A high-precision extraction method for thermal thaw lakes and freeze-thaw collapse based on satellite data
Through a high-precision extraction method based on satellite data, combined with manual interpretation and deep learning models, the problem of difficult extraction of hot melt lake ponds and frozen thaw collapses in the existing technology is solved, and efficient and accurate remote sensing extraction effect is achieved.
Patent Information
- Application Number
- CN202311486159.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-11-09
- Publication Date
- 2025-01-14
- Estimated Expiration
- 2043-11-09
AI Technical Summary
The prior art is difficult to extract heat-melting lake ponds and frozen-thaw collapses with small-area development characteristics with high accuracy. Traditional methods are prone to leakage of small-area land objects, especially when the whole scene image is directly classified, the extraction efficiency and accuracy are low.
Using a high-precision extraction method based on satellite data, through manual interpretation and labeling, combined with deep learning model training and transfer learning, the remote sensing images in the target area are predicted and identified by thermally melted lakes and ponds and frozen thaw collapses, including obtaining high-resolution satellites. Image preprocessing, image-by-image visual interpretation, buffer analysis, threshold segmentation and deep learning model training to achieve high-precision extraction.
The remote sensing extraction efficiency, integrity and accuracy of hot-melting lakes and ponds and freeze-melting collapses is improved, and it is especially suitable for high-precision extraction of small-area land objects, ensuring the accuracy and integrity of land objects objects, while improving work efficiency.
Smart Images
Figure CN117671511B_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of target positioning, in particular to a high-precision extraction method of thermal thaw lakes and ponds and freeze-thaw collapses based on satellite data. Background Art
[0002] Natural or human factors cause the thickening of the active layer of permafrost and local melting of underground ice or ice-rich permafrost. The ground sinks and forms contiguous, scattered, small-area melt depressions and collapse pits. The parts without water accumulation are called freeze-thaw collapses, and the parts with water accumulation are called melt lakes and ponds. Melt lakes and ponds and freeze-thaw collapses are characterized by small-area development, and melt lakes and ponds and freeze-thaw collapses with an area of less than 10,000 square meters account for about 80% of the total.
[0003] In the existing technology, the manual visual interpretation is usually used to extract large-area thermal thaw lakes and freeze-thaw collapses, which is a huge workload. However, in the extraction of ponds, lakes, collapse pits or bare land, small-area ground objects are easily missed by traditional supervised classification, unsupervised classification and other machine learning methods, band ratio method or threshold segmentation method, especially when the whole scene image is directly classified and extracted, the omission of small-area ground objects is more obvious, so it is difficult to completely extract thermal thaw lakes and freeze-thaw collapses with small-area development characteristics.
[0004] Therefore, based on the above reasons, it is necessary to design a high-precision extraction method for thaw lakes and freeze-thaw collapses based on satellite data. Through manual interpretation and labeling, deep learning model training and transfer learning, the thaw lakes and freeze-thaw collapses in the remote sensing images of the target area are predicted and identified, which improves the efficiency, completeness and accuracy of remote sensing extraction of thaw lakes and freeze-thaw collapses. Summary of the invention
[0005] The purpose of the present invention is to overcome the shortcomings of the prior art and provide a high-precision extraction method for thaw lakes and freeze-thaw collapses based on satellite data. By manually interpreting and marking, deep learning model training and transfer learning, thaw lakes and freeze-thaw collapses in remote sensing images of target areas are predicted and identified, thereby improving the efficiency, integrity and accuracy of remote sensing extraction of thaw lakes and freeze-thaw collapses.
