An urban land cover change image intelligent detection method, system, terminal and storage medium
By dynamically matching remote sensing image data and interpretation models, and combining human-machine collaborative correction and iterative optimization training, the problem of poor flexibility in urban land cover change detection of remote sensing interpretation systems has been solved, and efficient and intelligent urban land cover change detection has been achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- 深圳市规划和自然资源数据管理中心(深圳市空间地理信息中心)
- Filing Date
- 2026-04-23
- Publication Date
- 2026-05-29
AI Technical Summary
Existing remote sensing interpretation systems struggle to dynamically match models and data to operational needs in urban land cover change detection, resulting in poor flexibility and low adaptability.
By acquiring the detection task configuration parameters, dynamically matching the interpretation model and the remote sensing image data to be tested, using a deep learning model for inference analysis, combining human-machine collaborative correction to generate changed patches, and using the patches and original image data as new samples for iterative optimization training to form the target interpretation model.
It has enabled automated processing and efficient change detection of remote sensing images, improved the system's flexibility and adaptability, ensured that the detection results meet business standards, reduced the cost of manual intervention, and achieved adaptive optimization and efficiency of the model.
Smart Images
Figure CN122116056A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of remote sensing image processing technology, and in particular to an intelligent detection method, system, terminal, and computer-readable storage medium for urban land cover change images. Background Technology
[0002] With the acceleration of urbanization, urban land cover is changing rapidly. Accurate and timely acquisition of land cover change information is of great significance for urban planning, environmental monitoring, resource management, and other fields. Currently, change detection based on remote sensing imagery is the main means of obtaining this information.
[0003] Traditional remote sensing interpretation systems typically employ fixed processing flows and algorithm models, making it difficult to dynamically adjust to changing business needs (such as different detection targets and accuracy requirements). When faced with different task scenarios, it is often necessary to manually reconfigure underlying parameters or replace software, resulting in poor system adaptability and difficulty in meeting diverse business requirements.
[0004] Therefore, existing technologies still need to be improved and developed. Summary of the Invention
[0005] The main objective of this invention is to provide an intelligent detection method, system, terminal, and computer-readable storage medium for urban land cover change images. This aims to solve the problem that existing systems are unable to dynamically match models and data according to business needs when detecting urban land cover change, resulting in poor flexibility and low adaptability.
[0006] To achieve the above objectives, the present invention provides an intelligent detection method for urban land cover change images, which includes the following steps: Obtain the detection task configuration parameters, and determine the interpretation model and the remote sensing image data to be tested based on the detection task configuration parameters; The interpretation model is used to perform inference analysis on multi-temporal remote sensing images to generate initial change detection patches; Based on the user's correction operation on the initial change detection patch, a corrected change patch is generated; The changed patches and their associated original image data are stored as new samples in the sample library. The interpretation model is iteratively optimized and trained using the updated data in the sample library to obtain the target interpretation model. The target interpretation model is then used to infer the remote sensing image data to be measured, and the urban land cover change detection results are output.
[0007] Furthermore, determining the interpretation model and the remote sensing image data to be tested based on the detection task configuration parameters includes: The type of target land cover change detection task is determined based on the detection task configuration parameters. Based on the type of the target land cover change detection task, the corresponding remote sensing image data to be tested and the interpretation model are matched.
[0008] Furthermore, the step of using the interpretation model to perform inference analysis on multi-temporal remote sensing images and generate initial change detection patches includes: The remote sensing image data to be measured is preprocessed to generate multi-temporal remote sensing image slices; The interpretation model is used to infer the multi-temporal remote sensing image slices to generate AI cue patches, and the AI cue patches are aggregated to generate initial change detection patches.
[0009] Furthermore, the step of generating a corrected change patch based on the user's correction operation on the initial change detection patch includes: Based on the user's modification instructions for the initial change detection patch on the visual interface, the geometric boundary of the initial change detection patch is corrected to obtain the target change detection patch; Based on the user's classification confirmation instruction for the target change detection patch, the land feature category of the target change detection patch is determined, and a corrected change patch is generated.
