysis and interaction program is developed based on PySimpleGUI, mainly to realize some customized data analysis services. Similarly, this interaction and visualization process can also be developed based on the VUE front-end framework and Openlayers library as an alternative solution.
[0091] Therefore, this system automates the classification of LULC. The GEE script includes codes for data preprocessing, model training, classification calculation, and accuracy evaluation. High resolution (10 meters), based on Sentinel-2 satellite remote sensing image data, and can support autonomous uploading of higher-precision data. Near real-time (5 days / scene), calculation and analysis are performed for each scene of remote sensing image. Accuracy evaluation, the average overall accuracy reached 79%, and the Kappa coefficient reached 0.67. In the calculation, different spectral features and attributes play a decisive role in the classification accuracy; cross-platform data automatic analysis and transmission and local data visualization are also realized.
[0092] The system takes into account the long-term changes in land use and land cover types, and has built-in early warning judgment criteria based on the changes in land feature types in different periods. It can distinguish development and construction activities based on the change detection results, and use different color heat maps to distinguish the intensity of the activities, which are then displayed on the early warning interface to assist humans in improving work efficiency.
[0093] The technical scope of the present invention is not limited to the contents in the above description. Those skilled in the art can make various deformations and modifications to the above embodiments without departing from the technical idea of the present invention, and these deformations and modifications should all fall within the protection scope of the present invention.
Claims
1. The GEE-based ecological space control area development identification and early warning method is characterized by: It includes the following steps: S1. Obtain remote sensing image data of the study area and annotate them by category, perform image preprocessing on the annotated remote sensing images, input the preprocessed images into a pre-built random forest model for training, and obtain a trained random forest model; S2. Collect remote sensing image data of the target area according to a pre-set collection period, input it into the trained random forest model, obtain the probability distribution of each category of the target area within the collection period and store it; S3. Compare the probability distribution of each category with the probability distribution of each category in the target area in the previous collection cycle, determine whether the land use and land cover type of the target area has changed based on the difference value, and make the following decisions: (1) If the difference value is within the preset error range, it is determined that the land use and land cover type of the target area has not changed and is marked as unchanged; (2) If the difference value is not within the preset error range, the land use and land cover type changes in the target area are determined and marked; S4. Calculate and visualize the category, number of pixels, and area of the target area with reference to the marking information. Use a heat map to visualize the category change matrix of the target area in different acquisition cycles, and use different colors to visualize the number of pixels corresponding to the target area images in different acquisition cycles to obtain pixel-level change detection results for early warning.
2. The method for identifying and warning the development of ecological space control areas based on GEE according to claim 1 is characterized in that: In step S1, image preprocessing of the remote sensing image includes using different cloud masking techniques to process image clouds and shadows and normalizing the image reflectivity.
3. The method for identifying and warning the development of ecological space control areas based on GEE according to claim 1 is characterized in that: In step S1, the remote sensing image data of the study area includes at least 1000 samples obtained from open source data.
4. The method for identifying and warning the development of ecological space control areas based on GEE according to claim 1 is characterized in that: In step S2, the probability distribution data of each category of the target area is stored in the cloud database in time series, and the updated probability distribution data is transmitted to the local database regularly.
5. Based on GEE, the ecological space control area development identification and early warning system is characterized by: It includes: The model training module is used to obtain remote sensing image data of the study area and mark the categories, perform image preprocessing on the remote sensing images with category marking, and input the preprocessed images into a pre-built classification model for training to obtain a trained classification model; A classification calculation module is used to collect remote sensing image data of the target area according to a preset collection period, input it into a trained classification model, obtain the probability distribution of each category of the target area within the collection period and store it; The change detection module is used to compare the probability distribution of each category with the probability distribution of each category in the target area in the previous acquisition cycle, determine whether the land use and land cover type of the target area has changed based on the difference value, and make corresponding marks based on the judgment result; The visualization module is used to calculate and visualize the category, number of pixels, and area of the target area with reference to the marking information. A heat map is used to visualize the category change matrix of the target area in different acquisition cycles, and different colors are used to visualize the number of pixels corresponding to the target area images in different acquisition cycles, to obtain pixel-level change detection results for early warning.
6. The GEE-based ecological space control area development identification and early warning system according to claim 5 is characterized in that: The GEE-based ecological space control area development identification and early warning system also includes a data upload module for users to upload target area data by themselves. The formats uploaded by users include shp, zip, dbf, prj, shx, cpg, fix, qix, sbn, shp or xml.
7. The GEE-based ecological space control area development identification and early warning system according to claim 5 is characterized in that: The user obtains the local disk download and sharing permissions based on the client ID and key.
8. The GEE-based ecological space control area development identification and early warning system according to claim 7 is characterized in that: The pixel-level change detection results downloaded by the user are in the form of tif images or shp files.
9. The GEE-based ecological space control area development identification and early warning system according to claim 5 is characterized in that: The period of the collection cycle ranges from 3 to 7 days, depending on the latitude of the target area.
10. The GEE-based ecological space control area development identification and early warning system according to claim 5, characterized in that: Classification models include random forest models, decision tree models, or neural network models.