[0006] In order to achieve the above object, the present invention provides a method for extracting thermal thaw lakes and ponds and freeze-thaw collapse with high precision based on satellite data, comprising the following steps:
[0007] The following steps are involved:
[0008] S1, obtain satellite remote sensing images with a resolution better than 2 meters in the working area and perform radiation correction, atmospheric correction, orthorectification, and image fusion preprocessing;
[0009] S2, at a large scale, visually interpret and mark the center points of thermal thaw ponds and freeze-thaw collapses image by image, and obtain the point targets of thermal thaw ponds and freeze-thaw collapses in the work area;
[0010] S3, perform buffer analysis on the point target according to a radius of 300 meters, use the buffer zone to crop the satellite remote sensing image, and obtain the satellite remote sensing image of the target area;
[0011] S4, select typical areas of thermal thaw ponds and freeze-thaw collapse in the work area, perform threshold segmentation on the near-infrared band of the satellite remote sensing image of the target area in the typical development area, preliminarily obtain the vector boundaries of thermal thaw ponds and freeze-thaw collapse in the typical development area, and manually revise the precise vector boundaries;
[0012] S5, using precise vector boundaries and satellite remote sensing images of target areas to create training samples of thermal thaw lakes and freeze-thaw collapses with a pixel size of 256×256, respectively, for deep learning model training;
[0013] S6, apply the training model transfer learning to the satellite remote sensing images of the target area in the entire work area to predict thermal thaw lakes and freeze-thaw collapse.
[0014] The specific preprocessing in S1 is:
[0015] Obtain multispectral satellite remote sensing images covering the work area in the summer phase from June to August with no or little cloud, and the remote sensing image resolution should be better than 2 meters.
[0016] S2 is specifically:
[0017] The thermal thaw ponds and freeze-thaw collapses are visually interpreted from remote sensing images at a large scale of 1:1500 to 1:2000. The thermal thaw ponds and freeze-thaw collapses are visually interpreted grid by grid in the remote sensing images to obtain point target vectors with the geographic coordinates of the thermal thaw ponds and freeze-thaw collapses.
[0018] S3 is specifically:
[0019] The buffer zone surface layer vector is obtained after the target center point of the thermal thaw lake and freeze-thaw collapse is buffered; the target area image obtained by clipping the remote sensing image of the buffer zone includes the image pixels of the thermal thaw lake and freeze-thaw collapse and the surrounding areas.
[0020] S4 is specifically:
[0021] Typical areas of thermal thawing lakes and freeze-thaw collapses are selected to prepare deep learning training samples. Taking into account the phenomenon of different spectra of the same object in remote sensing images, multiple typical development areas can be selected. The precise vector boundaries of thermal thawing lakes and freeze-thaw collapses in typical development areas are obtained through threshold segmentation classification and manual visual revision.
[0022] S5 is as follows:
[0023] Taking into account the small-area development characteristics of thaw lakes and freeze-thaw collapses and the efficiency of deep model training, the training sample size is set to 256×256 pixels; deep learning model training can simultaneously classify thaw lakes and freeze-thaw collapses, or perform binary classification training separately.
[0024] The computer software program, when executed by the processor, implements a high-precision extraction method for thermal thaw lakes and freeze-thaw collapses based on satellite data.
[0025] An electronic device comprises a housing and a control panel, wherein the control panel comprises a storage medium.
[0026] The beneficial technical effects of the present invention are:
[0027] The advantages of accurate target area positioning through manual visual interpretation and the good semantic segmentation and classification capabilities of deep learning are used to achieve complete and high-precision extraction of thermal melt lakes and freeze-thaw collapses with small-area development characteristics.
[0028] Through manual visual interpretation and point marking, the locations of thaw lakes and freeze-thaw collapses with small-area development characteristics can be located relatively quickly; then, through buffer analysis and remote sensing image cropping, the target area remote sensing images of thaw lakes and freeze-thaw collapses can be obtained; then, through threshold segmentation and classification and manual visual revision process, high-precision thaw lake and freeze-thaw collapse sample data sets can be obtained more efficiently, and the sample data sets are used to train the deep learning model; finally, the deep learning model is used to predict the remote sensing images of the target area to achieve complete and high-precision extraction of thaw lakes and freeze-thaw collapses.
