Remote sensing image mixed pixel unmixing method and device and electronic equipment

Abstract

This invention belongs to the field of remote sensing image processing technology, and relates to a method, apparatus, and electronic equipment for unmixing mixed pixels in remote sensing images. The method includes acquiring high spatial resolution image data to obtain a land feature map; obtaining coarse-resolution pure pixels from the land feature map; sampling the reflectance corresponding to the distribution of each land category; determining the average and variance of the reflectance; constructing a linear regression model for the mixed pixels; calculating the probability distribution parameters of the coarse-resolution image pixels containing several land categories; determining the conditional expectation of the probability distribution of each land category; and substituting the average and variance of the reflectance into the conditional expectation to obtain an estimated value of the reflectance corresponding to each land category within the coarse-resolution image. This invention is not limited to unmixing a single image; it considers the spectral changes of coarse pixels over time, extracts nonlinear trends from the elements within the mixed pixels, and provides readily available data for subsequent STF models.

Description

Technical Field

[0001] This invention belongs to the field of remote sensing image processing technology, and more specifically, relates to a method, apparatus and electronic equipment for unmixing mixed pixels of remote sensing images based on Gaussian models. Background Technology

[0002] Multi-temporal remote sensing imagery is a method of acquiring images of the Earth's surface at different times, and it is widely used in various fields such as environmental monitoring, agriculture, forestry, and urban planning. This technology can capture the dynamic processes of changes in the Earth's surface over time, such as crop growth cycles, urban expansion, and environmental changes before and after natural disasters. By analyzing these images, changes in various environmental and socio-economic activities can be effectively monitored and predicted, which is of great value for resource management, disaster prevention, and response.

[0003] While multi-temporal remote sensing imagery offers significant advantages, current remote sensing technology faces a key challenge: existing remote sensing sensors typically cannot simultaneously possess high temporal and high spatial resolution. Although temporally continuous image data (high temporal resolution) can be acquired, the level of detail in these images (spatial resolution) is limited. Conversely, some sensors with high spatial resolution cannot acquire images frequently enough, limiting their temporal resolution. This limitation makes it difficult to obtain detailed and continuous information on surface changes for tasks such as environmental monitoring, disaster assessment, and resource management. Remote sensing satellite sensors can produce images with high spatial resolution, but due to their long revisit periods (e.g., 16 days for Landsat, 26 days for SPOT5), they can only provide temporally sparse image sequences. This is a trade-off between pixel range and detector scan width; currently, almost no remote sensing equipment can provide temporally dense (e.g., a one-day revisit period) and simultaneously high spatial resolution imagery.

[0004] Current spatiotemporal fusion techniques for remote sensing images, namely the fusion of high temporal-variable resolution images and high spatial resolution images, produce a series of temporally dense and spatially detailed images. Existing spatiotemporal fusion methods can be divided into five categories: weight function-based methods, demixing-based methods, learning-based methods, Bayesian methods, and hybrid methods. The prior assumption of the STARFM model (Spatial and Temporal Adaptive Reflectance Fusion Model) is that the reflectance change rate of each endmember is relatively stable. In some cases, such as vegetation phenological changes, this can lead to significant errors. Most decomposition-based methods require land cover maps generated from one or more high-resolution images, but rarely consider phenological changes during the fusion period.

[0005] Learning-based spatiotemporal fusion only utilizes the statistical relationship between coarse-resolution and fine-resolution images, without taking advantage of any physical properties related to ground reflectance. Learning-based remote sensing spatiotemporal image fusion methods have obvious limitations. First, the time span of the spatiotemporal fusion images cannot be too large. Second, their experiments are usually limited to small areas with a limited number of pixels, failing to conduct large-area fusion experiments.