[0010] Furthermore, storing the changed patches and their associated original image data as new samples in the sample library includes: Extract the original remote sensing image data for the time period corresponding to the changed patches; The changed patches are used as label data, and the original remote sensing image data are used as sample data to construct a training sample set and store it in the sample library.
[0011] Furthermore, the step of iteratively optimizing and training the interpretation model using updated data from the sample library to obtain a target interpretation model, and then using the target interpretation model to infer the remote sensing image data to be measured, outputting urban land cover change detection results, includes: The interpretation model is incrementally trained using the updated training sample set in the sample library to obtain the target interpretation model; The target interpretation model is used to perform secondary inference on the current batch of remote sensing image data to be measured or the newly acquired remote sensing image data to be measured, and output the detection results of urban land cover change.
[0012] Furthermore, the process of generating the corrected changed patch and using the changed patch and its associated original image data as new samples also includes: The corrected changes in the image patches are subjected to quality checks, and unqualified patches are selected. Remove the unqualified patches from the corrected changed patches; The unqualified patches include patches that do not meet the preset geometric precision and patches that do not meet the preset logical rules.
[0013] Furthermore, to achieve the above objectives, the present invention also provides an intelligent detection system for urban land cover change images. This system is used to implement the intelligent detection method for urban land cover change images as described above. The intelligent detection system includes: The parameter configuration module is used to obtain the detection task configuration parameters and determine the interpretation model and the remote sensing image data to be tested based on the detection task configuration parameters. The reasoning and analysis module is used to perform reasoning and analysis on multi-temporal remote sensing images using the interpretation model to generate initial change detection patches; The patch correction module is used to generate corrected changed patches based on the user's correction operation on the initial change detection patches; The sample update module is used to store the changed patches and their associated original image data as new samples into the sample library. The result output module is used to iteratively optimize and train the interpretation model using the updated data in the sample library to obtain the target interpretation model, and to use the target interpretation model to infer the remote sensing image data to be tested, and output the detection results of urban land cover change.
[0014] In addition, to achieve the above objectives, the present invention also provides a terminal, wherein the terminal includes: a memory, a processor, and an intelligent detection program for urban land cover change images stored in the memory and executable on the processor, wherein when the intelligent detection program for urban land cover change images is executed by the processor, it implements the steps of the intelligent detection method for urban land cover change images as described above.
[0015] In addition, to achieve the above objectives, the present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores an intelligent detection program for urban land cover change images, and when the intelligent detection program for urban land cover change images is executed by a processor, it implements the steps of the intelligent detection method for urban land cover change images as described above.
[0016] The beneficial effects of this invention are as follows: By acquiring the configuration parameters of the detection task, this invention achieves dynamic matching between business requirements, the interpretation model, and the remote sensing image data to be tested, endowing the system with extremely high flexibility and adaptability; by using the interpretation model to perform inference analysis on multi-temporal remote sensing images, it realizes automated processing of massive images and generation of initial change detection patches, significantly improving operational efficiency. The introduction of a human-machine collaboration mechanism, utilizing user correction operations on initial change detection patches, effectively compensates for the algorithm's accuracy shortcomings in complex scenarios, ensuring that the corrected change patches meet business standards. This application overcomes the bottleneck of difficult sample acquisition in the remote sensing field by storing change patches and associated original image data as new samples in the sample library; then, iterative optimization training of the interpretation model is performed using updated data in the sample library to obtain the target interpretation model and apply it to subsequent inference, achieving adaptive optimization of the model. This closed-loop design enables the system to have a self-evolving capability of "becoming more accurate with use," continuously reducing the cost of manual intervention as business accumulates, ultimately achieving efficient, intelligent, and sustainable development of urban land cover change detection. Attached Figure Description
[0017] Figure 1 This is a flowchart of a preferred embodiment of the intelligent detection method for urban land cover change images of the present invention; Figure 2 This is a closed-loop business framework diagram of the intelligent detection method for urban land cover change images in this invention, driven by both business and model. Figure 3 This is a diagram of the five-layer technical architecture of the intelligent detection method for urban land cover change images in this invention, which is characterized by layered decoupling. Figure 4 This is a technical roadmap of the land cover classification model based on the HR-Net architecture in the intelligent detection method for urban land cover change images of this invention; Figure 5 This is a technical roadmap of the land cover change detection model based on the Transformer architecture in the intelligent detection method for urban land cover change images of this invention; Figure 6 This is a flowchart of model training and iterative optimization in the intelligent detection method for urban land cover change images of the present invention; Figure 7 This is a structural diagram of a preferred embodiment of the intelligent detection system for urban land cover change images of the present invention; Figure 8 This is a structural diagram of a preferred embodiment of the terminal of the present invention. Detailed Implementation
[0018] This application provides an intelligent detection method, system, terminal, and storage medium for urban land cover change images. To make the objectives, technical solutions, and effects of this application clearer and more explicit, the following detailed description is provided with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only for explaining this application and are not intended to limit this application.
