Change region identification method and related device

Abstract

The invention discloses a change region identification method and a related device, which are applied to scenes such as artificial intelligence. The method comprises the following steps: acquiring a first target difference image between a first target image at a first target time and a second target image at a second target time and a second target difference image between the first target image and a data vectorization image at a third target time, feature extraction is carried out on the first target difference image and the second target difference image through a feature extraction layer in the change region recognition model to obtain a plurality of target feature images, and feature fusion based on different feature dimensions is carried out on the plurality of target feature images through a feature fusion layer to obtain a plurality of target fusion images; and performing change area identification on the plurality of target fusion images through the identification layer to obtain a target change area. According to the method, change information of spatial data represented by a new target data vectorized image compared with an old target image is considered, so that a target change area is more real and effective.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to a method and related apparatus for identifying changing regions. Background Technology

[0002] With the rapid development of image processing technology, it is possible to identify areas of spatial change by using old and new spatial imagery. For example, by using old and new geospatial imagery, it is possible to identify areas of building change in geospatial areas, thereby discovering changed buildings in geospatial space.

[0003] In related technologies, the change region identification method refers to: using a change region identification model to identify the change regions in the differential image between a new spatial image and an old spatial image, thereby obtaining the change regions in the space.

[0004] However, when spatial data obtained from other channels changes compared to the spatial data represented by old images, the above method only identifies the changed areas through the difference image between the new images and the spatial data obtained from other channels without acquiring the data difference information. This results in a certain deviation in the identification results of the changed areas, which are not realistic and effective enough. Summary of the Invention

[0005] To address the aforementioned technical problems, this application provides a method and related apparatus for identifying change regions, which can more accurately identify change regions of the target space at the first target time, making the target change regions of the target space at the first target time more realistic and effective.

[0006] The embodiments of this application disclose the following technical solutions:

[0007] On one hand, embodiments of this application provide a method for identifying change regions, the method comprising:

[0008] Acquire a first target difference image between a first target image at a first target time and a second target image at a second target time, and a second target difference image between the first target image and the data vectorized image of the target space at a third target time; the first target time and the third target time are after the second target time;

[0009] By using the feature extraction layer in the change region recognition model, features are extracted from the first target difference image and the second target difference image to obtain multiple target feature images; each target feature image includes a feature image corresponding to the first target difference image and a feature image corresponding to the second target difference image, and different target feature images correspond to different feature dimensions;

[0010] The feature fusion layer in the change region identification model performs feature fusion on multiple target feature images based on different feature dimensions to obtain multiple target fused images; different target fused images correspond to different feature dimensions.

[0011] The change region identification model uses an identification layer to identify change regions in multiple target fused images, thereby obtaining the target change region in the target space at the first target time.

[0012] On the other hand, embodiments of this application provide a change region identification device, the device comprising: an acquisition unit, an extraction unit, a fusion unit, and an identification unit;

[0013] The acquisition unit is used to acquire a first target difference image between a first target image of the target space at a first target time and a second target image at a second target time, and a second target difference image between the first target image and the data vectorized image of the target space at a third target time; the first target time and the third target time are after the second target time;

[0014] The extraction unit is used to extract features from the first target difference image and the second target difference image through the feature extraction layer in the change region recognition model to obtain multiple target feature images; each target feature image includes a feature image corresponding to the first target difference image and a feature image corresponding to the second target difference image, and different target feature images correspond to different feature dimensions;

[0015] The fusion unit is used to perform feature fusion on multiple target feature images based on different feature dimensions through the feature fusion layer in the change region identification model to obtain multiple target fused images; different target fused images correspond to different feature dimensions.

[0016] The recognition unit is used to perform change region recognition on multiple target fusion images through the recognition layer in the change region recognition model, and obtain the target change region of the target space at the first target time.

[0017] On the other hand, embodiments of this application provide a computer device, the computer device including a processor and a memory:

[0018] The memory is used to store computer programs and to transfer the computer programs to the processor;

[0019] The processor is configured to execute the method described in any of the foregoing aspects according to instructions in the computer program.

[0020] On the other hand, embodiments of this application provide a computer-readable storage medium for storing a computer program that, when run on a computer device, causes the computer device to perform the methods described in any of the foregoing aspects.

[0021] On the other hand, embodiments of this application provide a computer program product, including a computer program that, when run on a computer device, causes the computer device to perform the method described in any of the foregoing aspects.

[0022] As can be seen from the above technical solution, firstly, a first target difference image is acquired between a first target image at a first target time and a second target image at a second target time, and a second target difference image is acquired between the first target image and the data vectorized image of the target space at a third target time, wherein the first target time and the third target time are after the second target time. This method not only acquires the difference image between the new target image and the old target image in the target space, providing image difference information of the target space, but also acquires the difference image between the new target image and the relatively new target data vectorized image in the target space, providing data difference information of the target space.

[0023] Then, the first target difference image and the second target difference image are input into the feature extraction layer of the change region recognition model. Feature extraction outputs multiple target feature images, where each target feature image includes the feature image corresponding to the first target difference image and the feature image corresponding to the second target difference image. Different target feature images correspond to different feature dimensions. The multiple target feature images are then input into the feature fusion layer of the change region recognition model. Based on feature fusion of different feature dimensions, multiple target fused images are output, where different target fused images correspond to different feature dimensions. This method extracts multiple feature images corresponding to both the difference image between the new target image and the old target image in the target space, and the difference image between the new target image and the relatively new target data vectorized image in the target space. Image difference information and data difference information are represented by different receptive fields, and multiple feature images with different feature dimensions are fused into multiple fused images, thus interactively fusing image difference information and data difference information with different receptive fields.

[0024] Finally, the fused images of multiple targets are input into the recognition layer of the change region recognition model, and the change region recognition outputs the target change region of the target space at the first target time. Based on the image difference information and data difference information of different receptive fields after interactive fusion, this method considers the change information of spatial data of the target space relative to the new target data vectorized image compared with the old target image image, and more accurately identifies the change region of the target space at the first target time, making the target change region of the target space at the first target time more realistic and effective. Attached Figure Description

[0025] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0026] Figure 1 This is a system schematic diagram of a change region identification method provided in an embodiment of this application;

[0027] Figure 2 A flowchart illustrating a method for identifying change regions provided in an embodiment of this application;

[0028] Figure 3 A schematic diagram of a data vectorized image provided in an embodiment of this application;

[0029] Figure 4 This is a schematic diagram of the structure of a change region identification model provided in an embodiment of this application;

[0030] Figure 5 This is a schematic diagram of a change region identification result provided in an embodiment of this application;

[0031] Figure 6 A structural diagram of a change area identification device provided in an embodiment of this application;

[0032] Figure 7 A structural diagram of a server provided in an embodiment of this application;

[0033] Figure 8 This is a structural diagram of a terminal provided in an embodiment of this application. Detailed Implementation

[0034] The embodiments of this application will now be described with reference to the accompanying drawings.

[0035] Currently, change region identification models can be used to identify change regions in space by analyzing the difference image between new and old imagery. However, in practical applications, spatial data is often acquired through multiple channels. When the spatial data obtained from other channels differs from the spatial data represented by the old imagery, the change regions identified by the aforementioned methods through the difference imagery between the new and old imagery may deviate, resulting in less accurate and effective identification results.

[0036] This application provides a method for identifying change regions. It obtains a difference image between a new target image and an old target image in the target space to provide image difference information for the target space, and also obtains a difference image between the new target image and a relatively new target data vectorized image in the target space to provide data difference information for the target space. Multiple feature images corresponding to the difference images between the new and old target images in the target space, and the difference images between the new and relatively new target data vectorized images in the target space, are extracted to represent image difference information and data difference information with different receptive fields. Multiple feature images with different feature dimensions are fused into multiple fused images to interactively fuse image difference information and data difference information from different receptive fields. Based on the interactively fused image difference information and data difference information from different receptive fields, the method considers the change information of spatial data represented by the relatively new target data vectorized image in the target space compared to the old target image, thus more accurately identifying the change region of the target space at the first target time, making the target change region of the target space at the first target time more realistic and effective.

[0037] Next, the system architecture of the change region identification method will be introduced. See [link / reference] Figure 1 , Figure 1 This is a schematic diagram of a method for identifying changing regions provided in an embodiment of this application. The system includes a computer device 100, which is used to execute the method for identifying changing regions.

[0038] Computer device 100 acquires a first target difference image between a first target image at a first target time and a second target image at a second target time, and a second target difference image between the first target image and a data vectorized image of the target space at a third target time.

[0039] Here, the target space is any space obtained by continuous spatial partitioning. The first target time and the third target time are after the second target time, that is, the second target time is the old target time, the third target time is the relatively new target time, and the first target time is the new target time; therefore, the first target image image of the first target time is the new target image image, the second target image image of the second target time is the old target image image, the data vectorized image of the third target time is the relatively new target data vectorized image, and the data vectorized image of the third target time is the target space data vectorized image of the target space. The first target difference image is the difference image between the first target image image and the second target image image, and the second target difference image is the difference image between the first target image image and the data vectorized image.

[0040] As an example, let the target space be M, the first target time be T1, the first target image be F1, the second target time be T2, the second target image be F2, and the third target time be T3, where T2 < T3 < T1, and the vectorized image of the third target time be F3. Then, computer device 100 can acquire the first target difference image D1 between F1 of M at T1 and F2 at T2, and the second target difference image D2 between F1 and F3 of M at T3. Here, the target space can be geographic space, the first target image is a new target satellite image, and the second target image is an old target satellite image.

[0041] The computer device 100 extracts features from the first target difference image and the second target difference image through the feature extraction layer in the change region recognition model to obtain multiple target feature images.

