Building identification method and device, computer device and storage medium
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- 曙光信息产业(河南)有限公司
- Filing Date
- 2022-12-16
- Publication Date
- 2026-06-30
AI Technical Summary
The accuracy of ground feature identification in traditional remote sensing images is low, mainly because the presence of mixed pixels affects the spatial resolution of hyperspectral images, making it difficult to accurately extract ground feature distribution information.
By acquiring the homogeneity and heterogeneity information of each sub-pixel in the initial remote sensing image, and using high-resolution sub-pixel remote sensing images combined with a pre-set positioning model, including backpropagation neural network, long short-term memory network and classification network, feature fusion and classification are performed to identify various land features in the initial remote sensing image.
It improves the accuracy of identifying various land features in remote sensing images, makes full use of the spectral and spatial structure information of sub-pixel remote sensing images, and enhances the precision of land feature identification.
Smart Images
Figure CN115984611B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of remote sensing science and technology, and in particular to a method, apparatus, computer equipment and storage medium for identifying ground features. Background Technology
[0002] Remote sensing images can record the magnitude of electromagnetic waves from various ground features, and can be applied to ground feature identification. Hyperspectral images, as a type of remote sensing image, have strong ground feature identification capabilities by providing high-resolution spectral images. However, due to the complexity of ground feature types and the limitations of the instantaneous field of view of hyperspectral instruments, mixed pixels are commonly present in hyperspectral images. The presence of mixed pixels affects the spatial resolution of hyperspectral images, making it difficult to accurately extract ground feature distribution information from hyperspectral images.
[0003] In traditional techniques, the identification of ground features in remote sensing images involves classifying the ground features in the images based on the abundance values obtained after spectral unmixing of the images, thus obtaining the category information of the ground features in the remote sensing images.
[0004] However, traditional techniques suffer from low accuracy in identifying ground features in remote sensing images. Summary of the Invention
[0005] Therefore, it is necessary to provide a method, apparatus, computer equipment, and storage medium for identifying ground features in remote sensing images that can improve the accuracy of identifying ground features in remote sensing images, in order to address the aforementioned technical problems.
[0006] Firstly, this application provides a method for identifying land features. The method includes:
[0007] Based on the initial remote sensing image, obtain the homogeneity and heterogeneity information of each sub-pixel in the initial remote sensing image;
[0008] Based on the initial remote sensing image, a sub-pixel remote sensing image corresponding to the initial remote sensing image is obtained; wherein, the resolution of the sub-pixel remote sensing image is higher than the resolution of the initial remote sensing image;
[0009] Based on the sub-pixel remote sensing image, the homogeneity information, the heterogeneity information, and the preset positioning model, the identification results of various land features in the initial remote sensing image are obtained.
[0010] In one embodiment, the localization model includes a first backpropagation neural network, a second backpropagation neural network, a long short-term memory network, and a classification network; the step of obtaining the identification results of various land features in the initial remote sensing image based on the sub-pixel remote sensing image, the homogeneity information, the heterogeneity information, and the preset localization model includes:
[0011] The homogeneity information is input into the first backpropagation neural network to obtain homogeneity features;
[0012] The heterogeneity information is input into the second backpropagation neural network to obtain heterogeneous features;
[0013] The sub-pixel remote sensing image is input into the long short-term memory network to obtain the spectral feature information corresponding to the sub-pixel remote sensing image;
[0014] The homogeneity features, heterogeneity features, and spectral features are input into the classification network to obtain the recognition result.
[0015] In one embodiment, the classification network includes a feature fusion layer and a classification layer; the step of inputting the homogeneity features, the heterogeneity features, and the spectral feature information into the classification network to obtain the recognition result includes:
[0016] The feature fusion layer fuses the homogeneous features, the heterogeneous features, and the spectral feature information to obtain the fused features.
[0017] The fused features are classified using the classification layer to obtain the recognition result.
[0018] In one embodiment, obtaining the homogeneity and heterogeneity information of each sub-pixel in the initial remote sensing image based on the initial remote sensing image includes:
[0019] The initial remote sensing image is subjected to spectral decomposition to obtain a fractional map of the initial remote sensing image; the fractional map includes the proportion of each type of land feature in the initial remote sensing image;
[0020] Based on the initial remote sensing image and the fractional map, obtain the homogeneity and heterogeneity information of each sub-pixel in the initial remote sensing image.
[0021] In one embodiment, obtaining the homogeneity and heterogeneity information of each sub-pixel in the initial remote sensing image based on the initial remote sensing image and the fractional map includes:
[0022] Based on spatial correlation theory, the law of universal gravitation, and the fractional map, calculations are performed on each sub-pixel in the initial remote sensing image to obtain the homogeneity information and the heterogeneity information.
[0023] In one embodiment, obtaining the sub-pixel remote sensing image corresponding to the initial remote sensing image based on the initial remote sensing image includes:
[0024] The initial remote sensing image is processed using bicubic interpolation to obtain the sub-pixel remote sensing image.
[0025] Secondly, this application also provides a feature identification device. The device includes:
[0026] The first acquisition module acquires the homogeneity and heterogeneity information of each sub-pixel in the initial remote sensing image based on the initial remote sensing image.
