Remote sensing land cover classification method, device, equipment and storage medium based on transfer learning

Abstract

The present application provides a remote sensing land cover classification method, device, equipment and storage medium under transfer learning. The method includes: obtaining a source period image, a classification result label corresponding to the source period image and multiple target period images; respectively calculating the spectral slope vector corresponding to the source period image and the spectral slope vector corresponding to each target period image; calculating the change intensity of each target period image and the source period image according to the spectral slope vector corresponding to the source period image and the spectral slope vector corresponding to each target period image; determining the initial prediction sample for supervised classification according to the change intensity and the classification result label; purifying the initial prediction sample to obtain a credible sample for supervised classification; according to the credible sample, supervised classification is performed on multiple target period images to obtain a target land cover product; so as to achieve the effect of improving the temporal resolution, classification efficiency and accuracy of remote sensing land cover classification products.

Description

Technical Field

[0001] The present application relates to the technical field of remote sensing information processing, and in particular to a remote sensing land cover classification method, device, equipment and storage medium under transfer learning. Background Art

[0002] Land cover classification is one of the important areas of application of remote sensing technology. It aims to classify each pixel in the remote sensing image into a specific land cover type. Analyzing the classification and changes of watershed land use / cover is the basis for adaptive watershed management. Therefore, developing more efficient remote sensing land cover classification methods has become a promising direction.

[0003] In the existing technology, land cover classification methods can be divided into supervised classification and unsupervised classification. Supervised classification is suitable for situations where prior knowledge is known. The workload of sample annotation is large, but it has high accuracy and a wide range of applications. It is currently the most widely used method. Unsupervised classification does not require prior knowledge and directly performs classification.

[0004] However, the supervised classification method in the existing technical methods requires manual labeling of samples of different categories of land objects, which greatly increases the time and labor costs, while the unsupervised classification method cannot eliminate the interference of the same spectrum but different objects and the same object but different spectra, resulting in reduced accuracy. There are technical problems such as low temporal resolution of remote sensing land cover classification products, low classification efficiency and poor classification accuracy. Summary of the invention

[0005] This application provides a remote sensing land cover classification method, device, equipment and storage medium under transfer learning, so as to achieve the effect of improving the temporal resolution, classification efficiency and accuracy of remote sensing land cover classification products.

[0006] In the first aspect, the present application provides a remote sensing land cover classification method under transfer learning, comprising:

[0007] Obtain source period images, classification result labels corresponding to source period images, and multiple target period images;

[0008] Calculate the spectral slope vector corresponding to the source period image and the spectral slope vector corresponding to each target period image respectively;

[0009] According to the spectral slope vector corresponding to the source period image and the spectral slope vector corresponding to each target period image, the change intensity of each target period image and the source period image is calculated;

[0010] According to the change intensity and the classification result label, the initial prediction sample for supervised classification is determined;

[0011] Purify the initial prediction samples to obtain reliable samples for supervised classification;

[0012] Based on the credible samples, supervised classification is performed on images of multiple target periods to obtain the target land cover products.

[0013] In a possible implementation, the initial prediction sample is purified to obtain a reliable sample for supervised classification, including:

[0014] According to the clustering characteristics of the spatial distribution of the objects, the broken labels in the initial prediction sample are deleted to obtain the first purified sample;

[0015] Performing feature analysis on the first purified sample to retain tags with change intensities within a change intensity range, thereby obtaining a second purified sample;

[0016] The Markov random field model and information entropy theory are used to perform computational processing on the second purified sample to obtain a credible sample for supervised classification.

[0017] In a possible implementation, the Markov random field model and information entropy theory are used to perform computational processing on the second purified sample to obtain a credible sample for supervised classification, including:

[0018] The second purified sample is used as a label to pre-classify the images of multiple target periods using a support vector machine model to obtain a pre-classification posterior probability and a pre-classification result;

[0019] According to the pre-classification results, a Markov random field model is introduced to output the normalized total energy image;

[0020] For the pre-classification posterior probability, the information entropy theory is introduced to calculate the entropy value graph;

[0021] A weighted map is generated according to the normalized total energy image and the entropy map, and a credible area is determined based on the weighted map;

[0022] The trusted region is spatially overlapped with the second purified sample to obtain a trusted sample for supervised classification.

[0023] In a possible implementation, determining an initial prediction sample for supervised classification according to the change intensity and the classification result label includes:

[0024] According to the classification result label and the change intensity, multiple change intensity partitions are determined;

[0025] For multiple change intensity partitions, multiple partition thresholds are determined;

[0026] According to multiple partition thresholds and change intensities, the unchanged areas are determined;

[0027] The unchanged area and the classification result label are used as output for overlay processing to obtain the initial prediction sample for supervised classification.

[0028] In a possible implementation, performing feature analysis on the first purified sample to retain tags with variation intensities within a variation intensity range to obtain a second purified sample includes:

[0029] According to the change intensity and the unchanged area, determine the mean value and variance of the change intensity corresponding to each change intensity partition;

[0030] According to the mean value and variance, determine the change intensity range corresponding to each change intensity partition;

[0031] The first purified sample is screened according to the variation intensity range to retain tags with variation intensities within the variation intensity range, thereby obtaining a second purified sample.

[0032] In a possible implementation, respectively calculating the spectral slope vector corresponding to the source period image and the spectral slope vector corresponding to each target period image includes:

[0033] Calculate the spectral slope corresponding to the source period image and the spectral slope corresponding to each target period image respectively;

[0034] According to the spectral slope corresponding to the source period image and the spectral slope corresponding to each target period image, the spectral slope vector corresponding to the source period image and the spectral slope vector corresponding to each target period image are calculated.