[0019] It will be understood by those skilled in the art that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. It should also be understood that terms such as those defined in general dictionaries should be understood to have the same meaning as in the context of the prior art, and should not be interpreted in an idealized or overly formal sense unless specifically defined as herein.
[0020] Furthermore, if the embodiments of this invention involve descriptions such as "first" or "second," these descriptions are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined with "first" or "second" may explicitly or implicitly include at least one of those features. Additionally, the technical solutions of the various embodiments can be combined with each other, but this must be based on the ability of those skilled in the art to implement them. If the combination of technical solutions is contradictory or impossible to implement, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed by this invention.
[0021] like Figure 1 and Figure 2 As shown, the intelligent detection method for urban land cover change images provided in this embodiment of the invention aims to construct a closed-loop business framework driven by both business needs and models. This framework demonstrates a complete closed-loop logic starting from business requirement analysis (corresponding to step S10 below), through AI inference to generate clues (corresponding to step S20 below), human-machine collaborative correction (corresponding to step S30 below), and finally feeding back high-quality results as samples (corresponding to step S40 below) to drive model iteration (corresponding to step S50 below).
[0022] This method runs in such Figure 3 The architecture is based on a layered and decoupled five-layer technical architecture. This architecture includes a basic support layer, a data layer, a model layer, a training layer, and a service layer. The basic support layer provides computing power and framework support; the data layer manages multi-source images and samples; the model layer encapsulates various interpretation algorithms; the training layer is responsible for iterative optimization of the model; and the service layer provides inference interfaces. This architectural design enables the layers to interact through standardized interfaces, achieving end-to-end engineering support from data loading to business deployment.
[0023] Specifically, the intelligent detection method for urban land cover change images includes the following steps: S10. Obtain the detection task configuration parameters, and determine the interpretation model and the remote sensing image data to be tested based on the detection task configuration parameters.
[0024] The purpose of this step is to dynamically match the most suitable algorithm model and data source according to the specific business application scenario (such as land change survey, illegal building investigation and handling, or water body monitoring), so as to achieve a precise mapping from "business needs" to "technical solutions".
[0025] In this embodiment, the service layer first receives the detection task configuration parameters issued by the service terminal. The detection task configuration parameters include at least the task type (such as "land cover change detection" or "land cover classification"), the target area range, and the accuracy requirements.
[0026] Further, in step S10, determining the interpretation model and the remote sensing image data to be tested based on the detection task configuration parameters specifically includes the following sub-steps: Step S11: Determine the type of target land cover change detection task based on the detection task configuration parameters.
[0027] The system parses the configuration parameters and identifies the target task type. For example, if the parameter is specified as "refined land cover classification", the system determines that high-resolution feature extraction capability is required; if the parameter is specified as "large-scale land cover change", the system determines that global semantic analysis capability is required.
[0028] Step S12: Match the corresponding remote sensing image data to be tested and the interpretation model according to the type of the target land cover change detection task.
[0029] Based on the task type, the data layer automatically retrieves and loads multi-source remote sensing image data (such as Gaofen series satellite imagery or UAV orthophotos) of the corresponding time phase and resolution.