[0029] This method is particularly suitable for large-area high-precision and complete extraction of ground objects with small area and extensive development characteristics. It can improve work efficiency while ensuring the accuracy and completeness of ground object extraction. BRIEF DESCRIPTION OF THE DRAWINGS
[0030] Figure 1 It is a schematic diagram of the overall processing flow of the present invention.
[0031] Figure 2 It is a schematic diagram of the hardware structure including the method of the present invention. DETAILED DESCRIPTION
[0032] See also Figures 1-2 The present invention provides a high-precision extraction method for thermal thaw lakes and ponds and freeze-thaw collapses based on satellite data, which is executed by an electronic device containing a computer storage medium of the method, and specifically includes the following steps:
[0033] like Figure 1 As shown:
[0034] S110, obtain satellite remote sensing images with a high resolution better than 2 meters for the working area and perform preprocessing such as radiation correction, atmospheric correction, orthorectification, and image fusion.
[0035] The working area here refers to the watershed scale, the secondary or primary geographical division scale, and usually corresponds to multi-scene satellite remote sensing images, each of which covers a part of the working area. Radiation correction, atmospheric correction, orthorectification, and image fusion are data processing for each multispectral remote sensing image. This embodiment obtains satellite remote sensing images of the working area with a high resolution better than 2 meters as a data source for subsequent high-precision and complete extraction of thermal thaw lakes and freeze-thaw collapses. The Level1A data product of the GF-1 remote sensing satellite is used as the data source.
[0036] Obtain a certain period of cloudless or lightly clouded multispectral satellite imagery with a resolution better than 2 meters in the summer phase that completely covers the working area, generally including at least four bands: near infrared band, red band, green band and blue band.
[0037] S120. At a large scale, visually interpret and mark the center points of thermal thaw ponds and freeze-thaw collapses image by image, and obtain point targets of thermal thaw ponds and freeze-thaw collapses in the work area.
[0038] The large scale here is to ensure that land features such as thermal melt lakes and freeze-thaw collapses with small-area development characteristics are not missed. The point targets here are point vector files with geographic coordinates.
[0039] The images are visually inspected image by image and grid by grid to determine thermal thaw lakes and freeze-thaw collapses, and the center points are marked.
[0040] S130, performing a buffer analysis on the point target according to a radius of 300 meters, and using the buffer zone to crop the satellite remote sensing image to obtain a satellite remote sensing image of the target area.
[0041] The buffer radius of 300 meters here is for the development characteristics of thermal thaw lakes and freeze-thaw collapses. If other ground objects are extracted, the buffer radius can be adjusted accordingly. The satellite remote sensing image of the target area obtained here is the image pixel with the background image removed or only covering the ground object and the surrounding area.
[0042] After buffering analysis of all point targets, a surface layer file composed of many vector circular patches is obtained. After clipping the image using the surface layer file, the target area image composed of all thermal thaw lakes, freeze-thaw collapses and other ground objects and surrounding area pixels is obtained.
[0043] S140. Select typical areas of thermal thawing lakes and freeze-thaw collapse in the work area, perform threshold segmentation on the near-infrared band of the target area image, preliminarily obtain the vector boundaries of thermal thawing lakes and freeze-thaw collapse in the typical development areas, and manually revise to obtain the precise vector boundaries.
[0044] In order to ensure the number of deep learning training samples, the typical areas of thermal thaw lakes and freeze-thaw collapse can be multiple areas. Threshold segmentation is to save some of the workload of manual visual interpretation and improve interpretation efficiency, but in order to ensure the quality of training samples, the threshold segmentation results still need to be manually revised and checked.
[0045] Through threshold segmentation and manual visual revision, deep learning samples can be produced semi-automatically, efficiently and with high precision. This process can be classified through the threshold segmentation algorithm and reduce the interference of background images, which can significantly improve the efficiency of training sample production.