[0042] The change region identification model comprises a feature extraction layer, a feature fusion layer, and a recognition layer. The feature extraction layer extracts features from the image to obtain a feature image. The feature fusion layer fuses features from the feature images to obtain a new feature image. The recognition layer identifies change regions from the new feature image. Each target feature image includes a feature image corresponding to a first target difference image and a feature image corresponding to a second target difference image. In other words, each target feature image includes feature images of the same dimension superimposed on the channels, corresponding to the first target difference image and the second target difference image. Different target feature images correspond to different feature dimensions, and target feature images of different feature dimensions correspond to different receptive fields in the target space. Feature dimension and receptive field are inversely proportional; that is, the smaller the feature dimension, the larger the receptive field, and vice versa.

[0043] As an example, based on the above example, computer device 100 can extract features from D1 and D2 through the feature extraction layer in the change region recognition model to obtain multiple target feature images including C1, C2, ..., C m m is a positive integer, and m≥3.

[0044] Computer device 100 performs feature fusion on multiple target feature images based on different feature dimensions through the feature fusion layer in the change region recognition model to obtain multiple target fused images.

[0045] In this embodiment, at least one of the multiple target fusion images includes target feature images with different feature dimensions after fusion; that is, at least one target fusion image includes target feature images with different receptive fields after fusion. Different target fusion images correspond to different feature dimensions, and target fusion images with different feature dimensions correspond to different receptive fields in the target space.

[0046] As an example, based on the above example, computer device 100 can identify C1, C2, ..., C through the feature fusion layer in the change region identification model. m Feature fusion based on different feature dimensions is performed to obtain a fused image of multiple targets, including P. m-1 P m ...

[0047] Computer device 100 uses the recognition layer in the change region recognition model to perform change region recognition on multiple target fused images, and obtains the target change region in the target space at the first target time.

[0048] Among them, the target change region is the region in the target space that changes during the first target time, as identified by the change region identification model.

[0049] As an example, based on the above example, computer device 100 can identify P through the identification layer in the change region identification model. m-1 P m ...to identify change areas and obtain the target change area of ​​M at T1. The target change area can be a building change area where the geographic space changes at the first target time.

[0050] In other words, on the one hand, a first target difference image is acquired between a first target image at a first target time and a second target image at a second target time, where the first target time is after the second target time, providing image difference information for the target space; on the other hand, a second target difference image is acquired between the first target image and a data vectorized image of the target space at a third target time, where the third target time is after the second target time, providing data difference information for the target space. First, the first and second target difference images are input into the feature extraction layer of the change region recognition model. Feature extraction outputs multiple target feature images, where each target feature image includes a feature image corresponding to the first target difference image and a feature image corresponding to the second target difference image. Different target feature images correspond to different feature dimensions, representing image difference information and data difference information with different receptive fields. Then, the multiple target feature images are input into the feature fusion layer of the change region recognition model. Feature fusion based on different feature dimensions outputs multiple target fused images, where different target fused images correspond to different feature dimensions, interactively fusing image difference information and data difference information with different receptive fields. Multiple target fused images are input into the recognition layer of the change region recognition model. The change region recognition outputs the target space change region at the first target time. Considering the change information of the spatial data vectorized image of the target space compared with the spatial data represented by the second target image, the change region of the target space at the first target time is more accurately identified, making the target space change region at the first target time more realistic and effective.

[0051] It should be noted that the variation area identification method in this application involves artificial intelligence. Artificial intelligence is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results. In other words, artificial intelligence is a comprehensive technology within computer science that attempts to understand the essence of intelligence and produce a new type of intelligent machine that can react in a way similar to human intelligence. Artificial intelligence studies the design principles and implementation methods of various intelligent machines, enabling them to possess perception, reasoning, and decision-making functions.

[0052] Artificial intelligence (AI) is a comprehensive discipline encompassing a wide range of fields, including both hardware and software technologies. Fundamental AI technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, pre-trained model technology, operating / interactive systems, and mechatronics. Pre-trained models, also known as large-scale models or foundational models, can be widely applied to downstream tasks across various AI fields after fine-tuning. AI software technologies primarily include computer vision, speech processing, natural language processing, and machine learning / deep learning. In this application's embodiments, AI technologies mainly involve computer vision and machine learning / deep learning.

[0053] Computer vision (CV) is the science that studies how to enable machines to "see." More specifically, it refers to machine vision, which uses cameras and computers to replace human eyes in tasks such as target recognition, tracking, and measurement, and further performs image processing to create images more suitable for human observation or transmission to instruments for detection. As a scientific discipline, computer vision studies related theories and technologies, attempting to build artificial intelligence systems capable of extracting information from images or multidimensional data. Large model technology has brought significant changes to the development of computer vision technology. Pre-trained models in the vision field, such as Swin-transformer, ViT, V-MOE, and MAE, can be quickly and widely applied to specific downstream tasks after fine-tuning. Computer vision technology typically includes image processing, image recognition, and image semantic understanding.

[0054] Machine learning / deep learning is a multidisciplinary field involving probability theory, statistics, approximation theory, convex analysis, and algorithm complexity theory, among others. It specifically studies how computers can simulate or implement human learning behavior to acquire new knowledge or skills and reorganize existing knowledge structures to continuously improve their performance. Machine learning is the core of artificial intelligence and the fundamental way to endow computers with intelligence; its applications span all areas of artificial intelligence. Machine learning and deep learning typically include techniques such as artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and instruction-based learning. Pre-trained models represent the latest development in deep learning, integrating all of these techniques.

[0055] It should be noted that, in the embodiments of this application, the computer device can be a server or a terminal. The method provided in the embodiments of this application can be executed by the terminal or the server alone, or it can be executed by the terminal and the server in cooperation. Specifically, when the method provided in the embodiments of this application is executed by the terminal or the server alone, its execution method is similar to... Figure 1The corresponding embodiments are similar, mainly replacing the computer device with a terminal or server. Furthermore, when the method provided in this application is executed by a terminal and a server, steps that need to be displayed on the front-end interface can be executed by the terminal, while steps that require background calculations and do not need to be displayed on the front-end interface can be executed by the server.

[0056] The terminal can be a smartphone, tablet, laptop, desktop computer, intelligent voice interaction device, vehicle terminal, extended reality device, or aircraft, but is not limited to these. The server can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services, but is not limited to these. The terminal and server can be connected directly or indirectly through wired or wireless communication, and this application does not impose any restrictions. For example, the terminal and server can be connected through a network, which can be wired or wireless.

[0057] Furthermore, the embodiments of this application can be applied to various scenarios, including but not limited to cloud technology, artificial intelligence, smart transportation, assisted driving, autonomous driving, digital human, virtual human, virtual reality, augmented reality, mixed reality, audio and video, etc.

[0058] Next, the method for identifying change regions provided in the embodiments of this application will be described in detail using a computer device executing the method provided in the embodiments of this application as an example, in conjunction with the accompanying drawings.

[0059] See Figure 2 , Figure 2 A flowchart of a method for identifying change regions provided in this application embodiment, the method including:

[0060] S201: Acquire a first target difference image between a first target image at a first target time and a second target image at a second target time, and a second target difference image between the first target image and the data vectorized image of the target space at a third target time; the first target time and the third target time are after the second target time.

[0061] In related technologies, the difference image between a new target image and an old target image can be used to identify changed regions in the target space. However, when spatial data obtained from other sources changes compared to the spatial data represented by the old target image, the changed regions identified by continuing to use the difference image between the new and old target images will have certain deviations, resulting in the problem that the identification results of changed regions are not accurate and effective enough.

[0062] Based on this, in this embodiment of the application, when identifying changed regions in the target space, not only are new and old target image images of the target space needed, but also vectorized images of relatively new target data obtained from other channels are required. This is to take into account the changes in spatial data represented by the vectorized images of relatively new target data in the target space compared to the old target image images, making the identification results for changed regions in the target space more realistic and effective. Based on this, any space obtained from continuous spatial division is defined as the target space. The first target time, the second target time, and the third target time are all defined. If the first target time is after the second target time, and the third target time is after the second target time, then the first time is the new target time, the second target time is the old target time, and the third target time is the relatively new target time. Correspondingly, the first target image image of the target space at the first target time is the new target image image of the target space, the second target image image of the target space at the second target time is the old target image image of the target space, and the vectorized image of the target space data at the third target time, i.e., the vectorized image of the target space data at the third target time, is the relatively new target data vectorized image of the target space.

[0063] On the one hand, it is necessary to acquire the difference image between the new target image and the old target image in the target space, that is, to acquire the first target difference image between the first target image at the first target time and the second target image at the second target time in the target space; on the other hand, it is necessary to acquire the difference image between the new target image and the vectorized image of the relatively new target data in the target space, that is, to acquire the second target difference image between the first target image and the vectorized image of the target space at the third target time.

[0064] S201 not only acquires the difference image between the new target image and the old target image in the target space, providing image difference information of the target space, but also acquires the difference image between the new target image and the vectorized image of the relatively new target data in the target space, providing data difference information of the target space. This provides basic difference information for subsequent identification of change areas, considering the change information of spatial data represented by the vectorized image of the relatively new target data compared to the old target image, and more accurately identifying the change area of ​​the target space in the first target time, making the target change area of ​​the target space in the first target time more realistic and effective.

[0065] As an example of S201, the target space is M, the first target time is T1, the first target image is F1, the second target time is T2, the second target image is F2, the third target time is T3, T2 < T3 < T1, and the vectorized image of the third target time is F3. A first target difference image D1 can be obtained between F1 of M at T1 and F2 at T2, and a second target difference image D2 can be obtained between F1 and F3 of M at T3. Here, the target space can be geographic space, the first target image is a new target satellite image, and the second target image is an old target satellite image.

[0066] S202: By using the feature extraction layer in the change region recognition model, features are extracted from the first target difference image and the second target difference image to obtain multiple target feature images; each target feature image includes the feature image corresponding to the first target difference image and the feature image corresponding to the second target difference image, and different target feature images correspond to different feature dimensions.