[0030] At the same time, the model layer calls the corresponding interpretation model from the model library according to the task type: If the task is land cover classification, then a model based on the HR-Net (High-Resolution Network) architecture is matched. This model maintains a high-resolution feature flow through multi-scale parallel branches, making it suitable for extracting fine land cover boundaries.
[0031] If the task is change detection, a model based on the Transformer architecture is matched. This model introduces a self-attention mechanism to extract global semantic representations from dual-temporal images and utilizes cross-attention to fuse features, making it suitable for detecting complex ground feature changes.
[0032] S20. Using the interpretation model, perform reasoning analysis on multi-temporal remote sensing images to generate initial change detection patches.
[0033] The purpose of this step is to leverage the powerful feature extraction capabilities of deep learning models to quickly identify potential areas of change from massive amounts of remote sensing imagery, generating "clues" at the computer vision level.
[0034] Furthermore, the step of using the interpretation model to perform inference analysis on multi-temporal remote sensing images to generate initial change detection patches specifically includes: Step S21: Preprocess the remote sensing image data to be tested to generate multi-temporal remote sensing image slices.
[0035] The data layer performs standardization processing on the loaded raw remote sensing images, including geometric correction, radiometric calibration, and image registration, to ensure spatial consistency between the two temporal images. Subsequently, the large-scale images are divided into image slices of the model input size (e.g., 512×512 pixels) for batch reading by the model.
[0036] Step S22: Use the interpretation model to infer the multi-temporal remote sensing image slices, generate AI cue patches, and aggregate the AI cue patches to generate initial change detection patches.
[0037] The model layer loads the selected model (such as a Transformer change detection model) for forward propagation: Feature extraction: The dual-temporal images are digitized into a matrix, and a self-attention mechanism is introduced by segmenting image patches to extract features with global semantics.
[0038] Change information decoding: A multi-scale change information tensor is constructed by fusing dual-temporal features using a decoder. In this embodiment, an OCR (Object-Contextual Representation) module is introduced to aggregate object-level contextual information, enhance pixel semantic consistency, and reduce "salt-and-pepper noise".
[0039] Patch generation: The output probability map of change is converted into a binary change map by setting a threshold (such as 0.5), and after morphological processing (denoising, hole filling), it is aggregated into vectorized AI cue patches.
[0040] As can be seen, step S20 uses a deep learning model to quickly filter out potential change areas from massive remote sensing images.
[0041] In one embodiment, the model inference task is executed using the PyTorch framework and Ascend computing power provided by the basic support layer. Based on the task type determined in S10, the system calls are as follows: Figure 3 or Figure 4 Inference is performed using the different model architectures shown: Scenario 1: Land cover classification task (corresponding to) Figure 3 ) If the task is land cover classification, the system loads a land cover classification model based on the HR-Net architecture. For example... Figure 3 As shown, the model maintains high-resolution feature flow through multi-scale parallel branches. During the training phase, a contrastive learning module is introduced to construct positive and negative sample pairs for easily confused features output by the encoder. By constraining the contrastive loss, the model increases the inter-class distance and reduces the intra-class difference, thereby enhancing its discriminative ability. During inference, the model outputs a pixel-level classification probability map and optimizes the generation of vectorized initial classification patches.
[0042] Scenario 2: Land cover change detection task (corresponding to) Figure 4 ) If the task is land cover change detection, the system loads a change detection model based on the Transformer architecture. For example... Figure 4 As shown, the model first digitizes the dual-temporal images and extracts dual-temporal features with global semantic representation using a self-attention mechanism; then, it fuses the features through cross-attention. In the decoding stage, an OCR module is introduced to aggregate object-level contextual information to enhance pixel semantic consistency and reduce "salt-and-pepper noise." Finally, a change probability map is generated and transformed into initial change detection patches. In other words, as... Figure 5 As shown, images from two different time phases are first digitized, and a self-attention mechanism is introduced through image patch segmentation to extract features with global semantic representation. Then, a cross-attention mechanism is used to fuse the features from the two time phases. In the decoding stage, an OCR (Object Context Representation) module is introduced to aggregate object-level contextual information, thereby enhancing the consistency of pixel semantics and reducing "salt and pepper noise". Finally, the output change probability map is transformed into a binary change map, and after morphological processing, a vectorized AI cue patch is generated.