[0006] In summary, existing methods based on hybrid pixel decomposition, such as MMT (multisensor multiresolution technique), do not consider the temporal variation of the spectrum during demixing and are only applicable to hybrid pixel decomposition rather than spatiotemporal image fusion. Existing methods based on hybrid pixel decomposition only perform hybrid pixel decomposition on images from a single time period, without considering the temporal spectral variation of coarse pixels, and without fully considering the regular changes in ground reflectance over time or incorporating geological patterns. Summary of the Invention

[0007] To address the aforementioned technical problems, this invention provides a method, apparatus, and electronic device for demixing mixed pixels in remote sensing images.

[0008] In a first aspect, the present invention provides a method for unmixing hybrid pixels in remote sensing images, comprising:

[0009] Acquire high spatial resolution image data to obtain land feature maps;

[0010] Low spatial resolution image data is acquired, and detection is performed using pixels as windows. Pure pixels of coarse resolution image are obtained based on land feature maps.

[0011] The reflectance corresponding to each land category is obtained by sampling pure pixels from a coarse-resolution image.

[0012] Based on the fact that the reflectance of each land category within the mixed pixel conforms to a Gaussian distribution, the mean and variance of the reflectance corresponding to each land category distribution are determined.

[0013] The pixel values ​​of coarse-resolution images are represented as a linear combination of reflectance corresponding to several land categories, and a linear regression model of mixed pixels is constructed.

[0014] The linear regression model based on mixed pixels shows that the mixed pixels conform to a Gaussian distribution. Based on the area ratio of each land category in the coarse resolution image and the average and variance of the reflectance of each land category, the probability distribution parameters of the pixels in the coarse resolution image containing several land categories are calculated.

[0015] Based on the probability distribution of pixels in a coarse-resolution image containing several land categories, determine the conditional expectation of the probability distribution of each land category.

[0016] By substituting the average and variance of the reflectance corresponding to each land category into the conditional expectation of the probability distribution of each land category, we can obtain the estimated reflectance values ​​for each land category within the coarse-resolution image.

[0017] In a second aspect, the present invention provides a remote sensing image hybrid pixel demixing device, comprising a first acquisition unit, a second acquisition unit, a sampling unit, a first processing unit, a linear regression model construction unit, a parameter determination unit, a second processing unit, a third processing unit, and a fourth processing unit;

[0018] The first acquisition unit is used to acquire high spatial resolution image data to obtain a land feature map;

[0019] The second acquisition unit is used to acquire low spatial resolution image data, perform detection using pixels as windows, and obtain coarse resolution pure pixels based on the land feature map.

[0020] The sampling unit is used to sample pure pixels of a coarse-resolution image to obtain the reflectance corresponding to each land category.

[0021] The first processing unit is used to determine the average and variance of the reflectance corresponding to each land category distribution within the mixed pixel, based on the fact that the reflectance of each land category distribution conforms to a Gaussian distribution.

[0022] The linear regression model building unit is used to characterize the pixel values ​​of coarse-resolution images as a linear combination of reflectance corresponding to several land categories, and to build a linear regression model for mixed pixels.

[0023] The parameter determination unit is used to obtain the Gaussian distribution of the mixed pixels based on the linear regression model of the mixed pixels. It calculates the probability distribution parameters of the pixels in the coarse resolution image containing several land categories based on the area ratio of each land category in the coarse resolution image and the average and variance of the reflectance of each land category.

[0024] The second processing unit is used to determine the conditional expectation of the probability distribution of each land category based on the probability distribution of pixels in a coarse-resolution image containing several land categories.

[0025] The third processing unit is used to substitute the average and variance of the reflectance corresponding to each land category into the conditional expectation of the probability distribution of each land category to obtain the estimated value of the reflectance corresponding to each land category in the coarse resolution image.

[0026] Thirdly, the present invention provides an electronic device, comprising:

[0027] Processor and memory;

[0028] The memory is used to store computer operation instructions;

[0029] The processor is configured to execute the remote sensing image mixing pixel demixing method by invoking the computer operation instructions.