[0067] In this embodiment of the application, after executing the above-described S201 to obtain the first target difference image and the second target difference image, in order to consider the change information of the spatial data of the target space relative to the new target data vectorized image compared with the old target image image through different receptive fields, and to more accurately identify the change area of ​​the target space in the first target time, it is first necessary to extract the feature information of different receptive fields in the first target difference image and the second target difference image, and use different receptive fields to represent image difference information and data difference information. That is, based on the change region recognition model, which includes a feature extraction layer for extracting features from the image to obtain a feature image, a feature fusion layer for fusing features from the feature image to obtain a new feature image, and a change region recognition layer for identifying change regions from the new feature image, the first target difference image and the second target difference image share the feature extraction layer in the change region recognition model. The first target difference image and the second target difference image are input into the feature extraction layer in the change region recognition model for feature extraction, and multiple target feature images are output. Each target feature image includes a feature image corresponding to the first target difference image and a feature image corresponding to the second target difference image. In other words, the first target difference image and the second target difference image are input into the feature extraction layer for feature extraction, and multiple feature images corresponding to the first target difference image and the second target difference image are obtained, representing image difference information and data difference information with different receptive fields. The feature images corresponding to the first target difference image and the second target difference image of the same dimension are superimposed on the channel to output multiple target feature images. Different target feature images correspond to different feature dimensions, and target feature images of different feature dimensions correspond to different receptive fields in the target space.

[0068] By extracting feature information from different receptive fields in the first and second target difference images, and using different receptive fields to represent image difference information and data difference information, multiple target feature images obtained through feature extraction layer have different receptive fields. During feature extraction, a larger feature dimension results in a smaller receptive field for the target feature image, which is then used to identify regional changes in smaller areas. Conversely, a smaller feature dimension results in a larger receptive field for the target feature image, which is then used to identify regional changes in larger areas.

[0069] S202 extracts multiple feature images corresponding to both the difference image between the new target image and the old target image in the target space, and the difference image between the new target image and the relatively new target data vectorized image in the target space, through the feature extraction layer in the change region identification model. The image difference information and data difference information are represented by different receptive fields, so that when the change region is identified in the subsequent process, the spatial data change information represented by the relatively new target data vectorized image in the target space is considered by different receptive fields compared with the old target image, and the change region of the target space in the first target time is identified more accurately. This makes the target change region in the target space in the first target time more realistic and effective in providing feature extraction information from different receptive fields.

[0070] As an example of S202, based on the example of S201 above, feature extraction is performed on D1 and D2 through the feature extraction layer in the change region recognition model to obtain multiple target feature images including C1, C2, ..., C m m is a positive integer, m≥3, where C1, C2, ..., C m Corresponding to different feature dimensions.

[0071] S203: By using the feature fusion layer in the change region recognition model, multiple target feature images are fused based on different feature dimensions to obtain multiple target fused images; different target fused images correspond to different feature dimensions.

[0072] In this embodiment, after performing S202 to extract multiple target feature images with different feature dimensions, in order to more accurately consider the changes in spatial data represented by the new target data vectorized image relative to the old target image image through different receptive fields, and to more accurately identify the change region of the target space at the first target time, it is also necessary to interactively fuse image difference information and data difference information of different receptive fields. That is, multiple target feature images are input into the feature fusion layer of the change region recognition model for feature fusion based on different feature dimensions, realizing the fusion of target feature images with different feature dimensions, and outputting multiple target fused images; at least one target fused image in the multiple target fused images includes target feature images with different feature dimensions after fusion, that is, at least one target fused image includes target feature images with different receptive fields after fusion; different target fused images correspond to different feature dimensions, and target fused images with different feature dimensions correspond to different receptive fields of the target space.

[0073] Target fusion images with different receptive fields can identify regional changes in target space of different sizes; the larger the feature dimension of the target fusion image, the larger the receptive field, and the more regional changes it can identify in larger areas; conversely, the smaller the feature dimension of the target fusion image, the smaller the receptive field, and the more regional changes it can identify in smaller areas.

[0074] The S203 uses a feature fusion layer in the change region identification model to fuse multiple feature images of different feature dimensions into multiple fused images. This allows for the interactive fusion of image difference information and data difference information from different receptive fields, resulting in deep fusion of these information. This enables more accurate identification of the change region in the target space by considering the spatial data changes between the new target data vectorized image and the old target image image in different receptive fields. Consequently, it provides more accurate identification of the change region in the target space at the first target time, making the target change region in the target space at the first target time more realistic and effective in providing identification feature information.

[0075] As an example of S203, based on the example of S202 above, the feature fusion layer in the change region identification model is used to process C1, C2, ..., C1 of different feature dimensions. m Feature fusion based on different feature dimensions is performed to obtain a fused image of multiple targets, including P. m-1 P m ..., where P m-1 P m ... correspond to different feature dimensions.

[0076] S204: By using the recognition layer in the change region recognition model, change region recognition is performed on the fused images of multiple targets to obtain the target change region in the target space at the first target time.

[0077] In this embodiment of the application, after performing the above S203 to obtain multiple target fusion images with different feature dimensions, since the multiple target fusion images represent the image difference information and data difference information of different receptive fields after interactive fusion, the multiple target fusion images are input into the recognition layer of the change region recognition model to perform change region recognition. This allows for a more accurate consideration of the spatial data change information of the target space relative to the new target data vectorized image compared to the old target image image through different receptive fields, so as to more accurately identify the area where the target space changes at the first target time, and output the target change area of ​​the target space at the first target time.

[0078] The S204, through the recognition layer in the change region recognition model, based on the image difference information and data difference information of different receptive fields after interactive fusion, considers the change information of spatial data of the target space relative to the new target data vectorized image compared with the old target image, and more accurately identifies the change region of the target space in the first target time, making the target change region of the target space in the first target time more realistic and effective.

[0079] As an example of S204, based on the examples of S201-S203 above, P is identified through the recognition layer in the change region recognition model. m-1 P m ...to identify change areas and obtain the target change area of ​​M at T1. The target change area can be a building change area where the geographic space changes at the first target time.

[0080] As can be seen from the above technical solution, the method acquires a first target difference image between a first target image at a first target time and a second target image at a second target time, and a second target difference image between the first target image and the data vectorized image of the target space at a third target time, wherein the first target time and the third target time are after the second target time. This method not only acquires the difference image between the new target image and the old target image in the target space, providing image difference information of the target space, but also acquires the difference image between the new target image and the relatively new target data vectorized image in the target space, providing data difference information of the target space. The first target difference image and the second target difference image are input into the feature extraction layer of the change region recognition model, and feature extraction is performed to output multiple target feature images, wherein each target feature image includes the feature image corresponding to the first target difference image and the feature image corresponding to the second target difference image, and different target feature images correspond to different feature dimensions; the multiple target feature images are input into the feature fusion layer of the change region recognition model, and feature fusion based on different feature dimensions is performed to output multiple target fused images, wherein different target fused images correspond to different feature dimensions. This method extracts multiple feature images corresponding to both the difference image between the new target image and the old target image in the target space, and the difference image between the new target image and the relatively new target data vectorized image. It uses different receptive fields to represent image and data difference information, and fuses multiple feature images from different feature dimensions into multiple fused images. This interactive fusion of image and data difference information from different receptive fields results in the input of the multiple fused target images into the recognition layer of the change region recognition model. The change region recognition outputs the target change region in the target space at the first target time. Based on the interactively fused image and data difference information from different receptive fields, this method considers the spatial data change information represented by the relatively new target data vectorized image compared to the old target image, thus more accurately identifying the change region of the target space at the first target time, making the target change region in the target space at the first target time more realistic and effective.

[0081] In the above embodiments, when specifically implementing S201 to obtain the second target difference image between the first target image and the data vectorized image, considering that the first target image and the second target image are multi-channel images and the data vectorized image is a single-channel image, in order to facilitate the difference calculation between the first target image and the data vectorized image to obtain the difference image, it is necessary to expand the channels of the data vectorized image, that is, expand the single channel of the data vectorized image into a multi-channel image to obtain a data extended image. The difference calculation is then performed on the multi-channel first target image and the multi-channel data extended image to obtain the second target difference image between the first target image and the data extended image. Based on this, this application provides a possible implementation method where the first target image and the second target image are multi-channel images and the data vectorized image is a single-channel image. S201 includes the following S2011-S2013 (not shown in the figure).

[0082] S2011: Perform difference calculation on the first target image and the second target image to obtain the first target difference image.

[0083] S2012: Perform channel expansion on the vectorized image to obtain a data-expanded image.

[0084] S2013: Perform differential calculation on the first target image and the data-extended image to obtain the second target differential image.

[0085] Among them, the data-extended image is a multi-channel image.

[0086] The above-mentioned S2011-S2013 expands the single-channel data vectorized image into a multi-channel data expanded image, which can quickly, effectively and accurately calculate the difference image between the multi-channel first target image and the multi-channel data expanded image to obtain the second target difference image, thereby quickly, effectively and accurately obtaining the difference image between the new target image and the relative new target data vectorized image.

[0087] As an example, based on the example of S201 above, S2011-S2013 may include: performing a difference calculation on F1 and F2 to obtain D1; F3 is a grayscale single-channel image, performing channel expansion on F3 to obtain a data-expanded image F4, F4 is a three-primary-color (RGB) three-channel image; performing a difference calculation on F1 and F4 to obtain D2.

[0088] In the above embodiments, the data vectorized image at the third target time is obtained by vectorizing the target space data at the third target time, that is, the target space data is first acquired and then vectorized into a data vectorized image. Based on this, this application provides a possible implementation method, and the steps of obtaining the data vectorized image in S201 and S2012 include the following S1-S2 (not shown in the figure).

[0089] S1: Obtain the target space data at the third target time.

[0090] S2: Vectorize the target space data to obtain a vectorized image.

[0091] Among them, the target spatial data for the third target time is relatively new spatial data obtained from other channels.

[0092] The above S1-S2 vectorizes the target space data at the third target time into a visualized vectorized image of the relative new target data, which facilitates the subsequent direct calculation of data difference using the vectorized image of the relative new target data and the new target image.

[0093] As an example, based on the example of S201 above, S1-S2 may include: obtaining the target spatial data of M in T3 as G1; vectorizing G1 to obtain F3. Here, the target spatial data is geospatial data.