[0043] It should be noted that, Figure 4 The meaning of the four-dimensional tensor structure in the text is as follows: Height (H): The height refers to the number of pixels in the vertical direction of the image.
[0044] Width (W): Width refers to the number of pixels in the horizontal direction of the image.
[0045] Channels (C): The number of channels refers to the number of color or feature channels in an image. For example, a regular RGB color image has 3 channels (red, green, and blue); while a feature map processed by a neural network may have dozens or even hundreds of channels, each representing an abstract feature.
[0046] Batch (B): Batch size or batch size. In deep learning model training or inference, to improve efficiency, multiple images are usually processed simultaneously. This "number of images processed at once" is the batch size. The cube in the diagram represents the tensor of an image. Multiple images (B images) stacked in the "batch" dimension constitute the complete four-dimensional input data [B,H,W,C].
[0047] It should be noted that, Figure 5 This describes the normalization of images from two different time points in a land cover change detection task. Specifically: T1 is the image from time phase 1, representing remote sensing images acquired at an earlier time point (such as last year's satellite image); T2 is the image from time phase 2, representing remote sensing images acquired at a later time point (such as this year's satellite image). T1 and T2 serve as the baseline for change comparison. norm The result is the normalized image of phase 1; T2 norm This is the result of normalizing the Phase 2 image.
[0048] S30. Based on the user's correction operation on the initial change detection patch, generate the corrected change patch.
[0049] The purpose of this step is to refine the rough map patches generated by AI through a "human-machine collaboration" mechanism, combining the experience and knowledge of human experts, to ensure that the output meets strict business standards (such as the topological rules of land surveys).
[0050] In this embodiment, the service layer overlays the generated AI clue patches onto the original image and pushes them to the visualization terminal interface of the business personnel. The step of generating corrected change patches based on the user's correction operation on the initial change detection patches specifically includes: Step S31: Based on the user's modification instructions for the initial change detection patch on the visualization interface, the geometric boundary of the initial change detection patch is corrected to obtain the target change detection patch.
[0051] Business personnel use interactive tools (such as editing vertices and smoothing boundaries) to adjust the geometry of the patches so that they precisely fit the edges of ground features in the image, eliminating jagged edges or overflows generated by AI.
[0052] Step S32: Based on the user's classification confirmation instruction for the target change detection patch, determine the land feature category of the target change detection patch and generate the corrected change patch.
[0053] Business personnel correct potential classification errors by using image textures and contextual information (e.g., correcting "temporary soil stockpiles" to "construction land") and assign standard feature codes to generate change patches that conform to business standards.
[0054] S40. The changed patches and their associated original image data are stored as new samples in the sample library.
[0055] The purpose of this step is to transform high-quality output data generated during business production into valuable training samples, thereby solving the problems of difficult sample acquisition and high annotation costs in the field of remote sensing.
[0056] In this embodiment, storing the changed patches and their associated original image data as new samples in the sample library specifically includes: Step S41: Extract the original remote sensing image data within the time period corresponding to the changed patch.
[0057] It should be noted that the data layer precisely cuts out the corresponding dual-temporal image blocks from the historical image pool based on the spatial location and temporal attributes of the changed image patches.
[0058] Step S42: Use the changed patches as label data and the original remote sensing image data as sample data to construct a training sample set and store it in the sample library.
[0059] It should be noted that the system uses the cropped image as input data, rasterizes the vector boundaries of the changed patches to serve as label data, and automatically extracts their land cover category attributes. After quality checks (such as non-empty checks and format validation), the (Image, Label) pair is stored in the sample library of the data layer. This allows the sample library to continuously "grow" as business operations progress, accumulating localized features.