[0030] Based on the above technical solution, the present invention can be further improved as follows.

[0031] Furthermore, cluster maps or land feature maps are used to obtain land categories.

[0032] Furthermore, the pixel values ​​of the coarse-resolution image are represented as a linear combination of reflectance corresponding to several land categories, and a linear regression model for mixed pixels is constructed, including: assuming the pixel value of the coarse-resolution image is y... model Land category number i, the reflectance corresponding to land category i is x. i The area proportions of each land category within the mixed pixel are: The residuals of the linear regression model are ε, the linear combination parameters are N, and the average reflectance of land category i is . The variance of reflectance for land category i is If the reflectance distribution of each land category within a mixed pixel conforms to a Gaussian distribution, then:

[0033]

[0034] The pixel values ​​for a coarse-resolution image are:

[0035] Furthermore, the pixel values ​​of the coarse-resolution image are represented as a linear combination of reflectance corresponding to several land categories, and a linear regression model for mixed pixels is constructed, including: assuming the pixel value of the coarse-resolution image is y... model The area proportions corresponding to the distribution of each land category within the mixed pixel are: The average reflectance of land category i is The variance of reflectance for land category i is The linear combination parameter is N, and the coarse resolution image pixel value is y. model If the mixture follows a Gaussian distribution, then the Gaussian distribution of the mixed pixels is:

[0036]

[0037] Furthermore, based on the probability distribution of pixels in a coarse-resolution image containing several land categories, the conditional expectation of the probability distribution for each land category is determined, including: assuming the average reflectance of land category i is... The reflectance of land category i is x i The coarse-resolution image pixel observation value is y, and the coarse-resolution image pixel observation value y is related to the reflectance x of land category i. iThe covariance matrix is ​​Cov(x) i The average reflectance of land category i is (y), where y is the average reflectance of land category i. The average value of coarse-resolution pixel observations is μ y The variance of reflectance for land category i is The conditional expectation is E[x] i |y], the area proportions of each land category within the mixed pixel are: If y is a coarse-resolution pixel observation, then:

[0038]

[0039] but:

[0040]

[0041] Furthermore, by substituting the average and variance of the reflectance corresponding to each land category into the conditional expectation of the probability distribution of each land category, we obtain the estimated reflectance values ​​for each land category within the coarse-resolution image. This includes: assuming the estimated reflectance value for land category i is... The area proportions of each land category within the mixed pixel are as follows: The variance of reflectance for land category i is The average reflectance of land category i is If the observed pixel value of a coarse-resolution image is y, then the estimated reflectance is:

[0042]

[0043]

[0044] The beneficial effects of this invention are: the application of the SGFFM model proposed in this invention is not limited to demixing a single image. Considering the spectral changes of coarse pixels over time, it can be extended to demixing annual time series of coarse resolution data, which helps to extract physically meaningful nonlinear trends from the elements within the mixed pixels and provides ready-made data for subsequent STF models. Attached Figure Description

[0045] Figure 1 This is a schematic diagram of the remote sensing image hybrid pixel demixing method provided in Embodiment 1 of the present invention;

[0046] Figure 2 Box plot of pure pixel NDVI value samples;

[0047] Figure 3 Simulation graph of Gaussian distribution sampling curve for NDVI values;

[0048] Figure 4A schematic diagram illustrating the principle of pixel unmixing in remote sensing images;

[0049] Figure 5 This is a schematic diagram of an electronic device.

[0050] Icons: 30 - Electronic device; 310 - Processor; 320 - Bus; 330 - Memory; 340 - Transceiver. Detailed Implementation

[0051] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0052] Example 1

[0053] As an example, see the attached document. Figure 1 As shown, to solve the above-mentioned technical problems, this embodiment provides a method for demixing hybrid pixels in remote sensing images, including:

[0054] Acquire high spatial resolution image data to obtain land feature maps;

[0055] Low spatial resolution image data is acquired, and detection is performed using pixels as windows. Pure pixels of coarse resolution image are obtained based on land feature maps.