[0046] S150. Use precise vector boundaries and target area images to create 256×256 pixel training samples of thermal thaw lakes and freeze-thaw collapses for deep learning model training.
[0047] The 256×256 pixel deep learning training samples here are designed for the spatial scale of thermal thaw lakes and freeze-thaw collapses. If you are making training samples for other land objects, you can adjust the sample size appropriately.
[0048] The training samples obtained here are generally positive samples containing ground objects. If the ground object is similar to other ground objects in spectrum or texture, resulting in "different objects with the same spectrum", which leads to mis-extraction, the mis-extracted ground objects can be selected and negative samples can be made. Deep learning model training can be performed on thermal thaw lakes and freeze-thaw collapses at the same time, or they can be trained in two categories separately.
[0049] S160. Apply transfer learning of the training model to the satellite remote sensing images of the target area in the entire work area to predict thermal thaw lakes and freeze-thaw collapse.
[0050] The training model here is generated by training samples from typical development areas of the target area. When making predictions for the entire working area, transfer learning technology needs to be used.
[0051] Optionally, a deep learning model is used to predict only the satellite remote sensing images of the target area in the entire work area, rather than the background images, which greatly improves the prediction efficiency and reduces the erroneous extraction of other "different objects with the same spectrum" objects.
[0052] In a specific implementation method, the DeepLab V3+ algorithm is trained using a sample set to obtain a final DeepLabV3+ model; the trained DeepLab V3+ model is used to predict the target area image of the entire working area obtained by S130, and finally, the thermal thaw lake and frozen thaw lake targets with small area development characteristics are output with high precision and completeness.
[0053] The electronic device structure of the present invention is as follows Figure 2 As shown, the device includes a processor 50, a memory 51, an input device 52 and an output device 53; the number of processors 50 in the device can be one or more. Figure 2 A processor 50 is taken as an example; the processor 50, memory 51, input device 52 and output device 53 in the device can be connected by a bus or other means. Figure 2 The example of connecting through bus is taken in the following.
[0054] The memory 51, as a computer-readable storage medium, can be used to store software programs, computer executable programs and modules, such as program instructions / modules corresponding to the high-precision extraction method for large-area thermal thaw lakes and ponds and freeze-thaw collapses based on high-resolution satellite data in the embodiment of the present invention. The processor 50 executes various functional applications and data processing of the device by running the software programs, instructions and modules stored in the memory 51, that is, realizing the above-mentioned high-precision extraction method for large-area thermal thaw lakes and ponds and freeze-thaw collapses based on high-resolution satellite data.
[0055] The memory 51 may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system and at least one application required for a function; the data storage area may store data created according to the use of the terminal, etc. In addition, the memory 51 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some instances, the memory 51 may further include a memory remotely arranged relative to the processor 50, and these remote memories may be connected to the device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.
[0056] The input device 52 may be used to receive input digital or character information and generate key signal input related to user settings and function control of the device. The output device 53 may include a display device such as a display screen.
[0057] The storage medium comprising the method of the present invention can adopt any combination of one or more computer-readable media. Computer-readable media can be computer-readable signal media or computer-readable storage media, such as but not limited to electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or devices, or any combination of the above. Including electrical connections with one or more wires, portable computer disks, hard disks, random access memories (RAM), read-only memories (ROM), erasable programmable read-only memories (EPROM or flash memory), optical fibers, portable compact disk read-only memories (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the above. Computer-readable storage media can be any tangible medium containing or storing a program, which can be used by an instruction execution system, device or device or used in combination with it.
[0058] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, which carry computer-readable program code. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Computer-readable signal media may also be any computer-readable medium other than a computer-readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device.