[0060] Furthermore, the process between generating the corrected changed patch and using the changed patch and the associated original image data as new samples (steps S30 and S40) further includes: The corrected changed patches are subjected to quality checks to screen out unqualified patches, including patches that do not meet the preset geometric precision and patches that do not meet the preset logical rules.
[0061] Remove the unqualified patches from the corrected changed patches.
[0062] In this embodiment, the screening process involves performing a topological check on the corrected changed patches to remove unqualified patches with excessively small areas (e.g., less than 10 square meters) or distorted shapes (e.g., self-intersecting). Removal is also performed based on preset logical rules (e.g., "water bodies cannot appear on rooftops") to filter out patches that clearly violate geographical common sense. Only patches that pass the quality check are added to step S40 as new samples, thus ensuring the purity of the sample library and preventing "dirty data" from contaminating the model.
[0063] S50. The interpretation model is iteratively optimized and trained using the updated data in the sample library to obtain the target interpretation model. The target interpretation model is then used to infer the remote sensing image data to be measured, and the urban land cover change detection result is output.
[0064] The purpose of this step is to achieve "continuous evolution" of the model. By using newly accumulated samples to incrementally train the model, it adapts to new landforms and localized features, thereby addressing the problems of insufficient model generalization ability and obsolescence.
[0065] In this embodiment, the automatic or timed triggering of the model iteration process in the training layer specifically includes: Step S51: Incrementally train the interpretation model using the updated training sample set in the sample library to obtain the target interpretation model.
[0066] It should be noted that the training layer extracts the latest accumulated samples (including historical difficult examples and new class samples) from the sample library, and configures the loss function (such as introducing contrastive learning loss and constructing positive and negative sample pairs to increase the inter-class distance) and optimizer parameters.
[0067] Distributed training is performed using the computing power of the underlying support layer. During training, the system monitors the accuracy of the validation set (e.g., IoU, intersection-union ratio) in real time. When the IoU value converges or reaches a preset threshold (e.g., 0.85), the optimal weights are saved, and the target interpretation model is generated.
[0068] Step S52: Use the target interpretation model to perform secondary inference on the current batch of remote sensing image data to be tested or the newly accessed remote sensing image data to be tested, and output the urban land cover change detection results.
[0069] The service layer automatically switches the old version of the model to the newly trained target interpretation model.
[0070] It should be noted that the system uses this target interpretation model to perform final inference calculations on the data to be measured loaded in S10 (or newly accessed real-time images in the service layer). The output results include change detection patch vector files, change type attribute tables, and accuracy verification reports. These results are directly pushed to business application systems (such as the land survey system) through the API interface, completing the entire closed-loop process.
[0071] As can be seen, step S50 achieves the "continuous evolution" of the model. In one embodiment, the training layer automatically triggers or periodically triggers the model iteration process. Figure 6 As shown, this process follows a closed-loop path of "sample preparation—model tuning—accuracy verification—results consolidation." The training layer extracts the latest accumulated samples (including historical difficult examples and new category samples) from the sample library for incremental training. When the model's accuracy (e.g., IoU) on the validation set converges or reaches a preset threshold (e.g., 0.85), the optimal weights are saved, and a target interpretation model is generated. Subsequently, the service layer automatically switches the model version and uses this target interpretation model to perform secondary inference on the current batch or newly added remote sensing image data to be tested, ultimately outputting high-precision urban land cover change detection results.
[0072] Through the above embodiments, the present invention achieves closed-loop optimization from "business requirements" to "model output" and then to "sample feedback", which significantly improves the accuracy and efficiency of urban land cover change detection.