[0094] See Figure 3 , Figure 3 This is a schematic diagram of a data vectorized image provided in an embodiment of this application. The data vectorized image is a two-dimensional image, with black as the background and white as the foreground. Specifically, the data vectorized image is a vectorized image of geospatial data, and white represents buildings.

[0095] In the above embodiment, when S203 inputs multiple target feature images into the feature fusion layer of the change region recognition model to output multiple target fusion images based on different feature dimensions, since multiple target feature images correspond to multiple feature dimensions, in order to quickly and effectively fuse multiple target feature images of different feature dimensions into multiple target fusion images, and interactively fuse image difference information and data difference information of different receptive fields, so as to quickly and effectively identify the target change region of the target space in the first target time, target feature images of some feature dimensions in multiple feature dimensions can be fused, thereby accelerating the recognition speed while ensuring recognition accuracy.

[0096] In a scenario where multiple target feature images include a first feature image, a second feature image, and a third feature image, where the feature dimension of the third feature image is smaller than that of the second feature image, and the feature dimension of the second feature image is smaller than that of the first feature image (i.e., the feature dimensions of the first, second, and third feature images decrease sequentially), a third fusion image with the same feature dimension as the third feature image is first determined using the third feature image. Then, the upsampled third fusion image, which is an upsampled third fusion image with the same feature dimension as the second feature image, is used to increase the feature dimension of the third fusion image to the feature dimension of the second feature image. This allows for the fusion of the upsampled third fusion image and the second feature image, resulting in a second fusion image with the same feature dimension as the second feature image. Based on this, multiple target fusion images can be determined. Therefore, this application provides a possible implementation where multiple target feature images include a first feature image, a second feature image, and a third feature image, where the feature dimension of the third feature image is smaller than that of the second feature image, and the feature dimension of the second feature image is smaller than that of the first feature image. The step of obtaining the data vectorized image in S203 includes the following S2031-S2034 (not shown in the figure).

[0097] S2031: Determine the third fused image based on the third feature image; the third fused image and the third feature image correspond to the same feature dimension.

[0098] S2032: Upsample the third fused image to obtain the upsampled third fused image; the upsampled third fused image and the second feature image have the same feature dimension.

[0099] S2033: Perform feature fusion on the upsampled third fused image and the second feature image to obtain the second fused image; the second fused image and the second feature image correspond to the same feature dimension.

[0100] S2034: Based on the second fused image and the third fused image, determine multiple target fused images.

[0101] The first feature image is obtained by downsampling features from the first target difference image and the second target difference image. The second feature image is obtained by downsampling features from the first feature image. The third feature image is obtained by downsampling features from the second feature image. The third fused image and the third feature image have the same feature dimension. The upsampled third fused image and the second feature image have the same feature dimension. The second fused image and the second feature image have the same feature dimension. That is, the feature dimension of the third fused image is smaller than that of the second fused image, and the receptive field of the third fused image is larger than that of the second fused image.

[0102] The above steps S2031-S2034 involve upsampling the third fused image determined by the third feature image and fusing it with the second feature image to obtain a second fused image. This effectively ensures that the second fused image includes not only the image difference information and data difference information corresponding to the receptive field of the second feature dimension, but also the image difference information and data difference information corresponding to the larger receptive field of the third feature dimension. In other words, it effectively fuses the third feature image of the third feature dimension and the second feature image of the second feature dimension into a second fused image, interactively fusing the image difference information and data difference information of different receptive fields, thus deepening the understanding of image difference information and data difference information of different receptive fields. The second and third fused images utilize the third and second feature images from multiple target feature images to achieve feature fusion of different feature dimensions. That is, feature fusion is performed using target feature images with partial feature dimensions, reducing the number of feature images to achieve feature fusion. This allows for faster and more effective identification of subsequent change regions by considering the spatial changes in the target space relative to the old target image image through different receptive fields. This more accurately identifies the change regions of the target space in the first target time, making the target change regions in the target space in the first target time more realistic and effective in providing identification feature information.

[0103] As an example, based on the example of S203 above, m is 3, that is, C1, C2, ..., C m Including a first feature image C1, a second feature image C2, and a third feature image C3, steps S2031-S2034 may include: determining a third fused image P3 based on C3, where C3 and P3 correspond to the same feature dimension; upsampling P3 to obtain an upsampled P3, where the upsampled P3 and C2 correspond to the same feature dimension; fusing the features of the upsampled P3 and C2 to obtain a second fused image P2, where P2 and C2 correspond to the same feature dimension; and determining P based on P2 and P3. m-1 P m ...including P2 and P3.

[0104] As another example, based on the example of S203 above, m is 5, that is, C1, C2, ..., C mIncluding C1, C2, C3, C4, and C5; S2031-S2034 may include: determining a fused image P5 based on C5, where C5 and P5 correspond to the same feature dimension; upsampling P5 to obtain an upsampled P5, where the upsampled P5 and C4 correspond to the same feature dimension; fusing features between the upsampled P5 and C4 to obtain a fused image P4, where P4 and C4 correspond to the same feature dimension; upsampling P4 to obtain an upsampled P4, where the upsampled P4 and C3 correspond to the same feature dimension; fusing features between the upsampled P4 and C3 to obtain a fused image P3, where P3 and C3 correspond to the same feature dimension; determining P based on P3, P4, and P5. m-1 P m ...including P3, P4 and P5.

[0105] In the above embodiments, when multiple target feature images are input into the feature fusion layer of the change region recognition model in S2031-S2034 to perform feature fusion based on partial feature dimensions and output multiple target fused images, considering that adding target feature images with different feature dimensions for feature fusion can further improve the recognition accuracy of subsequent change regions, the second fused image can be upsampled to have the same feature dimension as the first feature image. This increases the feature dimension of the second fused image to the feature dimension of the first feature image, so that the upsampled second fused image and the first feature image are fused to obtain a first fused image with the same feature dimension as the first feature image. Thus, multiple target fused images can be determined based on the first fused image, the second fused image, and the third fused image. Based on this, this application provides a possible implementation method, which includes the following S3-S4 (not shown in the figure) after S2031-S2033. Correspondingly, S2034 includes S2034a (not shown in the figure).

[0106] S3: Upsample the second fused image to obtain the upsampled second fused image; the upsampled second fused image and the first feature image have the same feature dimension.

[0107] S4: Perform feature fusion on the upsampled second fused image and the first feature image to obtain the first fused image; the first fused image and the first feature image correspond to the same feature dimension.

[0108] S2034a: Based on the first fused image, the second fused image, and the third fused image, determine multiple target fused images.

[0109] In this context, the second fused image and the second feature image have the same feature dimension, the upsampled second fused image and the first feature image have the same feature dimension, and the first fused image and the first feature image have the same feature dimension. That is, the feature dimension of the second fused image is smaller than the feature dimension of the first fused image, and the receptive field of the second fused image is larger than the receptive field of the first fused image.

[0110] The above-mentioned S3-S4 and S2034a upsample the second fused image and perform feature fusion with the first feature image to obtain the first fused image. This further effectively realizes that the first fused image not only includes the image difference information and data difference information of the receptive field corresponding to the first feature dimension, but also includes the image difference information and data difference information of the larger receptive field corresponding to the second feature dimension. That is, it further effectively fuses the second feature image of the second feature dimension and the first feature image of the first feature dimension into the first fused image, so as to further interactively fuse the image difference information and data difference information of different receptive fields, and further enable the deep fusion of the image difference information and data difference information of different receptive fields. The first, second, and third fused images utilize the first, second, and third feature images from multiple target feature images to achieve feature fusion across different feature dimensions. This means that feature fusion is performed using target feature images across all feature dimensions, making the feature fusion more comprehensive and accurate. This allows for more accurate identification of subsequent change regions by considering the spatial changes in the target space relative to the old target image image through different receptive fields. It also enables more accurate identification of the target space's change regions at the first target time, providing more realistic and effective identification feature information for the target space's change regions at the first target time.

[0111] As an example, based on the examples in S2031-S2034 above, where m is 3, S3-S4 and S2034a can be: upsampling P2 to obtain an upsampled P2, where the upsampled P2 and C1 correspond to the same feature dimension; fusing features of the upsampled P2 and C1 to obtain a first fused image P1, where P1 and C1 correspond to the same feature dimension; and determining P based on P1, P2, and P3. m-1 P m ...including P1, P2 and P3.

[0112] As another example, based on the examples S2031-S2034 above, where m is 5, S3-S4 and S2034a can be: upsampling P3 to obtain an upsampled P3, where the upsampled P3 and C2 correspond to the same feature dimension; fusing features of the upsampled P3 and C2 to obtain a fused image P2, where P2 and C2 correspond to the same feature dimension; upsampling P2 to obtain an upsampled P2, where the upsampled P2 and C1 correspond to the same feature dimension; fusing features of the upsampled P2 and C1 to obtain a fused image P1, where P1 and C1 correspond to the same feature dimension; and determining P based on P1, P2, P3, P4, and P5. m-1 P m ...including P1, P2, P3, P4 and P5.

[0113] In the above embodiments, when inputting multiple target feature images into the feature fusion layer of the change region recognition model in S2031-S2034 to perform feature fusion based on partial feature dimensions and output multiple target fusion images, considering that target fusion images with different receptive fields recognize different regions when performing change region recognition, the region change situation of a larger region in the target space can be recognized by adding a target fusion image with a larger receptive field. Since the receptive field and feature dimension are inversely proportional, the third fusion image with the smallest feature dimension can be convolved to obtain a fourth fusion image with a feature dimension smaller than the third fusion image, thereby increasing the receptive field. Thus, by combining the fourth fusion image with a larger receptive field with the second and third fusion images, multiple target fusion images can be determined. This achieves the further addition of a target fusion image with a larger receptive field to recognize the region change situation of a larger region, based on obtaining target fusion images of different dimensions by fusing target feature images with partial feature dimensions. Based on this, this application provides a possible implementation method, which includes the following S5 (not shown in the figure) after S2031-S2033. Correspondingly, S2034 includes S2034b (not shown in the figure).