[0027] The second acquisition module is used to acquire a sub-pixel remote sensing image corresponding to the initial remote sensing image based on the initial remote sensing image; wherein the resolution of the sub-pixel remote sensing image is higher than the resolution of the initial remote sensing image;
[0028] A recognition module is used to obtain the recognition results of various land features in the initial remote sensing image based on the sub-pixel remote sensing image, the homogeneity information, the heterogeneity information, and a preset positioning model. Thirdly, this application also provides a computer device. The computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to perform the following steps:
[0029] Based on the initial remote sensing image, obtain the homogeneity and heterogeneity information of each sub-pixel in the initial remote sensing image;
[0030] Based on the initial remote sensing image, a sub-pixel remote sensing image corresponding to the initial remote sensing image is obtained; wherein, the resolution of the sub-pixel remote sensing image is higher than the resolution of the initial remote sensing image;
[0031] Based on the sub-pixel remote sensing image, the homogeneity information, the heterogeneity information, and the preset positioning model, the identification results of various land features in the initial remote sensing image are obtained.
[0032] Fourthly, this application also provides a computer-readable storage medium. The computer-readable storage medium stores a computer program thereon, which, when executed by a processor, performs the following steps:
[0033] Based on the initial remote sensing image, obtain the homogeneity and heterogeneity information of each sub-pixel in the initial remote sensing image;
[0034] Based on the initial remote sensing image, a sub-pixel remote sensing image corresponding to the initial remote sensing image is obtained; wherein, the resolution of the sub-pixel remote sensing image is higher than the resolution of the initial remote sensing image;
[0035] Based on the sub-pixel remote sensing image, the homogeneity information, the heterogeneity information, and the preset positioning model, the identification results of various land features in the initial remote sensing image are obtained.
[0036] Fifthly, this application also provides a computer program product. The computer program product includes a computer program that, when executed by a processor, performs the following steps:
[0037] Based on the initial remote sensing image, obtain the homogeneity and heterogeneity information of each sub-pixel in the initial remote sensing image;
[0038] Based on the initial remote sensing image, a sub-pixel remote sensing image corresponding to the initial remote sensing image is obtained; wherein, the resolution of the sub-pixel remote sensing image is higher than the resolution of the initial remote sensing image;
[0039] Based on the sub-pixel remote sensing image, the homogeneity information, the heterogeneity information, and the preset positioning model, the identification results of various land features in the initial remote sensing image are obtained.
[0040] The aforementioned land cover identification method, apparatus, computer equipment, and storage medium acquire the homogeneity and heterogeneity information of each sub-pixel in the initial remote sensing image, as well as the sub-pixel remote sensing image of the initial remote sensing image. Based on the sub-pixel remote sensing image, the homogeneity and heterogeneity information of each sub-pixel in the initial remote sensing image, and a preset positioning model, the identification results of various land covers in the initial remote sensing image can be obtained. Since the resolution of the sub-pixel remote sensing image is higher than that of the initial remote sensing image, the positioning model can fully learn the spectral information in the sub-pixel remote sensing image. Combined with the homogeneity and heterogeneity information of each pixel in the initial remote sensing image, the various land covers in the initial remote sensing image can be accurately identified, thereby improving the accuracy of the identification results. Attached Figure Description
[0041] Figure 1 This is an application environment diagram of the feature recognition method in one embodiment;
[0042] Figure 2 This is a flowchart illustrating a feature identification method in one embodiment;
[0043] Figure 3 This is a flowchart illustrating the feature identification method in another embodiment;
[0044] Figure 4 This is a schematic diagram of the positioning model in one embodiment;
[0045] Figure 5 This is a flowchart illustrating the feature identification method in another embodiment;
[0046] Figure 6 This is a flowchart illustrating the feature identification method in another embodiment;
[0047] Figure 7 This is a structural block diagram of a feature identification device in one embodiment;
[0048] Figure 8 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation
[0049] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.
[0050] The feature identification method provided in this application can be applied to, for example... Figure 1 The application environment shown. Figure 1 A computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 1 As shown, the computer device includes a processor, memory, and a network interface connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The database stores image data. The network interface communicates with external terminals via a network connection. When executed by the processor, the computer program implements a feature recognition method.
[0051] In one embodiment, such as Figure 2 As shown, a method for identifying ground features is provided, which can be applied to... Figure 1 Taking a computer device as an example, the explanation includes the following steps:
[0052] S201. Based on the initial remote sensing image, obtain the homogeneity and heterogeneity information of each sub-pixel in the initial remote sensing image.
[0053] Remote sensing imagery refers to images that record the magnitude of electromagnetic waves emitted by various land features. It can include information about various land features. Sub-pixels are the smallest imaging units in remote sensing images. The homogeneity of sub-pixels in remote sensing images refers to the spatial attraction between particles of the same land feature type, i.e., the property that particles belong to the same land feature type. The heterogeneity of sub-pixels in remote sensing images is used to describe the correlation between neighboring pixels and sub-pixels of different categories.
[0054] Typically, the homogeneity of sub-pixels largely determines their land cover category. In this embodiment, the homogeneity between sub-pixels and neighboring pixels can be used to obtain their homogeneity information. However, since the homogeneity of different sub-pixel categories may be the same, it affects the determination of the sub-pixel category. Therefore, the heterogeneity describing the correlation between neighboring pixels and sub-pixels of different categories can also be used to determine the sub-pixel category. In this embodiment, by introducing both homogeneity and heterogeneity information of sub-pixels, the attribute relationship between sub-pixels and neighboring pixels is jointly described. This description quantifies the influence of the entire neighboring pixel on the sub-pixel, rather than simply the spatial association of a portion of the land cover in the neighboring pixel on the sub-pixel. This makes the attribute characteristics of the sub-pixel more obvious, facilitating better analysis of the land cover attributes of the sub-pixel and improving positioning accuracy.