[0056] The reflectance corresponding to each land category is obtained by sampling pure pixels from a coarse-resolution image.

[0057] Based on the fact that the reflectance of each land category within the mixed pixel conforms to a Gaussian distribution, the mean and variance of the reflectance corresponding to each land category distribution are determined.

[0058] The pixel values ​​of coarse-resolution images are represented as a linear combination of reflectance corresponding to several land categories, and a linear regression model of mixed pixels is constructed.

[0059] The linear regression model based on mixed pixels shows that the mixed pixels conform to a Gaussian distribution. Based on the area ratio of each land category in the coarse resolution image and the average and variance of the reflectance of each land category, the probability distribution parameters of the pixels in the coarse resolution image containing several land categories are calculated.

[0060] Based on the probability distribution of pixels in a coarse-resolution image containing several land categories, determine the conditional expectation of the probability distribution of each land category.

[0061] By substituting the average and variance of the reflectance corresponding to each land category into the conditional expectation of the probability distribution of each land category, we can obtain the estimated reflectance values ​​for each land category within the coarse-resolution image.

[0062] This invention uses the SDGFM (Spectral Dynamic Gaussian Fusion Model) method to decompose a coarse-resolution image.

[0063] Optionally, land categories can be obtained using clustering maps or land feature maps.

[0064] In practical applications, firstly, clustering maps or land feature maps are used to generate i land categories. Secondly, the area proportions corresponding to each land category within the mixed pixel are calculated. Sampling is performed on coarse-resolution image pixels of 1.

[0065] Reflectance x of land category i i Sampling is performed on pure pixels of a coarse-resolution image. Since it contains spectral values ​​for only one land cover type, a Gaussian distribution can be used to analyze the reflectance x. i Modeling is performed.

[0066] Optionally, the pixel values ​​of the coarse-resolution image can be represented as a linear combination of reflectance corresponding to several land categories, and a linear regression model of the mixed pixels can be constructed, including: assuming the pixel value of the coarse-resolution image is y... model Land category number i, the reflectance corresponding to land category i is x. i The area proportions of each land category within the mixed pixel are: The residuals of the linear regression model are ε, the linear combination parameters are N, and the average reflectance of land category i is . The variance of reflectance for land category i is If the reflectance distribution of each land category within a mixed pixel conforms to a Gaussian distribution, then:

[0067]

[0068] The pixel values ​​for a coarse-resolution image are:

[0069] The residuals are Gaussian distributed over any given time between pixels y in a coarse-resolution image, and the residuals are ignored.

[0070] coarse resolution image pixel value y model Coarse-resolution image pixel values ​​containing one or more land categories, where the coarse-resolution image pixel values ​​are represented by reflectance x. i A linear combination, therefore the pixel value y of the coarse-resolution image modelIt also conforms to a Gaussian distribution.

[0071] Optionally, the pixel values ​​of the coarse-resolution image can be represented as a linear combination of reflectance corresponding to several land categories, and a linear regression model of the mixed pixels can be constructed, including: assuming the pixel value of the coarse-resolution image is y... model The area proportions corresponding to the distribution of each land category within the mixed pixel are: The average reflectance of land category i is The variance of reflectance for land category i is The linear combination parameter is N, and the coarse resolution image pixel value is y. model If the mixture follows a Gaussian distribution, then the Gaussian distribution of the mixed pixels is:

[0072]

[0073] Due to factors such as observation angle, different aerosol conditions, and terrain, the coarse-resolution pixel observation value y cannot always be equal to y. model Therefore, y can be corrected through the model. model The difference between y and y is used to determine the pixel values ​​y that constitute the coarse-resolution image. model Each component in the matrix reverts to its true value. If reflectance x i A larger variance means that the reflectivity x of class i is higher. i It is more susceptible to environmental effects or sensor observation effects, therefore y model The larger the deviation between y and y, the more it can be attributed to the reflectivity x of class i. i deviation.