[0114] S5: Perform convolution processing on the third fused image to obtain the fourth fused image; the feature dimension of the fourth fused image is smaller than the feature dimension of the third fused image.

[0115] S2034b: Based on the second fused image, the third fused image, and the fourth fused image, determine multiple target fused images.

[0116] Among them, the feature dimension of the fourth fused image is smaller than that of the third fused image, that is, the receptive field of the fourth fused image is larger than that of the third fused image.

[0117] The above-mentioned S5 and S2034b obtain a fourth fused image by convolutional processing of the third fused image. This reduces the feature dimension of the fourth fused image compared to the third fused image, thereby increasing the receptive field of the fourth fused image compared to the third fused image. This allows for the identification of regional changes in a larger area of ​​the target space through the fourth fused image with a larger receptive field. Consequently, it provides more accurate information for subsequent identification of changed areas by considering the spatial data changes in the target space relative to the old target image image through more different receptive fields. This more accurately identifies the changed areas of the target space at the first target time, providing more realistic and effective identification feature information for the target space at the first target time.

[0118] As an example, based on the examples S2031-S2034 above, where m is 3, S5 and S2034b can be: performing convolution processing on P3 to obtain the fourth fused image P4, where the feature dimension of P4 is smaller than the feature dimension of P3; determining P based on P2, P3, and P4. m-1 P m ...including P2, P3 and P4.

[0119] As another example, based on the examples S2031-S2034 above, where m is 5, S5 and S2034b can be: performing convolution processing on P5 to obtain a fused image P6, where the feature dimension of P6 is smaller than the feature dimension of P5; determining P based on P3, P4, P5, and P6. m-1 P m ...including P3, P4, P5 and P6.

[0120] In the above embodiments, considering the need to obtain a fused image with a larger receptive field, multiple convolution processes can be performed to continuously increase the receptive field and further improve the recognition accuracy of larger regions. Therefore, after convolving the third fused image to obtain a fourth fused image with an increased receptive field, the fourth fused image can be further convolved to obtain a fifth fused image with a larger receptive field than the fourth fused image. Then, based on the second, third, fourth, and fifth fused images, multiple target fused images are determined.

[0121] As an example, based on the example of S5 above, where m is 3, it can also include: performing convolution processing on P4 to obtain a fifth fused image P5, where the feature dimension of P5 is smaller than the feature dimension of P4; determining P based on P2, P3, P4, and P5. m-1 P m ...including P2, P3, P4 and P5.

[0122] As another example, based on the example of S5 above, where m is 5, it can also include: performing convolution processing on P6 to obtain a fused image P7, where the feature dimension of P7 is smaller than the feature dimension of P6; determining P based on P3, P4, P5, P6, and P7. m-1 P m ...including P3, P4, P5, P6 and P7.

[0123] In the above embodiments, when inputting multiple target feature images into the feature fusion layer of the change region recognition model in S3-S4 and S2034a to perform feature fusion based on all feature dimensions and output multiple target fused images, considering that target feature images with different receptive fields recognize different regions when performing change region recognition, the recognition of regional changes in larger regions of the target space can be achieved by adding target feature images with larger receptive fields. Since the receptive field and feature dimension are inversely proportional, the third fused image with the smallest feature dimension can be convolved to obtain a fourth fused image with a feature dimension smaller than the third fused image, thereby increasing the receptive field. Thus, by combining the fourth fused image with a larger receptive field with the first fused image, the second fused image, and the third fused image, multiple target fused images can be obtained. This achieves the recognition of regional changes in larger regions by further adding fused images with larger receptive fields on the basis of fusing target feature images with all feature dimensions. Based on this, this application provides a possible implementation method, which includes the above-mentioned S5 (not shown in the figure) after S2031-S2033. Correspondingly, S2034a includes the following S2034a1 (not shown in the figure).

[0124] S5: Perform convolution processing on the third fused image to obtain the fourth fused image; the feature dimension of the fourth fused image is smaller than the feature dimension of the third fused image.

[0125] S2034a1: Based on the first fused image, the second fused image, the third fused image, and the fourth fused image, determine multiple target fused images.

[0126] Among them, the feature dimension of the fourth fused image is smaller than that of the third fused image, that is, the receptive field of the fourth fused image is larger than that of the third fused image.

[0127] The above-mentioned S5 and S2034a1 perform convolution processing on the third fused image, thereby reducing the feature dimension of the third fused image after convolution processing. That is, the receptive field of the fourth fused image obtained after convolution processing of the third fused image is increased, so that the fourth fused image with a larger receptive field can be used to identify the regional changes of a larger area of ​​the target space. This allows for faster and more effective identification of the change area in the target space by considering the spatial data changes of the target space relative to the old target image image through more different receptive fields. This makes the target space change area at the first target time more realistic and effective in providing identification feature information.

[0128] As an example, based on the examples S3-S4 and S2034a above, S5 and S2034a1 could be: performing convolution processing on P3 to obtain a fourth fused image P4, where the feature dimension of P4 is smaller than the feature dimension of P3; determining P based on P1, P2, P3, and P4. m-1 P m ...including P1, P2, P3 and P4.

[0129] As another example, based on the examples S3-S4 and S2034a above, S5 and S2034a1 could be: performing convolution processing on P5 to obtain a fused image P6, where the feature dimension of P6 is smaller than the feature dimension of P5; determining P based on P1, P2, P3, P4, P5, and P6. m-1 P m ...including P1, P2, P3, P4, P5 and P6.

[0130] In the above embodiments, considering the need to obtain a fused image with a larger receptive field, multiple convolution processes can be performed to continuously increase the receptive field and further improve the recognition accuracy of larger regions. Therefore, after convolving the third fused image to obtain a fourth fused image with an increased receptive field, the fourth fused image can be further convolved to obtain a fifth fused image with a larger receptive field than the fourth fused image. Then, based on the second, third, fourth, and fifth fused images, multiple target fused images are determined.

[0131] As an example, based on the examples S3-S4 and S2034a above, it may further include: performing convolution processing on P4 to obtain a fifth fused image P5, where the feature dimension of P5 is smaller than the feature dimension of P4; determining P based on P1, P2, P3, P4, and P5. m-1 P m ...including P1, P2, P3, P4 and P5.

[0132] As another example, based on the examples S3-S4 and S2034a above, it may further include: performing convolution processing on P6 to obtain a fused image P7, where the feature dimension of P7 is smaller than the feature dimension of P6; determining P based on P1, P2, P3, P4, P5, P6, and P7. m-1 P m ...including P1, P2, P3, P4, P5, P6 and P7.

[0133] In the above embodiment, when S204 inputs multiple target fused images into the recognition layer of the change region recognition model to perform change region recognition and output the target space target change region at the first target time, considering that multiple target fused images correspond to different feature dimensions, that is, multiple target fused images correspond to different receptive fields, when inputting multiple target fused images into the recognition layer of the change region recognition model to perform change region recognition, one change region of the target space at the first target time may correspond to multiple predicted change boxes. In order to reduce the repeated recognition results of one change region corresponding to multiple predicted change boxes, it is necessary to suppress and filter the multiple predicted change boxes, and retain one predicted change box for one change region. Therefore, inputting multiple target fused images into the recognition layer of the change region recognition model for change region recognition requires not only identifying multiple predicted change boxes, but also identifying the change confidence of each predicted change box. By using the multiple change confidences corresponding to multiple predicted change boxes, and the multiple overlaps between multiple predicted change boxes, multiple predicted change boxes are suppressed and filtered. This ensures that for multiple predicted change boxes with an overlap greater than or equal to a preset overlap, the predicted change box corresponding to the highest change confidence is retained, thus obtaining the target change box. Therefore, the target change region of the target space at the first target time can be determined through the target change box. Based on this, this application provides a possible implementation, where S204 includes the following S2041-S2043 (not shown in the figure).

[0134] S2041: Perform change region identification on the fused image of multiple targets to obtain multiple predicted change boxes and the change confidence of each predicted change box.

[0135] S2042: Based on the multiple change confidence levels corresponding to multiple predicted change boxes and the multiple overlaps between multiple predicted change boxes, suppress and filter multiple predicted change boxes to obtain the target change box.

[0136] S2043: Determine the target change area based on the target change box.

[0137] Among them, the change confidence score is used to reflect the confidence level of the predicted change box; the overlap between two predicted change boxes is used to determine the degree of overlap between the two predicted change boxes, that is, whether the two predicted change boxes correspond to a change region; the target change region is the change region of the identified target space.

[0138] The above steps S2041-S2043 involve identifying change regions in multiple target fused images to obtain multiple predicted change boxes and change confidence scores for each predicted change box. This allows for the suppression and filtering of multiple predicted change boxes based on their multiple change confidence scores and multiple overlaps. Specifically, for multiple predicted change boxes with overlap scores greater than or equal to a preset overlap, the predicted change box with the highest change confidence score is retained. This predicted change box with the highest change confidence score is then identified as the target change box, ensuring the effectiveness of the suppressed multiple predicted change boxes. In this way, the target change box is used to determine the target change region in the target space at the first target time, reducing the repetitive identification results of the target change region.

[0139] As an example, based on the example of S204 above, S2041-S2043 can be: for multiple P m-1 P m ... to identify change regions and obtain multiple predicted change boxes Z = {Z1, ..., Z...} n The confidence level of the change in Z1 is K1, Z n The confidence level of the change is K n n≥2; based on the K1, ..., Z corresponding to Z1 n Corresponding K n Z1, ..., Z n Multiple overlaps between Z1, ..., Z n Suppression screening is performed to obtain the target change box; that is, based on the K1, ..., Z1 corresponding to Z1. n Corresponding K n Determine K = {K1, ... K} n The confidence level for the largest change in} is K. max and K max The corresponding prediction change box is Z. max Calculate ZZ max Each predicted change box in the Z-shape max The degree of overlap between them, for Z max Delete ZZs with an overlap greater than or equal to the preset overlap. max The predicted change box in the middle; continue based on the deleted ZZ max For each of the corresponding multiple change confidence levels, determine the maximum change confidence level and the corresponding predicted change box, and calculate the ZZ after deletion. maxFor each predicted bounding box (excluding the one corresponding to the highest confidence level of change), the overlap between that bounding box and the one corresponding to the highest confidence level of change is calculated. For the predicted bounding box corresponding to the highest confidence level of change, boxes with an overlap greater than or equal to a preset overlap are deleted. max The predicted change box is obtained by analogy with the target change box; the target change area is determined based on the target change box.