[0055] Optionally, in this embodiment, the homogeneity information of sub-pixels in the initial remote sensing image may include the homogeneity magnitude between the sub-pixel and its neighboring pixels, and the heterogeneity information may include the heterogeneity magnitude between the sub-pixel and its neighboring pixels. Optionally, the same sub-pixel may have multiple neighboring pixels. The average homogeneity magnitude between the sub-pixel and its neighboring pixels can be used as the homogeneity information between the sub-pixel and its neighboring pixels, or the maximum homogeneity magnitude between the sub-pixel and its neighboring pixels can be used as the homogeneity information between the sub-pixel and its neighboring pixels. Similarly, the average heterogeneity magnitude between the sub-pixel and its neighboring pixels can be used as the heterogeneity information between the sub-pixel and its neighboring pixels, or the maximum heterogeneity magnitude between the sub-pixel and its neighboring pixels can be used as the heterogeneity information between the sub-pixel and its neighboring pixels.
[0056] In this embodiment, the homogeneity of each sub-pixel in the initial remote sensing image when they belong to different land cover types can be calculated sequentially to obtain the first K column vector, where K is the number of land cover types. The greater the homogeneity of a sub-pixel in a certain category, the more likely it is to belong to that category. The heterogeneity of sub-pixels when they belong to different land cover types is calculated sequentially to obtain the second K column vector.
[0057] It is understood that hyperspectral images are a type of remote sensing image. In this embodiment, the initial remote sensing image can be a hyperspectral image, or it can be other types of remote sensing images.
[0058] S202, Based on the initial remote sensing image, acquire the sub-pixel remote sensing image corresponding to the initial remote sensing image; wherein, the resolution of the sub-pixel remote sensing image is higher than the resolution of the initial remote sensing image.
[0059] Optionally, in this embodiment, the initial remote sensing image can be processed using the distribution of ground features in the initial remote sensing image to obtain a sub-pixel remote sensing image corresponding to the initial remote sensing image. For example, interpolation processing can be performed on the initial remote sensing image to obtain the sub-pixel remote sensing image. It is understood that the resolution of the sub-pixel remote sensing image is higher than that of the initial remote sensing image. The higher the resolution of the remote sensing image, the more details it can display. In other words, the sub-pixel remote sensing image contains richer spectral information.
[0060] S203. Based on the sub-pixel remote sensing image, homogeneous information, heterogeneous information, and the preset positioning model, the identification results of various land features in the initial remote sensing image are obtained.
[0061] In this embodiment, the preset positioning model is a pre-trained model that can identify various land features in the initial remote sensing image and obtain the identification results of various land features in the initial remote sensing image.
[0062] Optionally, in this embodiment, the preset positioning model may include multiple sub-models, each of which may perform different operations. When training the positioning model, the model containing multiple sub-models can be trained, enabling it to identify ground features in the initial remote sensing image based on the homogeneity and heterogeneity information of each sub-pixel in the sub-pixel remote sensing image and the initial remote sensing image. Optionally, in this embodiment, when training the positioning model, 25% of the data from each type of ground feature data can be selected as training samples for training, resulting in a trained positioning model, and the weights of the trained positioning model can be saved.
[0063] In the aforementioned land cover identification method, by acquiring the homogeneity and heterogeneity information of each sub-pixel in the initial remote sensing image, and the sub-pixel remote sensing image of the initial remote sensing image, the identification results of various land covers in the initial remote sensing image can be obtained based on the sub-pixel remote sensing image, the homogeneity information of each sub-pixel in the initial remote sensing image, the heterogeneity information of each sub-pixel in the initial remote sensing image, and a preset positioning model. Since the resolution of the sub-pixel remote sensing image is higher than that of the initial remote sensing image, the positioning model can fully learn the spectral information in the sub-pixel remote sensing image. Combined with the homogeneity and heterogeneity information of each pixel in the initial remote sensing image, the various land covers in the initial remote sensing image can be accurately identified, thereby improving the accuracy of the identification results of various land covers in the initial remote sensing image.
[0064] In the scenario described above, where the identification results of various land features in the initial remote sensing image are obtained based on the sub-pixel remote sensing image, the homogeneity information of each sub-pixel in the initial remote sensing image, the heterogeneity information of each sub-pixel in the initial remote sensing image, and a preset localization model, the aforementioned localization model may include a first backpropagation neural network, a second backpropagation neural network, a long short-term memory network, and a classification network. In one embodiment, such as Figure 3 As shown, S203 includes:
[0065] S301, input the homogeneity information into the first backpropagation neural network to obtain homogeneous features.
[0066] Among them, the back propagation (BP) neural network is a multi-layer feedforward neural network trained according to the error back propagation algorithm. It utilizes gradient search technology to minimize the mean square error between the actual output value and the expected output value of the network. In this embodiment, the homogeneity information of each sub-pixel in the initial remote sensing image is input into the first BP neural network. The first BP neural network processes the homogeneity information of each sub-pixel to obtain the homogeneity features corresponding to each sub-pixel.