[0074] Optionally, based on the probability distribution of pixels in a coarse-resolution image containing several land categories, determine the conditional expectation of the probability distribution for each land category, including: assuming the average reflectance of land category i is... The reflectance of land category i is x i The coarse-resolution image pixel observation value is y, and the coarse-resolution image pixel observation value y is related to the reflectance x of land category i. i The covariance matrix is ​​Coy(x i The average reflectance of land category i is (y), where y is the average reflectance of land category i. The average value of coarse-resolution pixel observations is μ y The variance of reflectance for land category i is The conditional expectation is E[x] i |y], the area proportions of each land category within the mixed pixel are: If y is a coarse-resolution pixel observation, then:

[0075]

[0076] but:

[0077]

[0078] Optionally, the average and variance of the reflectance corresponding to each land category are substituted into the conditional expectation of the probability distribution of each land category to obtain an estimate of the reflectance corresponding to each land category in the coarse-resolution image, including: assuming the estimated reflectance of land category i is... The area proportions of each land category within the mixed pixel are as follows: The variance of reflectance for land category i is The average reflectance of land category i is If the observed pixel value of a coarse-resolution image is y, then the estimated reflectance is:

[0079]

[0080]

[0081] The localized reflectance of land class i at any given time is obtained from image observations at a coarse resolution.

[0082] Directly using the SGFFM model of this invention to decompose a coarse-resolution image yields a decomposition result. However, the SGFFM model's result is at the object level, meaning that pixels of the same land cluster type will share equal reflectance values ​​within a coarse-resolution image pixel. Therefore, for applications requiring detailed spatial distribution of reflectance, pixel-based STF refinement is necessary. The SGFFM model proposed in this invention can be used for unmixing single images and generating high-resolution images with dense temporal phases using long-term coarse-resolution images (such as MODIS images).

[0083] Patterns in spectral curves are influenced by various environmental factors, such as changes in solar radiation angle, surface temperature, and phenology. However, these fluctuations exhibit systematic behavior. The SGFFM model proposed in this invention addresses this issue by utilizing object-level unmixing and a Gaussian model, greatly facilitating prediction. Time series of land type reflectance are obtained by sampling reflectance. Typically, we can obtain time series of reflectance for many land classifications. Taking the NDVI (Normalized Difference Vegetation Index) value of the MOD09Q1 sensor in the MODIS (Moderate-resolution Imaging Spectroradiometer) product as an example, sampling the NDVI value of a land cover type yields many pure pixels in each image, as shown in the attached figure. Figure 2 As shown, the horizontal axis represents time, in days, and the vertical axis represents the normalized vegetation index value.

[0084] Ideally, the annual reflectance of land cover category should be related to the attached... Figure 2 The average values ​​shown show the same trend; however, there are many reasons for the differences in reflectance, and even so, the annual reflectance of land cover categories generally follows the trend of the average value.

[0085] It is assumed that the reflectance values ​​of pure pixels for each land cover category in the image follow a Gaussian distribution, which is a reasonable assumption in most cases. If this is not the case for some land cover situations, the results can be improved by setting more categories in the fine-resolution clustering step. A visualization is attached. Figure 3 As shown, the horizontal axis represents time in days, and the vertical axis represents the NDVI value.

[0086] Data generated by SDGFM can be integrated into other STF models, thus enhancing the results of SDGFM models by applying STF models. The advantage of the SDGFM model lies in its ability to quickly generate unmixed, coarse-grained observation time series. Once this time series is obtained, the original coarse-resolution data in other STF models can be replaced with unmixed data from the SDGFM model. Therefore, the application of the SDGFM model proposed in this invention is not limited to unmixing single images. Considering the spectral variations of coarse pixels over time, it can be extended to unmixing annual time series of coarse-resolution data, helping to extract physically meaningful nonlinear trends from elements within mixed pixels and providing readily available data for subsequent STF models.