[0140] In the above embodiments, when S2041 is specifically implemented to identify change regions of multiple target fused images to obtain multiple predicted change boxes, the change boxes corresponding to the change regions are determined by the prediction center, the prediction box position, and the prediction category. Therefore, change regions can be identified for multiple target fused images to obtain multiple prediction centers and the prediction box position and prediction category of each prediction center. Thus, multiple predicted change boxes can be determined by the multiple prediction centers and the prediction box position and prediction category of each prediction center.

[0141] In summary, see Figure 4 The diagram shown is a structural schematic of a change region recognition model provided in an embodiment of this application. The change region recognition model includes a feature extraction layer, a feature fusion layer, and a recognition layer. First, the first target difference image and the second target difference image are input into the feature extraction layer for feature extraction, outputting multiple target feature images including C1, C2, C3, C4, and C5, with the feature dimensions of C1, C2, C3, C4, and C5 gradually decreasing. The feature extraction layer may include multiple residual units, each of which may include a convolutional layer, a normalization layer, an activation layer, and an identity mapping layer. The bottom convolutional layer can be used to extract basic features such as edge textures from the first and second target difference images. The higher convolutional layers can be used to combine and abstract the extracted basic features such as edge textures. The normalization layer can be used to normalize the extracted features to a normal distribution. The activation layer can be used to perform nonlinear mapping on the extracted features, enhancing the generalization ability of the change region recognition model.

[0142] Then, C1, C2, C3, C4, and C5 are input into the feature fusion layer for feature fusion based on different feature dimensions, outputting multiple target fused images including P3, P4, P5, P6, and P7. That is, P5 is determined based on C5, and C5 and P5 correspond to the same feature dimension. On one hand, P5 is upsampled to obtain upsampled P5, and upsampled P5 and C4 correspond to the same feature dimension. Feature fusion is performed on upsampled P5 and C4 to obtain P4, and P4 and C4 correspond to the same feature dimension. P4 is upsampled to obtain upsampled P4, and upsampled P4 and C3 correspond to the same feature dimension. Feature fusion is performed on upsampled P4 and C3 to obtain P3, and P3 and C3 correspond to the same feature dimension. On the other hand, convolution processing is performed on P5 to obtain P6, and convolution processing is performed on P6 to obtain P7. The feature fusion layer is a pyramid feature fusion layer.

[0143] Finally, P3, P4, P5, P6, and P7 are input into the recognition layer. Change region identification is performed on P3, P4, P5, P6, and P7 to obtain the target change region in the target space at the first target time. Specifically, change region identification is performed on P3, P4, P5, P6, and P7 to obtain multiple predicted change boxes and the change confidence score for each predicted change box. Based on the multiple change confidence scores corresponding to the multiple predicted change boxes and the multiple overlaps between the multiple predicted change boxes, suppression filtering is performed on the multiple predicted change boxes to obtain the target change box. The target change region is then determined based on the target change box. Specifically, changing region identification on P3, P4, P5, P6, and P7 to obtain multiple predicted change boxes means: performing change region identification on P3, P4, P5, P6, and P7 to obtain multiple prediction centers and the prediction box position and prediction category for each prediction center. Thus, multiple predicted change boxes are determined through the multiple prediction centers, the prediction box position, and the prediction category of each prediction center.

[0144] In the above embodiments, the change region identification model in S202-S204 is obtained by training an initial prediction model based on a first historical difference image between a first historical image of a preset space at a first historical time and a second historical image at a second historical time, a second historical difference image between the first historical image and the data vectorized image of the preset space at a third historical time, and the labeled change region of the preset space at the first historical time. In essence, the first and second historical difference images are used as inputs to the initial prediction model, and the output is the identified change region of the preset space at the first historical time, making the identified change region of the preset space at the first historical time close to the labeled change region of the preset space at the first historical time.

[0145] The preset space is any space obtained by continuous spatial division. The first historical time and the third historical time are after the second historical time; that is, the second historical time is the old historical time, the third historical time is the relatively new historical time, and the first historical time is the new historical time. Therefore, the first historical image of the first historical time is the new historical image, the second historical image of the second historical time is the old historical image, the vectorized image of the third historical time is the relatively new historical data vectorized image, and the vectorized image of the third historical time is the vectorized image of the preset space's preset spatial data. The first historical difference image is the difference image between the first historical image and the second historical image, and the second historical difference image is the difference image between the first historical image and the vectorized image of the third historical time. The preset space can be geographic space, the first historical image is the new historical satellite image, and the second historical image is the old historical satellite image.

[0146] The initial prediction model comprises a feature extraction layer, a feature fusion layer, and a recognition layer. The feature extraction layer extracts features from the image to obtain feature images. The feature fusion layer fuses features from the feature images to obtain new feature images. The recognition layer identifies change regions in the new feature images to obtain change regions. The specific training process of the initial prediction model is as follows: First, a first historical difference image is acquired between a first historical image at a first historical time and a second historical image at a second historical time, and a second historical difference image is acquired between the first historical image and a vectorized data image of the preset space at a third historical time. These first and second historical difference images are input into the initial prediction model. The feature fusion layer of the initial prediction model extracts features from the first and second historical difference images to obtain multiple historical feature images. The feature fusion layer in the initial prediction model then fuses these multiple historical feature images based on different feature dimensions to obtain multiple historical fused images. Finally, the recognition layer in the initial prediction model identifies change regions in these multiple historical fused images to obtain the identified change regions of the preset space at the first historical time. The loss function of the initial prediction model is used to calculate the loss between the identified change region at the first historical time and the labeled change region at the first historical time. The model parameters of the initial prediction model are adjusted using this loss until the loss no longer decreases or a preset number of iterations is reached. At this point, the training of the initial prediction model is considered complete, and the trained initial prediction model is used as the change region identification model. Therefore, this application provides a possible implementation method. The training steps of the change region identification model in S202-S204 may, for example, include the following S6-S10 (not shown in the figure).

[0147] S6: Obtain the first historical difference image between the first historical image at the first historical time and the second historical image at the second historical time, and the second historical difference image between the first historical image and the data vectorized image of the preset space at the third historical time; the first historical time and the third historical time are after the second historical time.

[0148] S7: Through the feature extraction layer in the initial prediction model, feature extraction is performed on the first historical difference image and the second historical difference image to obtain multiple historical feature images; each historical feature image includes the feature image corresponding to the first historical difference image and the feature image corresponding to the second historical difference image, and different historical feature images correspond to different feature dimensions.

[0149] S8: Through the feature fusion layer in the initial prediction model, multiple historical feature images are fused based on different feature dimensions to obtain multiple historical fused images; different historical fused images correspond to different feature dimensions.

[0150] S9: Through the recognition layer in the initial prediction model, the change region is identified in multiple historical fused images to obtain the recognized change region in the preset space at the first historical time.

[0151] S10: Based on the difference between the identified change area and the labeled change area in the preset space at the first historical time, and the loss function of the initial prediction model, train the initial prediction model to obtain the change area identification model.

[0152] Each historical feature image includes a feature image corresponding to the first historical difference image and a feature image corresponding to the second historical difference image. That is, each historical feature image includes feature images of the same dimension superimposed on the channels, corresponding to the first historical difference image and the second historical difference image. Different historical feature images correspond to different feature dimensions, and historical feature images of different feature dimensions correspond to different receptive fields in the historical space. Feature dimension and receptive field are inversely proportional; that is, the smaller the feature dimension, the larger the receptive field, and vice versa. At least one of the multiple historical fused images includes fused historical feature images of different feature dimensions, that is, at least one historical fused image includes fused historical feature images of different receptive fields. Different historical fused images correspond to different feature dimensions, and historical fused images of different feature dimensions correspond to different receptive fields in the historical space. The identified change region is the region in the preset space that changed at the first historical time, as identified by the change region identification model.

[0153] S6-S10 calculates the loss between the identified change region and the labeled change region at the first historical time using a loss function. The model parameters of the initial prediction model are adjusted so that the identified change region at the first historical time gradually approaches the labeled change region at the first historical time. The trained change region identification model can improve the accuracy of the change region identification model in identifying target change regions in the target space.

[0154] In the above embodiments, when S9 obtains the identified change region of the preset space in the first historical time by identifying the change region of multiple historical fused images through the recognition layer in the initial prediction model, the change region is determined based on the recognition center, the recognition box position, and the recognition category. Therefore, change region identification can be performed on multiple historical fused images to obtain multiple recognition centers and the recognition box position and recognition category of each recognition center. Thus, the identified change region is determined through the multiple recognition centers and the recognition box position and recognition category of each recognition center. Based on this, this application provides a possible implementation method, and S9 may include the following S91-S92 (not shown in the figure).

[0155] S91: Perform change region identification on multiple historical fused images to obtain multiple identification centers and the location and category of the identification box for each center.

[0156] S92: Determine the recognition variation area based on multiple recognition centers and the recognition frame position and recognition category of each recognition center.

[0157] The recognition center is the center of the recognized change area, and the position of the recognition box is determined based on the recognition center and the distance from the recognition center to the four sides of the recognition box; the recognition category is the change category of the recognized change area.

[0158] The above steps S91-S92 involve identifying change regions from multiple historical fused images to obtain multiple identification centers and the position and category of the identification box for each center. Based on these multiple identification centers and the position and category of the identification box for each center, the change regions can be identified, thus enabling the identification of change regions to be obtained directly from multiple historical fused images.