[0067] Optionally, in this embodiment, the first BP neural network model may include an input layer, a hidden layer, and an output layer. The input layer may include K neurons, where K is the number of sub-pixels in the initial remote sensing image. The input layer is used to input the homogeneity information of each sub-pixel in the initial remote sensing image. After processing by the hidden layer, high-dimensional homogeneity features are obtained in the output layer.
[0068] S302, input the heterogeneity information into the second backpropagation neural network to obtain heterogeneous features.
[0069] In this embodiment, the heterogeneity information of each sub-pixel in the initial remote sensing image is input into the second BP neural network, and the heterogeneity information of each sub-pixel is processed by the second BP neural network to obtain the heterogeneity features corresponding to each sub-pixel.
[0070] Optionally, in this embodiment, the second BP neural network model may include an input layer, a hidden layer, and an output layer. The input layer may include K neurons, where K is the number of sub-pixels in the initial remote sensing image. The input layer is used to input the heterogeneity information of each sub-pixel in the initial remote sensing image. After processing by the hidden layer, high-dimensional heterogeneity features are obtained in the output layer.
[0071] S303: Input the sub-pixel remote sensing image into the long short-term memory network to obtain the spectral feature information corresponding to the sub-pixel remote sensing image.
[0072] Among them, Long Short-Term Memory (LSTM) is a type of recurrent neural network that is suitable for processing and predicting important events with very long intervals and delays in time series.
[0073] In this embodiment, a dual-layer LSTM network with 128 hidden nodes can be used to fully extract the spectral feature information corresponding to the sub-pixel remote sensing image.
[0074] S304: Input homogeneity features, heterogeneity features, and spectral features into the classification network to obtain the recognition results.
[0075] Optionally, the classification network can be a model with classification capabilities. In this embodiment, a Softmax classifier can be used. The Softmax classifier is a generalized induction of the logistic regression classifier for multiple classifications.
[0076] For example, in this embodiment, such as Figure 4 As shown, the first backpropagation neural network, the second backpropagation neural network, and the long short-term memory network are the three sub-models of the localization model. Furthermore, the homogeneous features, heterogeneous features, and spectral features obtained from the three sub-models can be input into the classification network of the localization model. The classification network further fuses and classifies the spectral features, homogeneous features, and heterogeneous features to obtain the identification results of various land features in the initial remote sensing image.
[0077] Optionally, as an alternative implementation, the classification network of the localization model may include a feature fusion layer and a classification layer. The feature fusion layer can fuse homogeneous features, heterogeneous features, and spectral features to obtain fused features. The classification layer can then be used to classify the fused features, obtaining the identification results of various land features in the initial remote sensing image. Feature fusion involves extracting different feature vectors from the same pattern and optimizing their combination, and can be performed using serial or parallel processing methods.
[0078] Optionally, in this embodiment, Indian Pines data can be used to verify the land cover recognition method. The Indian Pines data contains images with a size of 145×145×200, consisting of 16 land cover categories (excluding the background). By downsampling the data, a low-resolution image is generated. A scaling factor s=2 is set and the image is classified by an SVM classifier. Sub-pixel localization is performed by the SPSAM algorithm, the STHSPM algorithm, and the land cover recognition method of this application, respectively. In the land cover recognition results, the SPSAM algorithm shows severe mutual influence between categories, inaccurate localization of sub-pixels of different analogies, obvious jagged edges, and obvious fusion of closely spaced land cover features. The STHSPM algorithm is affected by the background, resulting in inaccurate sub-pixel localization, obvious jagged edges, and obvious fusion of closely spaced land cover features. In contrast, the localization effect of this application is significantly improved between categories and between categories and the background, removing jagged edges and fusion phenomena, and is almost close to the true land cover classification. When the scaling factors s=2 and s=3, the accuracy and Kappa coefficients of the three methods are shown in the table below. Kappa is an indicator of classification accuracy, and Accuracy is the accuracy. It can be seen that the land cover identification method proposed in this application is superior to the other two methods in both identification accuracy and recognition precision.
[0079]
[0080] In this embodiment, by inputting sub-pixel remote sensing images into a Long Short-Term Memory (LSTM) network, spectral feature information corresponding to the sub-pixel remote sensing images can be obtained. This allows the rich spectral information of the remote sensing images to be integrated into the land cover identification process, fully leveraging the value of the remote sensing images. The use of the LTM network not only considers spectral information but also fully considers the interrelationships between bands, making the acquired spectral feature information more accurate. In addition, by inputting homogeneous information into the first backpropagation neural network and heterogeneous information into the second backpropagation neural network, corresponding homogeneous and heterogeneous features can be obtained. This combines the spectral feature information of the remote sensing images with the homogeneity and heterogeneity of the sub-pixels in the remote sensing images, making full use of spectral and spatial structure information and improving the accuracy of land cover identification.
[0081] In the scenario described above, where homogeneity and heterogeneity information of each sub-pixel in the initial remote sensing image are obtained, in one embodiment, such as Figure 5 As shown, the above S201 includes:
[0082] S401, spectral decomposition of the initial remote sensing image is performed to obtain a fractional map of the initial remote sensing image; the fractional map includes the proportion of each type of land feature in the initial remote sensing image.