[0087] Example 2

[0088] Based on the same principle as the method shown in Embodiment 1 of the present invention, as illustrated in the appendix. Figure 4 As shown, the embodiments of the present invention also provide a remote sensing image hybrid pixel demixing device, including a first acquisition unit, a second acquisition unit, a sampling unit, a first processing unit, a linear regression model construction unit, a parameter determination unit, a second processing unit, a third processing unit and a fourth processing unit;

[0089] The first acquisition unit is used to acquire high spatial resolution image data to obtain a land feature map;

[0090] The second acquisition unit is used to acquire low spatial resolution image data, perform detection using pixels as windows, and obtain coarse resolution pure pixels based on the land feature map.

[0091] The sampling unit is used to sample pure pixels of a coarse-resolution image to obtain the reflectance corresponding to each land category.

[0092] The first processing unit is used to determine the average and variance of the reflectance corresponding to each land category distribution within the mixed pixel, based on the fact that the reflectance of each land category distribution conforms to a Gaussian distribution.

[0093] The linear regression model building unit is used to characterize the pixel values ​​of coarse-resolution images as a linear combination of reflectance corresponding to several land categories, and to build a linear regression model for mixed pixels.

[0094] The parameter determination unit is used to obtain the Gaussian distribution of the mixed pixels based on the linear regression model of the mixed pixels. It calculates the probability distribution parameters of the pixels in the coarse resolution image containing several land categories based on the area ratio of each land category in the coarse resolution image and the average and variance of the reflectance of each land category.

[0095] The second processing unit is used to determine the conditional expectation of the probability distribution of each land category based on the probability distribution of pixels in a coarse-resolution image containing several land categories.

[0096] The third processing unit is used to substitute the average and variance of the reflectance corresponding to each land category into the conditional expectation of the probability distribution of each land category to obtain the estimated value of the reflectance corresponding to each land category in the coarse resolution image.

[0097] Optionally, land categories can be obtained using clustering maps or land feature maps.

[0098] Optionally, the pixel values ​​of the coarse-resolution image can be represented as a linear combination of reflectance corresponding to several land categories, and a linear regression model of the mixed pixels can be constructed, including: assuming the pixel value of the coarse-resolution image is y... model Land category number i, the reflectance corresponding to land category i is x. i The area proportions of each land category within the mixed pixel are: The residuals of the linear regression model are ε, the linear combination parameters are N, and the average reflectance of land category i is . The variance of reflectance for land category i is If the reflectance distribution of each land category within a mixed pixel conforms to a Gaussian distribution, then:

[0099]

[0100] The pixel values ​​for a coarse-resolution image are:

[0101] Optionally, the pixel values ​​of the coarse-resolution image can be represented as a linear combination of reflectance corresponding to several land categories, and a linear regression model of the mixed pixels can be constructed, including: assuming the pixel value of the coarse-resolution image is y... model The area proportions corresponding to the distribution of each land category within the mixed pixel are: The average reflectance of land category i is The variance of reflectance for land category i is The linear combination parameter is N, and the coarse resolution image pixel value is y. model If the mixture follows a Gaussian distribution, then the Gaussian distribution of the mixed pixels is:

[0102]

[0103] Optionally, based on the probability distribution of pixels in a coarse-resolution image containing several land categories, determine the conditional expectation of the probability distribution for each land category, including: assuming the average reflectance of land category i is... The reflectance of land category i is x i The coarse-resolution image pixel observation value is y, and the coarse-resolution image pixel observation value y is related to the reflectance x of land category i. i The covariance matrix is ​​Cov(x) i The average reflectance of land category i is (y), where y is the average reflectance of land category i. The average value of coarse-resolution pixel observations is μ y The variance of reflectance for land category i is The conditional expectation is E[x] i |y], the area proportions of each land category within the mixed pixel are: If y is a coarse-resolution pixel observation, then:

[0104]

[0105] but:

[0106]

[0107] Optionally, the average and variance of the reflectance corresponding to each land category are substituted into the conditional expectation of the probability distribution of each land category to obtain an estimate of the reflectance corresponding to each land category in the coarse-resolution image, including: assuming the estimated reflectance of land category i is... The area proportions of each land category within the mixed pixel are as follows: The variance of reflectance for land category i is The average reflectance of land category i is If the observed pixel value of a coarse-resolution image is y, then the estimated reflectance is:

[0108]

[0109]

[0110] Example 3

[0111] Based on the same principles as the methods shown in the embodiments of the present invention, the embodiments of the present invention also provide an electronic device, as shown in the appendix. Figure 5 As shown, the electronic device may include, but is not limited to: a processor and a memory; the memory for storing computer programs; and the processor for executing the remote sensing image mixing pixel demixing method shown in the embodiments of the present invention by calling the computer program.

[0112] In one alternative embodiment, an electronic device is provided. Figure 5 The illustrated electronic device 30 includes a processor 310 and a memory 330. The processor 310 and the memory 330 are connected, for example, via a bus 320.

[0113] Optionally, the electronic device 30 may further include a transceiver 340, which can be used for data interaction between the electronic device and other electronic devices, such as sending and / or receiving data. It should be noted that in practical applications, the transceiver 340 is not limited to one type, and the structure of the electronic device 30 does not constitute a limitation on the embodiments of the present invention.

[0114] Processor 310 may be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It can implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this invention. Processor 310 may also be a combination that implements computational functions, such as including one or more microprocessor combinations, a combination of a DSP and a microprocessor, etc.

[0115] Bus 320 may include a pathway for transmitting information between the aforementioned components. Bus 320 may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. Bus 320 can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 5 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0116] The memory 330 may be a ROM (Read Only Memory) or other type of static storage device capable of storing static information and instructions, RAM (Random Access Memory) or other type of dynamic storage device capable of storing information and instructions, or an EEPROM (Electrically Erasable Programmable Read Only Memory), CD-ROM (Compact Disc Read Only Memory) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto.

[0117] The memory 330 is used to store application code (computer program) for executing the present invention, and its execution is controlled by the processor 310. The processor 310 is used to execute the application code stored in the memory 330 to implement the content shown in the foregoing method embodiments.

[0118] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for unmixing pixels in remote sensing images, characterized in that, include: Acquire high spatial resolution image data to obtain land feature maps; Low spatial resolution image data is acquired, and detection is performed using pixels as windows. Pure pixels of coarse resolution image are obtained based on land feature maps. The reflectance corresponding to the distribution of each land category is obtained by sampling pure pixels from a coarse-resolution image. Based on the fact that the reflectance of each land category within the mixed pixel conforms to a Gaussian distribution, the mean and variance of the reflectance corresponding to each land category distribution are determined. The pixel values ​​of coarse-resolution images are represented as a linear combination of reflectance corresponding to several land categories, and a linear regression model of mixed pixels is constructed. Based on the linear regression model of mixed pixels, the mixed pixels conform to a Gaussian distribution. According to the area proportions of each land category in the coarse-resolution image and the average and variance of the reflectance corresponding to each land category distribution, the probability distribution of pixels in the coarse-resolution image containing several land categories is calculated, including: assuming the pixel value of the coarse-resolution image is... Land category code is Land categories The corresponding reflectance is The area proportions of each land category within the mixed pixel are: The residuals of the linear regression model are The linear combination parameters are Land categories The average reflectance is Land categories The variance of the reflectivity is If the reflectance distribution of each land category within the mixed pixel conforms to a Gaussian distribution, then: The pixel values ​​for the coarse-resolution image are: Coarse resolution image pixel values If the mixture follows a Gaussian distribution, then the Gaussian distribution of the mixed pixels is: ; Based on the probability distribution of pixels in a coarse-resolution image containing several land categories, determine the conditional expectation of the probability distribution for each land category, including: assuming the observed values ​​of the coarse-resolution image pixels are... coarse resolution image pixel observations With land category reflectivity The covariance matrix is The average value of coarse-resolution pixel observations is Conditional expectation is ,but: ; ,but: ; By substituting the average and variance of the reflectance corresponding to each land category into the conditional expectation of the probability distribution of each land category, we can obtain the estimated reflectance values ​​for each land category within the coarse-resolution image.