[0073] The beneficial effects of this invention are as follows: By acquiring the configuration parameters of the detection task, this invention achieves dynamic matching between business requirements, the interpretation model, and the remote sensing image data to be tested, endowing the system with extremely high flexibility and adaptability; by using the interpretation model to perform inference analysis on multi-temporal remote sensing images, it realizes automated processing of massive images and generation of initial change detection patches, significantly improving operational efficiency. The introduction of a human-machine collaboration mechanism, utilizing user correction operations on initial change detection patches, effectively compensates for the algorithm's accuracy shortcomings in complex scenarios, ensuring that the corrected change patches meet business standards. This application overcomes the bottleneck of difficult sample acquisition in the remote sensing field by storing change patches and associated original image data as new samples in the sample library; then, iterative optimization training of the interpretation model is performed using updated data in the sample library to obtain the target interpretation model and apply it to subsequent inference, achieving adaptive optimization of the model. This closed-loop design enables the system to have a self-evolving capability of "becoming more accurate with use," continuously reducing the cost of manual intervention as business accumulates, ultimately achieving efficient, intelligent, and sustainable development of urban land cover change detection.
[0074] Furthermore, such as Figure 7 As shown, based on the above-mentioned intelligent detection method for urban land cover change images, the present invention also provides an intelligent detection system for urban land cover change images, the intelligent detection system for urban land cover change images comprising: Parameter configuration module 51 is used to obtain detection task configuration parameters and determine the interpretation model and remote sensing image data to be tested based on the detection task configuration parameters; The reasoning and analysis module 52 is used to perform reasoning and analysis on multi-temporal remote sensing images using the interpretation model to generate initial change detection patches; The patch correction module 53 is used to generate corrected changed patches based on the user's correction operation on the initial change detection patches; The sample update module 54 is used to store the changed patches and associated original image data as new samples into the sample library. The result output module 55 is used to iteratively optimize and train the interpretation model using the updated data in the sample library to obtain the target interpretation model, and use the target interpretation model to infer the remote sensing image data to be tested, and output the urban land cover change detection results.
[0075] Furthermore, such as Figure 8 As shown, based on the above-mentioned intelligent detection method and system for urban land cover change images, the present invention also provides a terminal, which includes a processor 10, a memory 20 and a display 30. Figure 8 Only some of the terminal components are shown; however, it should be understood that it is not required to implement all of the components shown, and more or fewer components may be implemented instead.
[0076] In some embodiments, the memory 20 may be an internal storage unit of the terminal, such as a hard disk or memory. In other embodiments, the memory 20 may be an external storage device of the terminal, such as a plug-in hard disk, smart media card (SMC), secure digital card (SD), flash card, etc. Further, the memory 20 may include both internal and external storage devices. The memory 20 is used to store application software and various types of data installed on the terminal, such as the program code installed on the terminal. The memory 20 can also be used to temporarily store data that has been output or will be output. In one embodiment, the memory 20 stores an intelligent detection program 40 for urban land cover change images, which can be executed by the processor 10 to implement the intelligent detection method for urban land cover change images in this application.
[0077] In some embodiments, the processor 10 may be a central processing unit (CPU), an intelligent computing bare metal server, a microprocessor, or other data processing chip, used to run program code stored in the memory 20 or process data, such as executing the intelligent detection method for urban land cover change images.
[0078] In some embodiments, the display 30 may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen. The display 30 is used to display information on the terminal and to display a visual user interface. The components of the terminal communicate with each other via a system bus.
[0079] The present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores an intelligent detection program for urban land cover change images, and the intelligent detection program for urban land cover change images, when executed by a processor, implements the steps of the intelligent detection method for urban land cover change images as described above.
[0080] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal that includes that element.
[0081] Of course, those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided by this invention can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.
[0082] It should be understood that the application of the present invention is not limited to the examples above. Those skilled in the art can make improvements or modifications based on the above description, and all such improvements and modifications should fall within the protection scope of the appended claims.
Claims
1. A method for intelligent detection of urban land cover change images, characterized in that, The intelligent detection method for urban land cover change images includes the following steps: Obtain the detection task configuration parameters, and determine the interpretation model and the remote sensing image data to be tested based on the detection task configuration parameters; The interpretation model is used to perform inference analysis on multi-temporal remote sensing images to generate initial change detection patches; Based on the user's correction operation on the initial change detection patch, a corrected change patch is generated; The changed patches and their associated original image data are stored as new samples in the sample library. The interpretation model is iteratively optimized and trained using the updated data in the sample library to obtain the target interpretation model. The target interpretation model is then used to infer the remote sensing image data to be measured, and the urban land cover change detection results are output.