[0083] Among them, spectral decomposition is a hybrid spectral decomposition technique used to determine the proportion of different ground cover spectral components or unknown components within the same pixel. The mixing of different ground cover spectral components will change the depth, position, width, area and absorption degree of the band. Hybrid spectral decomposition techniques usually use matrix equations, neural network methods and spectral absorption index techniques to find the proportion of each component spectrum within a given pixel.
[0084] In this embodiment, spectral decomposition is performed using low-resolution initial remote sensing images to obtain fractional maps of various land features in the initial remote sensing images. These fractional maps include the proportion of each type of land feature in the initial remote sensing images.
[0085] S402, Based on the initial remote sensing image and fractional map, obtain the homogeneity and heterogeneity information of each sub-pixel in the initial remote sensing image.
[0086] Optionally, in this embodiment, the homogeneity and heterogeneity information of each sub-pixel in the initial remote sensing image can be calculated separately based on spatial correlation theory, the law of universal gravitation, and fractional maps. According to spatial correlation theory, under the premise that the scale of spatial variables is larger than the pixel scale of the remote sensing image, targets that are spatially closer have more similar attribute values. In remote sensing images, this more precisely refers to the spatial autocorrelation between pixels, that is, within a mixed pixel and between different pixels, sub-pixels that are closer are more likely to belong to the same land cover type than sub-pixels that are relatively farther apart. For example, it is assumed that spatial autocorrelation exists between a sub-pixel and its eight neighboring mixed pixels, while the correlation between mixed pixels that are farther apart can be ignored.
[0087] Furthermore, in this embodiment, the theory of universal gravitation is introduced. According to this theory, any two objects in nature attract each other. The magnitude of the gravitational force is directly proportional to the product of the masses of the two objects and inversely proportional to the square of the distance between them. The law of universal gravitation accurately describes the pattern of mutual attraction between objects. For example, in this embodiment, each pixel and sub-pixel in the initial remote sensing image can be considered as a particle of different masses. The fractional map of the mixed pixels can then be represented as the mass of these particles, and the spatial attraction between sub-pixels and their surrounding neighboring pixels in the mixed pixels can be quantitatively expressed.
[0088] Optionally, in this embodiment, the spatial attraction between particles of the same land cover type is used to describe the homogeneity between particles, that is, the property that particles belong to the same land cover type, and is represented by spatial gravity. The homogeneity between sub-pixels and neighboring pixels can be expressed as:
[0089]
[0090] Where, pm and p n For neighboring pixels, Z(p) m Z(p) represents the percentage content of class Z in the central mixed pixel. n Z(p) represents the percentage content of class Z in the central mixed pixel, s is the scaling factor, and Z(p) represents the percentage content of class Z. n R represents the percentage of class Z in the neighborhood pixels. in For sub-pixels and neighboring pixels p m The center distance, Z(w) in ) is a sub-pixel p n The degree of homogeneity between sub-pixels and adjacent mixed pixels belonging to Class Z land cover. To qualitatively describe the homogeneity between sub-pixels and mixed pixels, equation (1) is simplified to obtain:
[0091]
[0092] For example, in this embodiment, when a sub-pixel has 8 neighboring pixels, the homogeneity attribute of the sub-pixel as a Z-type land cover under the influence of all neighboring pixels can be expressed as: Among them, Z(w i ) represents the homogeneous attribute of this sub-pixel.
[0093] Optionally, in this embodiment, the homogeneity of sub-pixels is somewhat accidental. Even with different score maps, sub-pixels of different categories may still exhibit the same homogeneity, affecting the determination of sub-pixel categories. Therefore, heterogeneity is proposed to describe the correlation between neighboring pixels and sub-pixels when they belong to different categories. The heterogeneity between sub-pixels and neighboring pixels can be expressed as:
[0094]
[0095] Among them, Z * (w in This indicates the correlation between the sub-pixel and other land cover categories in adjacent mixed pixels when the sub-pixel is a Z-class land cover.
[0096] For example, in this embodiment, when a sub-pixel has 8 neighboring pixels, the heterogeneous attribute of the sub-pixel as a Z-type land feature under the influence of all neighboring pixels can be expressed as: Among them, Z(w i ) represents the heterogeneous attribute of this sub-pixel.
[0097] In this embodiment, the initial remote sensing image is spectrally decomposed to obtain a fractional map of the initial remote sensing image. Using spatial correlation theory, the law of universal gravitation, and the fractional map, the homogeneity and heterogeneity of sub-pixels are introduced to jointly describe the attribute relationship between sub-pixels and neighboring pixels. The influence of neighboring pixels on sub-pixels is quantified, making the attribute characteristics of sub-pixels more obvious, which facilitates better analysis of the attribute characteristics of sub-pixels and obtains the accuracy of ground feature identification.
[0098] In the scenario described above, where a sub-pixel remote sensing image is obtained from an initial remote sensing image, in one embodiment, S202 includes: processing the initial remote sensing image using bicubic interpolation to obtain the sub-pixel remote sensing image.
[0099] Bicubic interpolation is the most commonly used interpolation method in two-dimensional space. It is a method used to "interpolate" or increase the number / density of "pixels" in an image. By using interpolation techniques to increase graphic data, the resolution of the image can be increased when output in other forms.