2. The remote sensing image pixel unmixing method according to claim 1, characterized in that, Use clustering diagrams or land feature maps to obtain land categories.

3. The remote sensing image pixel unmixing method according to claim 1, characterized in that, By substituting the average and variance of the reflectance corresponding to each land category into the conditional expectation of the probability distribution of each land category, we obtain the estimated values ​​of the reflectance corresponding to each land category in the coarse-resolution image, including: assuming the land category... The estimated reflectance is The area proportions of each land category within the mixed pixel are: Land categories The variance of the reflectivity is Land categories The average reflectance is The observed pixel values ​​of the coarse-resolution image are Then the estimated value of reflectivity is: ； 。 4. A remote sensing image mixing and demixing device, characterized in that, It includes a first acquisition unit, a second acquisition unit, a sampling unit, a first processing unit, a linear regression model construction unit, a parameter determination unit, a second processing unit, a third processing unit, and a fourth processing unit; The first acquisition unit is used to acquire high spatial resolution image data to obtain a land feature map; The second acquisition unit is used to acquire low spatial resolution image data, perform detection using pixels as windows, and obtain coarse resolution pure pixels based on the land feature map. The sampling unit is used to sample pure pixels of a coarse-resolution image to obtain the reflectance corresponding to the distribution of each land category. The first processing unit is used to determine the average and variance of the reflectance corresponding to each land category distribution within the mixed pixel, based on the fact that the reflectance of each land category distribution conforms to a Gaussian distribution. The linear regression model building unit is used to characterize the pixel values ​​of coarse-resolution images as a linear combination of reflectance corresponding to several land categories, and to build a linear regression model for mixed pixels. The parameter determination unit is used to obtain a Gaussian distribution of mixed pixels based on a linear regression model. It calculates the probability distribution of pixels in the coarse-resolution image containing several land categories based on the area proportions of each land category and the average and variance of the reflectance for each land category in the coarse-resolution image. Let the pixel value of the coarse-resolution image be... Land category code is Land categories The corresponding reflectance is The area proportions of each land category within the mixed pixel are: The residuals of the linear regression model are The linear combination parameters are Land categories The average reflectance is Land categories The variance of the reflectivity is If the reflectance distribution of each land category within the mixed pixel conforms to a Gaussian distribution, then: The pixel values ​​for the coarse-resolution image are: Coarse resolution image pixel values If the mixture follows a Gaussian distribution, then the Gaussian distribution of the mixed pixels is: ; The second processing unit is used to determine the conditional expectation of the probability distribution of each land category based on the probability distribution of pixels in a coarse-resolution image containing several land categories, including: assuming the observed value of the coarse-resolution image pixel is... coarse resolution image pixel observations With land category reflectivity The covariance matrix is The average value of coarse-resolution pixel observations is Conditional expectation is ,but: ; ,but: ; The third processing unit is used to substitute the average and variance of the reflectance corresponding to each land category into the conditional expectation of the probability distribution of each land category to obtain the estimated value of the reflectance corresponding to each land category in the coarse resolution image.

5. An electronic device, characterized in that, include: Processor and memory; The memory is used to store computer operation instructions; The processor is configured to execute the remote sensing image mixing pixel demixing method according to any one of claims 1 to 3 by invoking the computer operation instructions.

Images