2. The intelligent detection method for urban land cover change images according to claim 1, characterized in that, The step of determining the interpretation model and the remote sensing image data to be tested based on the detection task configuration parameters includes: The type of target land cover change detection task is determined based on the detection task configuration parameters. Based on the type of the target land cover change detection task, the corresponding remote sensing image data to be tested and the interpretation model are matched.
3. The intelligent detection method for urban land cover change images according to claim 1, characterized in that, The step of using the interpretation model to perform inference analysis on multi-temporal remote sensing images and generate initial change detection patches includes: The remote sensing image data to be measured is preprocessed to generate multi-temporal remote sensing image slices; The interpretation model is used to infer the multi-temporal remote sensing image slices to generate AI cue patches, and the AI cue patches are aggregated to generate initial change detection patches.
4. The intelligent detection method for urban land cover change images according to claim 1, characterized in that, The step of generating a corrected change patch based on the user's correction operation on the initial change detection patch includes: Based on the user's modification instructions for the initial change detection patch on the visual interface, the geometric boundary of the initial change detection patch is corrected to obtain the target change detection patch; Based on the user's classification confirmation instruction for the target change detection patch, the land feature category of the target change detection patch is determined, and a corrected change patch is generated.
5. The intelligent detection method for urban land cover change images according to claim 1, characterized in that, The step of storing the changed patches and their associated original image data as new samples in the sample library includes: Extract the original remote sensing image data for the time period corresponding to the changed patches; The changed patches are used as label data, and the original remote sensing image data are used as sample data to construct a training sample set and store it in the sample library.
6. The intelligent detection method for urban land cover change images according to claim 1, characterized in that, The process involves iteratively optimizing and training the interpretation model using updated data from the sample database to obtain a target interpretation model. This target interpretation model is then used to infer the results of the remote sensing image data to be measured, outputting urban land cover change detection results. This includes: The interpretation model is incrementally trained using the updated training sample set in the sample library to obtain the target interpretation model; The target interpretation model is used to perform secondary inference on the current batch of remote sensing image data to be measured or the newly acquired remote sensing image data to be measured, and output the detection results of urban land cover change.
7. The intelligent detection method for urban land cover change images according to claim 1, characterized in that, The process of generating the corrected changed patch and using the changed patch and associated original image data as new samples also includes: The corrected changes in the image patches are subjected to quality checks, and unqualified patches are selected. Remove the unqualified patches from the corrected changed patches; The unqualified patches include patches that do not meet the preset geometric precision and patches that do not meet the preset logical rules.
8. An intelligent detection system for urban land cover change images, characterized in that, The intelligent detection system for urban land cover change images is used to implement the intelligent detection method for urban land cover change images as described in any one of claims 1-7, and the intelligent detection system for urban land cover change images includes: The parameter configuration module is used to obtain the detection task configuration parameters and determine the interpretation model and the remote sensing image data to be tested based on the detection task configuration parameters. The reasoning and analysis module is used to perform reasoning and analysis on multi-temporal remote sensing images using the interpretation model to generate initial change detection patches; The patch correction module is used to generate corrected changed patches based on the user's correction operation on the initial change detection patches; The sample update module is used to store the changed patches and their associated original image data as new samples into the sample library. The result output module is used to iteratively optimize and train the interpretation model using the updated data in the sample library to obtain the target interpretation model, and to use the target interpretation model to infer the remote sensing image data to be tested, and output the detection results of urban land cover change.
9. A terminal, characterized in that, The terminal includes: a memory, a processor, and an intelligent detection program for urban land cover change images stored in the memory and executable on the processor. When the intelligent detection program for urban land cover change images is executed by the processor, it implements the steps of the intelligent detection method for urban land cover change images as described in any one of claims 1-7.
10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores an intelligent detection program for urban land cover change images, which, when executed by a processor, implements the steps of the intelligent detection method for urban land cover change images as described in any one of claims 1-7.