[0100] In this embodiment, bicubic interpolation is used to process the initial remote sensing image, which can increase the resolution of the remote sensing image and obtain a sub-pixel remote sensing image. For example, in this embodiment, taking the initial remote sensing image as a hyperspectral image, bicubic interpolation can be used to process the initial low-resolution hyperspectral image to obtain a high-resolution hyperspectral image.
[0101] In this embodiment, bicubic interpolation is used to process the initial remote sensing image, which can preserve better detail quality and obtain sub-pixel remote sensing images with smoother image edges, thereby obtaining more accurate ground feature identification results.
[0102] The following describes an embodiment of this disclosure using a specific ground feature recognition scenario, such as... Figure 6 As shown, the method includes the following steps:
[0103] S1. Perform spectral decomposition on the initial remote sensing image to obtain a fractional map of the initial remote sensing image; the fractional map includes the proportion of each type of land feature in the initial remote sensing image.
[0104] S2, based on spatial correlation theory, the law of universal gravitation, and fractional maps, calculates each sub-pixel in the initial remote sensing image to obtain homogeneous information and heterogeneous information.
[0105] S3. The initial remote sensing image is processed using bicubic interpolation to obtain a sub-pixel remote sensing image; the resolution of the sub-pixel remote sensing image is higher than that of the initial remote sensing image.
[0106] S4. Input the homogeneous information into the first backpropagation neural network of the preset positioning model to obtain homogeneous features; input the heterogeneous information into the second backpropagation neural network of the preset positioning model to obtain heterogeneous features.
[0107] S5. Input the sub-pixel remote sensing image into the long short-term memory network of the preset positioning model to obtain the spectral feature information corresponding to the sub-pixel remote sensing image.
[0108] S6, through the feature fusion layer of the preset positioning model, the homogeneous features, heterogeneous features and spectral features are fused to obtain the fused features;
[0109] S7 uses the classification layer of the preset positioning model to classify the fused features and obtain the identification results of ground objects in the initial remote sensing image.
[0110] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.
[0111] Based on the same inventive concept, this application also provides a feature identification device for implementing the feature identification method described above. The solution provided by this device is similar to the solution described in the above method; therefore, the specific limitations in one or more feature identification device embodiments provided below can be found in the limitations of the feature identification method described above, and will not be repeated here.
[0112] In one embodiment, such as Figure 7 As shown, a ground feature identification device is provided, comprising: a first acquisition module, a second acquisition module, and an identification module, wherein:
[0113] The first acquisition module acquires the homogeneity and heterogeneity information of each sub-pixel in the initial remote sensing image based on the initial remote sensing image.
[0114] The second acquisition module is used to acquire sub-pixel remote sensing images corresponding to the initial remote sensing images based on the initial remote sensing images; wherein the resolution of the sub-pixel remote sensing images is higher than the resolution of the initial remote sensing images.
[0115] The identification module is used to obtain the identification results of various land features in the initial remote sensing image based on the sub-pixel remote sensing image, various homogeneous information, various heterogeneous information and the preset positioning model.
[0116] The feature identification device provided in this embodiment can execute the above method embodiment, and its implementation principle and technical effect are similar, so it will not be described again here.
[0117] In one embodiment, the identification module includes: a first acquisition unit, a second acquisition unit, a third acquisition unit, and a fourth acquisition unit, wherein:
[0118] The first acquisition unit is used to input the homogeneous information into the first backpropagation neural network to obtain homogeneous features.
[0119] The second acquisition unit is used to input the heterogeneous information into the second backpropagation neural network to obtain heterogeneous features.
[0120] The third acquisition unit inputs the sub-pixel remote sensing image into the long short-term memory network to obtain the spectral feature information corresponding to the sub-pixel remote sensing image.
[0121] The fourth acquisition unit inputs homogeneity features, heterogeneity features, and spectral feature information into the classification network to obtain the recognition results.
[0122] The feature identification device provided in this embodiment can execute the above method embodiment, and its implementation principle and technical effect are similar, so it will not be described again here.
[0123] In one embodiment, the fourth acquisition unit is used to fuse homogeneous features, heterogeneous features and spectral feature information through a feature fusion layer to obtain fused features; and to classify the fused features using a classification layer to obtain recognition results.
[0124] The feature identification device provided in this embodiment can execute the above method embodiment, and its implementation principle and technical effect are similar, so it will not be described again here.
[0125] In one embodiment, the first acquisition module includes: a fifth acquisition unit and a sixth acquisition unit, wherein:
[0126] The fifth acquisition unit is used to perform spectral decomposition on the initial remote sensing image to obtain a fractional map of the initial remote sensing image; the fractional map includes the proportion of various land features in the initial remote sensing image.
[0127] The sixth acquisition unit is used to acquire the homogeneity and heterogeneity information of each sub-pixel in the initial remote sensing image based on the initial remote sensing image and the fractional map.
[0128] The feature identification device provided in this embodiment can execute the above method embodiment, and its implementation principle and technical effect are similar, so it will not be described again here.
[0129] In one embodiment, the sixth acquisition unit is used to calculate each sub-pixel in the initial remote sensing image according to the spatial correlation theory, the law of universal gravitation, and the fractional map, to obtain each homogeneous information and each heterogeneous information.
[0130] The feature identification device provided in this embodiment can execute the above method embodiment, and its implementation principle and technical effect are similar, so it will not be described again here.