[0159] In the above embodiments, considering that when identifying changing regions, the changing regions are determined based on multiple identification centers and the position and category of the bounding boxes of each identification center, the loss function of the initial prediction model needs to include a center loss function corresponding to the identification center, a category loss function corresponding to the identification category, and a position loss function corresponding to the bounding box position. Among these, the center loss function can effectively suppress low-quality bounding boxes. By using the center loss function, category loss function, and position loss function, relatively complex calculations can be avoided, thereby accelerating the identification and training speed of the initial prediction model.

[0160] As an example, the category loss function is the cross-entropy loss function, and the location loss function is the L2 least squares loss function.

[0161] See Figure 5 , Figure 5 This is a schematic diagram illustrating a change area identification result provided in an embodiment of this application. The target space is geographic space. Figure 5 (a) is a geospatial image of the second target at the second target time. Figure 5 (b) is the geospatial image of the first target at the first target time. Figure 5 The target change box in (b) identifies the target change area, which is the building change area. The identification category of the building change area can include addition, demolition, and reconstruction. A solid line target change box represents addition, a long dashed line target change box represents demolition, and a short dashed line target change box represents reconstruction. Figure 5 (a) is an older satellite image of the target in geospatial data. Figure 5 (b) is a satellite image of the new target in geospatial space.

[0162] It should be noted that, based on the implementation methods provided in the above aspects, this application can be further combined to provide more implementation methods.

[0163] based on Figure 2 In accordance with the change region identification method provided in the corresponding embodiments, this application also provides a change region identification device, see [link to relevant documentation]. Figure 6 , Figure 6 This is a structural diagram of a change region identification device provided in an embodiment of the present application. The change region identification device 600 includes: an acquisition unit 610, an extraction unit 620, a fusion unit 630, and an identification unit 640.

[0164] The acquisition unit 610 is used to acquire a first target difference image between a first target image at a first target time and a second target image at a second target time, and a second target difference image between the first target image and the data vectorized image of the target space at a third target time; the first target time and the third target time are after the second target time;

[0165] The extraction unit 620 is used to extract features from the first target difference image and the second target difference image through the feature extraction layer in the change region identification model to obtain multiple target feature images; each target feature image includes a feature image corresponding to the first target difference image and a feature image corresponding to the second target difference image, and different target feature images correspond to different feature dimensions;

[0166] The fusion unit 630 is used to perform feature fusion on multiple target feature images based on different feature dimensions through the feature fusion layer in the change region recognition model to obtain multiple target fused images; different target fused images correspond to different feature dimensions.

[0167] The recognition unit 640 is used to identify the change regions of multiple target fused images through the recognition layer in the change region recognition model, and obtain the target change regions in the target space at the first target time.

[0168] In one possible implementation, the acquisition unit 610 is used for:

[0169] The difference between the first target image and the second target image is calculated to obtain the first target difference image;

[0170] Perform channel expansion on the vectorized image to obtain a data-expanded image;

[0171] The difference between the first target image and the data-extended image is calculated to obtain the second target difference image.

[0172] In one possible implementation, the change region identification device 600 further includes: a vectorization unit;

[0173] Vectorized units are used for:

[0174] Acquire target space data at the third target time;

[0175] The target space data is vectorized to obtain a vectorized image.

[0176] In one possible implementation, the multiple target feature images include a first feature image, a second feature image, and a third feature image, wherein the feature dimension of the third feature image is smaller than the feature dimension of the second feature image, and the feature dimension of the second feature image is smaller than the feature dimension of the first feature image; the fusion unit 630 is used for:

[0177] The third fused image is determined based on the third feature image; the third fused image and the third feature image correspond to the same feature dimension.

[0178] The third fused image is upsampled to obtain the upsampled third fused image; the upsampled third fused image and the second feature image have the same feature dimension.

[0179] The upsampled third fused image and the second feature image are fused to obtain the second fused image; the second fused image and the second feature image have the same feature dimension.

[0180] Multiple target fused images are determined based on the second and third fused images.

[0181] In one possible implementation, the fusion unit 630 is also used for:

[0182] The second fused image is upsampled to obtain an upsampled second fused image; the upsampled second fused image and the first feature image have the same feature dimension.

[0183] The upsampled second fused image and the first feature image are fused to obtain the first fused image; the first fused image and the first feature image have the same feature dimension.

[0184] Multiple target fused images are determined based on the first fused image, the second fused image, and the third fused image.

[0185] In one possible implementation, the fusion unit 630 is also used for:

[0186] The third fused image is convolved to obtain the fourth fused image; the feature dimension of the fourth fused image is smaller than the feature dimension of the third fused image.

[0187] Multiple target fused images are determined based on the second fused image, the third fused image, and the fourth fused image.

[0188] In one possible implementation, the fusion unit 630 is also used for:

[0189] The third fused image is convolved to obtain the fourth fused image; the feature dimension of the fourth fused image is smaller than the feature dimension of the third fused image.

[0190] Based on the first fused image, the second fused image, and the third fused image, multiple target fused images are determined, including:

[0191] Multiple target fused images are determined based on the first fused image, the second fused image, the third fused image, and the fourth fused image.

[0192] In one possible implementation, the identification unit 640 is used for:

[0193] The system identifies regions of change in a fused image of multiple targets, and obtains multiple predicted bounding boxes and the change confidence of each predicted bounding box.

[0194] Based on the multiple change confidence scores corresponding to multiple predicted change boxes and the multiple overlaps between multiple predicted change boxes, the multiple predicted change boxes are suppressed and filtered to obtain the target change box.

[0195] Determine the target change area based on the target change box.

[0196] In one possible implementation, the change region identification device 600 further includes: a training unit;

[0197] Training units are used for:

[0198] Acquire a first historical difference image between a first historical image at a first historical time and a second historical image at a second historical time, and a second historical difference image between the first historical image and a data vectorized image of the preset space at a third historical time; the first historical time and the third historical time are after the second historical time;

[0199] The feature extraction layer in the initial prediction model extracts features from the first historical difference image and the second historical difference image to obtain multiple historical feature images. Each historical feature image includes the feature image corresponding to the first historical difference image and the feature image corresponding to the second historical difference image. Different historical feature images correspond to different feature dimensions.

[0200] The feature fusion layer in the initial prediction model performs feature fusion on multiple historical feature images based on different feature dimensions to obtain multiple historical fused images; different historical fused images correspond to different feature dimensions.

[0201] By using the recognition layer in the initial prediction model, the change region is identified in multiple historical fused images to obtain the recognized change region in the first historical time in the preset space.

[0202] Based on the difference between the identified changed region and the labeled changed region in the preset space at the first historical time, and the loss function of the initial prediction model, the initial prediction model is trained to obtain the changed region identification model.

[0203] In one possible implementation, the training unit is used for:

[0204] Change region identification is performed on multiple historical fused images to obtain multiple identification centers and the bounding box position and identification category of each identification center;

[0205] Based on multiple recognition centers and the recognition frame position and recognition category of each recognition center, the recognition variation area is determined.

[0206] In one possible implementation, the loss function of the initial prediction model includes a category loss function, a location loss function, and a center loss function.

[0207] As can be seen from the above technical solution, the change region identification device includes: an acquisition unit, a feature extraction unit, a feature fusion unit, and a change region identification unit. The acquisition unit acquires a first target difference image between a first target image at a first target time and a second target image at a second target time, and a second target difference image between the first target image and the data vectorized image of the target space at a third target time, wherein the first target time and the third target time are after the second target time. The acquisition unit not only acquires the difference image between the new target image and the old target image in the target space, providing image difference information for the target space, but also acquires the difference image between the new target image and the relatively new target data vectorized image, providing data difference information for the target space. The feature extraction unit inputs the first target difference image and the second target difference image into the feature extraction layer of the change region recognition model to extract features and output multiple target feature images. Each target feature image includes a feature image corresponding to the first target difference image and a feature image corresponding to the second target difference image. Different target feature images correspond to different feature dimensions. The feature fusion unit inputs the multiple target feature images into the feature fusion layer of the change region recognition model and outputs multiple target fused images based on different feature dimensions. Different target fused images correspond to different feature dimensions. The feature extraction unit extracts multiple feature images corresponding to both the difference image between the new target image and the old target image in the target space, and the difference image between the new target image and the relatively new target data vectorized image in the target space. Image difference information and data difference information are represented by different receptive fields. The feature fusion unit fuses multiple feature images with different feature dimensions into multiple fused images, interactively fusing image difference information and data difference information with different receptive fields. The change region recognition unit inputs multiple target fused images into the recognition layer of the change region recognition model, and the change region recognition outputs the target change region of the target space at the first target time. Based on the image difference information and data difference information of different receptive fields after interactive fusion, the change region recognition device considers the change information of the spatial data represented by the new target data vectorized image of the target space compared with the old target image image, and more accurately identifies the change region of the target space at the first target time, making the target change region of the target space at the first target time more realistic and effective.

[0208] This application also provides a computer device, which may be a server, see [link to previous document]. Figure 7 , Figure 7This application provides a structural diagram of a server 700. The server 700 can vary significantly due to different configurations or performance characteristics. It may include one or more processors, such as a central processing unit (CPU) 722, and a memory 732, as well as one or more storage media 730 (e.g., one or more mass storage devices) for storing application programs 742 or data 744. The memory 732 and storage media 730 can be temporary or persistent storage. The program stored in the storage media 730 may include one or more modules (not shown in the diagram), each module including a series of instruction operations on the server. Furthermore, the CPU 722 may be configured to communicate with the storage media 730 and execute the series of instruction operations stored in the storage media 730 on the server 700.

[0209] Server 700 may also include one or more power supplies 726, one or more wired or wireless network interfaces 750, one or more input / output interfaces 758, and / or one or more operating systems 741, such as Windows Server. TM Mac OS X TM Unix TM Linux TM FreeBSD TM etc.

[0210] In this embodiment, the central processing unit 722 in the server 700 can execute the methods provided in the various optional implementations of the above embodiments.