[0059] The program code contained on the computer-readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wire, optical cable, RF, etc., or any suitable combination of the foregoing. The computer program code for performing the operation of the present invention may be written in one or more programming languages or a combination thereof, including object-oriented programming languages such as Java, Smalltalk, C++, and conventional procedural programming languages such as "C" language or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider)
[0060] The above are only preferred embodiments of the present invention, which are only used to help understand the method and core ideas of the present application. The protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technicians in this technical field, some improvements and modifications without departing from the principle of the present invention should also be regarded as the protection scope of the present invention.
[0061] The present invention fundamentally solves the problems in the prior art of large manual visual workload and the traditional algorithm's easy omission of small-area ground objects. It predicts and identifies thermal thaw lakes and freeze-thaw collapses in remote sensing images of target areas through manual interpretation and labeling, deep learning model training, and transfer learning, thereby improving the efficiency, integrity, and accuracy of remote sensing extraction of thermal thaw lakes and freeze-thaw collapses.
Claims
1. A high-precision extraction method for thermal thaw lakes and ponds and freeze-thaw collapse based on satellite data, characterized in that: The following steps are involved: S1, obtain satellite remote sensing images with a resolution better than 2 meters in the working area and perform radiation correction, atmospheric correction, orthorectification, and image fusion preprocessing; S2, at a large scale, visually interpret and mark the center points of thermal thaw ponds and freeze-thaw collapses image by image, and obtain the point targets of thermal thaw ponds and freeze-thaw collapses in the work area; S3, perform buffer analysis on the point target according to a radius of 300 meters, use the buffer zone to crop the satellite remote sensing image, and obtain the satellite remote sensing image of the target area; S4, select typical areas of thermal thaw ponds and freeze-thaw collapse in the work area, perform threshold segmentation on the near-infrared band of the satellite remote sensing image of the target area in the typical development area, preliminarily obtain the vector boundaries of thermal thaw ponds and freeze-thaw collapse in the typical development area, and manually revise the precise vector boundaries; S5, using precise vector boundaries and satellite remote sensing images of target areas to create training samples of thermal thaw lakes and freeze-thaw collapses with a pixel size of 256×256, respectively, for deep learning model training; S6, apply the training model transfer learning to the satellite remote sensing images of the target area in the entire work area to predict thermal thaw lakes and freeze-thaw collapse; The preprocessing in S1 is specifically as follows: Obtain cloudless or lightly clouded multispectral satellite remote sensing images covering the work area in the summer period from June to August, with a resolution of better than 2 meters; The S2 is specifically: The thermal thaw ponds and freeze-thaw collapses are visually interpreted from remote sensing images at a large scale of 1:1500 to 1:2000; the thermal thaw ponds and freeze-thaw collapses are visually interpreted grid by grid of remote sensing images to obtain point target vectors with geographical coordinates of thermal thaw ponds and freeze-thaw collapses; The S3 is specifically: After the target center point of the thermal thaw lake and the freeze-thaw collapse is buffered, the buffer area surface layer vector is obtained; the target area image obtained by clipping the remote sensing image of the buffer area includes the image pixels of the thermal thaw lake and the freeze-thaw collapse and the surrounding areas; The S4 is specifically: Typical areas of thermal thawing lakes and freeze-thaw collapses were selected to prepare deep learning training samples. Considering the phenomenon of different spectra of the same object in remote sensing images, multiple typical areas were selected. The precise vector boundaries of thermal thawing lakes and freeze-thaw collapses in typical areas were obtained through threshold segmentation classification and manual visual revision. The S5 is specifically: Taking into account the small-area development characteristics of thaw lakes and freeze-thaw collapses and the efficiency of deep model training, the training sample size is set to 256×256 pixels; deep learning model training is used to classify thaw lakes and freeze-thaw collapses.
2. A computer storage medium, characterized in that: The storage medium stores a computer software program, and when the computer software program is executed by the processor, the high-precision extraction method for thermal thaw lakes and ponds and freeze-thaw collapse based on satellite data as claimed in claim 1 is implemented.
3. An electronic device, comprising a housing and a control panel, characterized in that: The control board includes the storage medium as claimed in claim 2.