[0131] In one embodiment, the second acquisition module includes a seventh acquisition unit, wherein:
[0132] The seventh acquisition unit is used to process the initial remote sensing image using bicubic interpolation to obtain sub-pixel remote sensing images.
[0133] The feature identification device provided in this embodiment can execute the above method embodiment, and its implementation principle and technical effect are similar, so it will not be described again here.
[0134] Each module in the aforementioned feature identification device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.
[0135] In one embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 8 As shown, the computer device includes a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, NFC (Near Field Communication), or other technologies. When executed by the processor, the computer program implements a method for identifying geographical features. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad mounted on the computer device's casing, or an external keyboard, touchpad, or mouse.
[0136] Those skilled in the art will understand that Figure 8 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.
[0137] In one embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:
[0138] Based on the initial remote sensing image, obtain the homogeneity and heterogeneity information of each sub-pixel in the initial remote sensing image;
[0139] Based on the initial remote sensing image, obtain the sub-pixel remote sensing image corresponding to the initial remote sensing image; wherein, the resolution of the sub-pixel remote sensing image is higher than the resolution of the initial remote sensing image;
[0140] Based on sub-pixel remote sensing images, homogeneous information, heterogeneous information, and a pre-set positioning model, the identification results of various land features in the initial remote sensing images are obtained.
[0141] In one embodiment, the processor, when executing a computer program, also performs the following steps:
[0142] The homogeneity information is input into the first backpropagation neural network to obtain homogeneous features;
[0143] The heterogeneity information is input into the second backpropagation neural network to obtain the heterogeneous features;
[0144] Sub-pixel remote sensing images are input into a long short-term memory network to obtain the spectral feature information corresponding to the sub-pixel remote sensing images;
[0145] Homogeneity features, heterogeneity features, and spectral features are input into the classification network to obtain the recognition results.
[0146] In one embodiment, the processor, when executing a computer program, also performs the following steps:
[0147] The feature fusion layer fuses homogeneous features, heterogeneous features, and spectral features to obtain fused features.
[0148] The fused features are classified using a classification layer to obtain the recognition results.
[0149] In one embodiment, the processor, when executing a computer program, also performs the following steps:
[0150] The initial remote sensing image is spectrally decomposed to obtain a fractional map of the initial remote sensing image; the fractional map includes the proportion of each type of land feature in the initial remote sensing image;
[0151] Based on the initial remote sensing image and the fractional map, obtain the homogeneity and heterogeneity information of each sub-pixel in the initial remote sensing image.
[0152] In one embodiment, the processor, when executing a computer program, also performs the following steps:
[0153] Based on spatial correlation theory, the law of universal gravitation, and fractional maps, calculations are performed on each sub-pixel in the initial remote sensing image to obtain homogeneous and heterogeneous information.
[0154] In one embodiment, the processor, when executing a computer program, also performs the following steps:
[0155] The initial remote sensing image was processed using bicubic interpolation to obtain sub-pixel remote sensing images.
[0156] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor:
[0157] Based on the initial remote sensing image, obtain the homogeneity and heterogeneity information of each sub-pixel in the initial remote sensing image;
[0158] Based on the initial remote sensing image, obtain the sub-pixel remote sensing image corresponding to the initial remote sensing image; wherein, the resolution of the sub-pixel remote sensing image is higher than the resolution of the initial remote sensing image;
[0159] Based on sub-pixel remote sensing images, homogeneous information, heterogeneous information, and a pre-set positioning model, the identification results of various land features in the initial remote sensing images are obtained.
[0160] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:
[0161] The homogeneity information is input into the first backpropagation neural network to obtain homogeneous features;
[0162] The heterogeneity information is input into the second backpropagation neural network to obtain the heterogeneous features;
[0163] Sub-pixel remote sensing images are input into a long short-term memory network to obtain the spectral feature information corresponding to the sub-pixel remote sensing images;
[0164] Homogeneity features, heterogeneity features, and spectral features are input into the classification network to obtain the recognition results.
[0165] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:
[0166] The feature fusion layer fuses homogeneous features, heterogeneous features, and spectral features to obtain fused features.
[0167] The fused features are classified using a classification layer to obtain the recognition results.
[0168] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:
[0169] The initial remote sensing image is spectrally decomposed to obtain a fractional map of the initial remote sensing image; the fractional map includes the proportion of each type of land feature in the initial remote sensing image;
[0170] Based on the initial remote sensing image and the fractional map, obtain the homogeneity and heterogeneity information of each sub-pixel in the initial remote sensing image.
[0171] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:
[0172] Based on spatial correlation theory, the law of universal gravitation, and fractional maps, calculations are performed on each sub-pixel in the initial remote sensing image to obtain homogeneous and heterogeneous information.
[0173] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:
[0174] The initial remote sensing image was processed using bicubic interpolation to obtain sub-pixel remote sensing images.
[0175] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, performs the following steps:
[0176] Based on the initial remote sensing image, obtain the homogeneity and heterogeneity information of each sub-pixel in the initial remote sensing image;
[0177] Based on the initial remote sensing image, obtain the sub-pixel remote sensing image corresponding to the initial remote sensing image; wherein, the resolution of the sub-pixel remote sensing image is higher than the resolution of the initial remote sensing image;
[0178] Based on sub-pixel remote sensing images, homogeneous information, heterogeneous information, and a pre-set positioning model, the identification results of various land features in the initial remote sensing images are obtained.