[0211] The computer device provided in this application embodiment can also be a terminal, see [link to relevant documentation]. Figure 8 , Figure 8 This is a structural diagram of a terminal provided in an embodiment of this application. Taking a smartphone as an example, the smartphone includes components such as a radio frequency (RF) circuit 810, a memory 820, an input unit 830, a display unit 840, a sensor 850, an audio circuit 860, a wireless Fidelity (WiFi) module 870, a processor 880, and a power supply 890. The input unit 830 may include a touch panel 831 and other input devices 832, the display unit 840 may include a display panel 841, and the audio circuit 860 may include a speaker 861 and a microphone 862. Those skilled in the art will understand that... Figure 8 The smartphone structure shown does not constitute a limitation on smartphones and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0212] The memory 820 can be used to store software programs and modules. The processor 880 executes various functions and data processing of the smartphone by running the software programs and modules stored in the memory 820. The memory 820 may mainly include a program storage area and a data storage area. The program storage area may store the operating system, applications required for at least one function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the smartphone (such as audio data, phonebook, etc.). In addition, the memory 820 may include high-speed random access memory, and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0213] The processor 880 is the control center of the smartphone, connecting various parts of the smartphone via various interfaces and lines. It performs various functions and processes data by running or executing software programs and / or modules stored in the memory 820, and by accessing data stored in the memory 820. Optionally, the processor 880 may include one or more processing units; preferably, the processor 880 may integrate an application processor and a modem processor, wherein the application processor mainly handles the operating system, user interface, and applications, and the modem processor mainly handles wireless communication. It is understood that the modem processor may also not be integrated into the processor 880.

[0214] In this embodiment, the processor 880 in the smartphone can execute the methods provided in the various optional implementations of the above embodiments.

[0215] According to one aspect of this application, a computer-readable storage medium is provided for storing a computer program that, when run on a computer device, causes the computer device to perform the methods provided in various optional implementations of the above embodiments.

[0216] According to one aspect of this application, a computer program product is provided, comprising a computer program stored in a computer-readable storage medium. A processor of a computer device reads the computer program from the computer-readable storage medium and executes the computer program, causing the computer device to perform the methods provided in various optional implementations of the above embodiments.

[0217] The descriptions of the processes or structures corresponding to the above figures each have their own emphasis. For parts of a process or structure that are not described in detail, please refer to the relevant descriptions of other processes or structures.

[0218] The terms "first," "second," etc., used in this application's specification and the accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented, for example, in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0219] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection between apparatuses or units through some interfaces, and may be electrical, mechanical, or other forms.

[0220] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0221] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0222] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing computer programs, such as USB flash drives, portable hard drives, read-only memory (ROM), RAM, magnetic disks, or optical disks.

[0223] In this application embodiment, the terms "module" or "unit" refer to a computer program or part of a computer program that has a predetermined function and works with other related parts to achieve a predetermined goal, and can be implemented wholly or partially using software, hardware (such as processing circuitry or memory), or a combination thereof. Similarly, a processor (or multiple processors or memory) can be used to implement one or more modules or units. Furthermore, each module or unit can be part of an overall module or unit that includes the functionality of that module or unit.

[0224] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for identifying changing regions, characterized in that, The method includes: Acquire a first target difference image between a first target image at a first target time and a second target image at a second target time, and a second target difference image between the first target image and the data vectorized image of the target space at a third target time; the first target time and the third target time are after the second target time; By using the feature extraction layer in the change region recognition model, features are extracted from the first target difference image and the second target difference image to obtain multiple target feature images; each target feature image includes a feature image corresponding to the first target difference image and a feature image corresponding to the second target difference image, and different target feature images correspond to different feature dimensions; The feature fusion layer in the change region identification model performs feature fusion on multiple target feature images based on different feature dimensions to obtain multiple target fused images; different target fused images correspond to different feature dimensions. The change region identification model uses an identification layer to identify change regions in multiple target fused images, thereby obtaining the target change region in the target space at the first target time.

2. The method according to claim 1, characterized in that, The first target image and the second target image are multi-channel images, and the data vectorized image is a single-channel image; the acquisition of the first target difference image between the first target image at the first target time and the second target image at the second target time, and the second target difference image between the first target image and the data vectorized image of the target space at the third target time, includes: The difference between the first target image and the second target image is calculated to obtain the first target difference image; The vectorized data image is subjected to channel expansion to obtain a data-expanded image; The second target difference image is obtained by performing a difference calculation on the first target image and the data-extended image.

3. The method according to claim 1 or 2, characterized in that, The steps for obtaining the data vectorized image include: Obtain the target space data of the target space at the third target time; The target spatial data is vectorized to obtain the vectorized image of the data.

4. The method according to claim 1, characterized in that, The plurality of target feature images include a first feature image, a second feature image, and a third feature image, wherein the feature dimension of the third feature image is smaller than the feature dimension of the second feature image, and the feature dimension of the second feature image is smaller than the feature dimension of the first feature image; the step of performing feature fusion on the plurality of target feature images based on different feature dimensions to obtain a plurality of target fused images includes: A third fused image is determined based on the third feature image; the third fused image and the third feature image correspond to the same feature dimension; The third fused image is upsampled to obtain an upsampled third fused image; the upsampled third fused image and the second feature image have the same feature dimension. The upsampled third fused image and the second feature image are fused to obtain a second fused image; the second fused image and the second feature image correspond to the same feature dimension; Based on the second fused image and the third fused image, a plurality of target fused images are determined.

5. The method according to claim 4, characterized in that, The method further includes: The second fused image is upsampled to obtain an upsampled second fused image; the upsampled second fused image and the first feature image correspond to the same feature dimension. The upsampled second fused image and the first feature image are fused to obtain a first fused image; the first fused image and the first feature image correspond to the same feature dimension; The step of determining multiple target fused images based on the second fused image and the third fused image includes: Based on the first fused image, the second fused image, and the third fused image, a plurality of target fused images are determined.

6. The method according to claim 4, characterized in that, The method further includes: The third fused image is convolved to obtain a fourth fused image; the feature dimension of the fourth fused image is smaller than the feature dimension of the third fused image. The step of determining multiple target fused images based on the second fused image and the third fused image includes: Based on the second fused image, the third fused image, and the fourth fused image, a plurality of target fused images are determined.

7. The method according to claim 5, characterized in that, The method further includes: The third fused image is convolved to obtain a fourth fused image; the feature dimension of the fourth fused image is smaller than the feature dimension of the third fused image. The step of determining a plurality of target fused images based on the first fused image, the second fused image, and the third fused image includes: Based on the first fused image, the second fused image, the third fused image, and the fourth fused image, a plurality of target fused images are determined.

8. The method according to claim 1, characterized in that, The step of identifying change regions in multiple fused target images to obtain the target change regions in the target space at the first target time includes: The changed regions are identified in the multiple target fused images to obtain multiple predicted change boxes and the change confidence of each predicted change box; Based on multiple change confidence levels corresponding to multiple predicted change boxes and multiple overlap levels between multiple predicted change boxes, the multiple predicted change boxes are suppressed and filtered to obtain the target change box; The target change region is determined based on the target change box.

9. The method according to claim 1, characterized in that, The training steps of the change region identification model include: Acquire a first historical difference image between a first historical image at a first historical time and a second historical image at a second historical time, and a second historical difference image between the first historical image and a data vectorized image of the preset space at a third historical time; the first historical time and the third historical time are after the second historical time; The feature extraction layer in the initial prediction model extracts features from the first historical difference image and the second historical difference image to obtain multiple historical feature images. Each historical feature image includes a feature image corresponding to the first historical difference image and a feature image corresponding to the second historical difference image. Different historical feature images correspond to different feature dimensions. The feature fusion layer in the initial prediction model performs feature fusion on multiple historical feature images based on different feature dimensions to obtain multiple historical fused images; different historical fused images correspond to different feature dimensions. The recognition layer in the initial prediction model is used to identify change regions in multiple historical fused images to obtain the identified change regions in the preset space at the first historical time. Based on the difference between the identified change region and the labeled change region in the preset space at the first historical time, and the loss function of the initial prediction model, the initial prediction model is trained to obtain the change region identification model.

10. The method according to claim 9, characterized in that, The step of identifying change regions in multiple historical fused images to obtain the identified change regions in the preset space at the first historical time includes: The changed region is identified in multiple historical fused images to obtain multiple identification centers and the identification box position and identification category of each identification center; The identification variation area is determined based on the multiple identification centers and the identification frame position and identification category of each identification center.

11. The method according to claim 10, characterized in that, The loss function of the initial prediction model includes a category loss function, a location loss function, and a center loss function.

12. A change area identification device, characterized in that, The device includes: an acquisition unit, an extraction unit, a fusion unit, and an identification unit; The acquisition unit is used to acquire a first target difference image between a first target image of the target space at a first target time and a second target image at a second target time, and a second target difference image between the first target image and the data vectorized image of the target space at a third target time; the first target time and the third target time are after the second target time; The extraction unit is used to extract features from the first target difference image and the second target difference image through the feature extraction layer in the change region recognition model to obtain multiple target feature images; each target feature image includes a feature image corresponding to the first target difference image and a feature image corresponding to the second target difference image, and different target feature images correspond to different feature dimensions; The fusion unit is used to perform feature fusion on multiple target feature images based on different feature dimensions through the feature fusion layer in the change region identification model to obtain multiple target fused images; different target fused images correspond to different feature dimensions. The recognition unit is used to perform change region recognition on multiple target fusion images through the recognition layer in the change region recognition model, and obtain the target change region of the target space at the first target time.

13. A computer device, characterized in that, The computer device includes a processor and memory: The memory is used to store computer programs and to transfer the computer programs to the processor; The processor is configured to execute the method according to any one of claims 1-11 according to instructions in the computer program.

14. A computer-readable storage medium, characterized in that, The computer-readable storage medium is used to store a computer program that, when run on a computer device, causes the computer device to perform the method according to any one of claims 1-11.

15. A computer program product, comprising a computer program, characterized in that, When the computer program is run on a computer device, it causes the computer device to perform the method according to any one of claims 1-11.

Images