[0179] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:
[0180] The homogeneity information is input into the first backpropagation neural network to obtain homogeneous features;
[0181] The heterogeneity information is input into the second backpropagation neural network to obtain the heterogeneous features;
[0182] Sub-pixel remote sensing images are input into a long short-term memory network to obtain the spectral feature information corresponding to the sub-pixel remote sensing images;
[0183] Homogeneity features, heterogeneity features, and spectral features are input into the classification network to obtain the recognition results.
[0184] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:
[0185] The feature fusion layer fuses homogeneous features, heterogeneous features, and spectral features to obtain fused features.
[0186] The fused features are classified using a classification layer to obtain the recognition results.
[0187] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:
[0188] The initial remote sensing image is spectrally decomposed to obtain a fractional map of the initial remote sensing image; the fractional map includes the proportion of each type of land feature in the initial remote sensing image;
[0189] Based on the initial remote sensing image and the fractional map, obtain the homogeneity and heterogeneity information of each sub-pixel in the initial remote sensing image.
[0190] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:
[0191] Based on spatial correlation theory, the law of universal gravitation, and fractional maps, calculations are performed on each sub-pixel in the initial remote sensing image to obtain homogeneous and heterogeneous information.
[0192] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:
[0193] The initial remote sensing image was processed using bicubic interpolation to obtain sub-pixel remote sensing images.
[0194] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.
[0195] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.
[0196] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0197] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.
Claims
1. A method of identifying a ground object, characterized by, The method includes: The initial remote sensing image is spectrally decomposed to obtain a fractional map of the initial remote sensing image; the fractional map includes the proportion of each type of land feature in the initial remote sensing image; Based on the spatial correlation theory, the law of universal gravitation, and the fractional map, the homogeneity information and heterogeneity information of each sub-pixel in the initial remote sensing image are calculated respectively. Based on the initial remote sensing image, a sub-pixel remote sensing image corresponding to the initial remote sensing image is obtained; wherein, the resolution of the sub-pixel remote sensing image is higher than the resolution of the initial remote sensing image; Based on the sub-pixel remote sensing image, the homogeneity information, the heterogeneity information, and the preset positioning model, the identification results of various land features in the initial remote sensing image are obtained; The localization model includes a first backpropagation neural network, a second backpropagation neural network, a long short-term memory network, and a classification network; the step of obtaining the identification results of various land features in the initial remote sensing image based on the sub-pixel remote sensing image, the homogeneous information, the heterogeneous information, and the preset localization model includes: The homogeneity information is input into the first backpropagation neural network to obtain homogeneity features; The heterogeneity information is input into the second backpropagation neural network to obtain heterogeneous features; The sub-pixel remote sensing image is input into the long short-term memory network to obtain the spectral feature information corresponding to the sub-pixel remote sensing image; The homogeneity features, heterogeneity features, and spectral features are input into the classification network to obtain the recognition result.
2. The method of claim 1, wherein, The classification network includes a feature fusion layer and a classification layer; the step of inputting the homogeneity features, the heterogeneity features, and the spectral feature information into the classification network to obtain the recognition result includes: The feature fusion layer fuses the homogeneous features, the heterogeneous features, and the spectral feature information to obtain the fused features. The fused features are classified using the classification layer to obtain the recognition result.
3. The method according to claim 1 or 2, characterized in that, The step of obtaining the sub-pixel remote sensing image corresponding to the initial remote sensing image includes: The initial remote sensing image is processed using bicubic interpolation to obtain the sub-pixel remote sensing image.
4. An object recognition apparatus characterized by comprising: The device includes: The first acquisition module performs spectral decomposition on the initial remote sensing image to obtain a fractional map of the initial remote sensing image; the fractional map includes the proportion of various land features in the initial remote sensing image; based on spatial correlation theory, the law of universal gravitation and the fractional map, it calculates the homogeneity information and heterogeneity information of each sub-pixel in the initial remote sensing image respectively. The second acquisition module is used to acquire a sub-pixel remote sensing image corresponding to the initial remote sensing image based on the initial remote sensing image; wherein the resolution of the sub-pixel remote sensing image is higher than the resolution of the initial remote sensing image; The identification module is used to obtain the identification results of various land features in the initial remote sensing image based on the sub-pixel remote sensing image, each of the homogeneous information, each of the heterogeneous information and the preset positioning model; The localization model includes a first backpropagation neural network, a second backpropagation neural network, a long short-term memory network, and a classification network; the step of obtaining the identification results of various land features in the initial remote sensing image based on the sub-pixel remote sensing image, the homogeneous information, the heterogeneous information, and the preset localization model includes: The homogeneity information is input into the first backpropagation neural network to obtain homogeneity features; The heterogeneity information is input into the second backpropagation neural network to obtain heterogeneous features; The sub-pixel remote sensing image is input into the long short-term memory network to obtain the spectral feature information corresponding to the sub-pixel remote sensing image; The homogeneity features, heterogeneity features, and spectral features are input into the classification network to obtain the recognition result. 5.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is configured to perform the method according to any one of claims 1-4 when the computer program is executed by the processor. When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 3.
6. A computer-readable storage medium having stored thereon a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 3.
7. A computer program product comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 3.