[0035] In a second aspect, the present application provides a remote sensing land cover classification device under transfer learning, comprising:

[0036] The acquisition module is used to obtain source period images, classification result labels and multiple target period images;

[0037] The first calculation module is used to calculate the spectral slope vector corresponding to the source period image and the spectral slope vector corresponding to each target period image respectively;

[0038] The second calculation module is used to calculate the change intensity of each target period image and the source period image according to the spectral slope vector corresponding to the source period image and the spectral slope vector corresponding to each target period image;

[0039] A first processing module is used to determine an initial prediction sample for supervised classification according to the change intensity and the classification result label;

[0040] The second processing module is used to purify the initial prediction samples to obtain reliable samples for supervised classification;

[0041] The third processing module is used to perform supervised classification on multiple target period images based on credible samples to obtain target land cover products.

[0042] In a possible implementation manner, the second processing module is further used for:

[0043] According to the clustering characteristics of the spatial distribution of the objects, the broken labels in the initial prediction sample are deleted to obtain the first purified sample;

[0044] Performing feature analysis on the first purified sample to retain tags with change intensities within a change intensity range, thereby obtaining a second purified sample;

[0045] The Markov random field model and information entropy theory are used to perform computational processing on the second purified sample to obtain a credible sample for supervised classification.

[0046] In a possible implementation manner, the second processing module is further used for:

[0047] The second purified sample is used as a label to pre-classify the images of multiple target periods using a support vector machine model to obtain a pre-classification posterior probability and a pre-classification result;

[0048] According to the pre-classification results, a Markov random field model is introduced to output the normalized total energy image;

[0049] For the pre-classification posterior probability, the information entropy theory is introduced to calculate the entropy value graph;

[0050] A weighted map is generated according to the normalized total energy image and the entropy map, and a credible area is determined based on the weighted map;

[0051] The trusted region is spatially overlapped with the second purified sample to obtain a trusted sample for supervised classification.

[0052] In a possible implementation manner, the first processing module is further configured to:

[0053] According to the classification result label and the change intensity, multiple change intensity partitions are determined;

[0054] For multiple change intensity partitions, multiple partition thresholds are determined;

[0055] According to multiple partition thresholds and change intensities, the unchanged areas are determined;

[0056] The unchanged area and the classification result label are used as output for overlay processing to obtain the initial prediction sample for supervised classification.

[0057] In a possible implementation manner, the second processing module is further used for:

[0058] According to the change intensity and the unchanged area, determine the mean value and variance of the change intensity corresponding to each change intensity partition;

[0059] According to the mean value and variance, determine the change intensity range corresponding to each change intensity partition;

[0060] The first purified sample is screened according to the variation intensity range to retain tags with variation intensities within the variation intensity range, thereby obtaining a second purified sample.

[0061] In a possible implementation manner, the second calculation module is further used for:

[0062] Calculate the spectral slope corresponding to the source period image and the spectral slope corresponding to each target period image respectively;

[0063] According to the spectral slope corresponding to the source period image and the spectral slope corresponding to each target period image, the spectral slope vector corresponding to the source period image and the spectral slope vector corresponding to each target period image are calculated.

[0064] In a third aspect, the present application provides a remote sensing land cover classification device under transfer learning, including: a memory, a processor;

[0065] The memory stores computer-executable instructions;

[0066] The processor executes the computer-executable instructions stored in the memory, so that the processor executes the above first aspect and / or various possible implementations of the first aspect.

[0067] In a fourth aspect, the present application provides a computer-readable storage medium, in which computer-executable instructions are stored. When the computer-executable instructions are executed by a processor, they are used to implement the above first aspect and / or various possible implementations of the first aspect.

[0068] In a fifth aspect, the present application provides a computer program product, including a computer program, which, when executed by a processor, implements the above first aspect and / or various possible implementations of the first aspect.

[0069] The present application provides a remote sensing land cover classification method, device, equipment and storage medium under transfer learning, which obtains a source period image, a classification result label corresponding to the source period image and multiple target period images; respectively calculates the spectral slope vector corresponding to the source period image and the spectral slope vector corresponding to each target period image; calculates the change intensity of each target period image and the source period image according to the spectral slope vector corresponding to the source period image and the spectral slope vector corresponding to each target period image; determines the initial prediction sample for supervised classification according to the change intensity and the classification result label; purifies the initial prediction sample to obtain a credible sample for supervised classification; and supervises the classification of multiple target period images according to the credible sample to obtain a target land cover product, thereby first obtaining the classification result label corresponding to the source period image, and based on these labels The spectral slope difference method is used to automatically generate and supervise the purification of samples, without the need for a large amount of manual labeling work, which greatly reduces the time and labor cost of manual sample preparation and labeling. Secondly, by using the source period images, the classification results of the source period images and multiple target period images, by calculating and comparing the change intensity of the spectral slope vector, land cover classification products can be quickly generated for multiple target period images. There is no need to re-produce samples and classify them for each target period image, which significantly improves the temporal resolution of the land cover product. At the same time, the spectral slope difference is considered to detect changes in ground objects, and the spectral slope vector is used to purify the initial prediction samples, which can remove noise and misclassified samples, improve the credibility of the samples, and then improve the accuracy of the classification, thereby achieving the effect of improving the temporal resolution, classification efficiency and accuracy of remote sensing land cover classification products. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0071] Figure 1 A schematic diagram of an application data processing system architecture provided in an embodiment of the present application;

[0072] Figure 2 Schematic diagram of the process of remote sensing land cover classification method under transfer learning provided in the embodiment of the present application Figure 1 ;

[0073] Figure 3 Schematic diagram of the process of remote sensing land cover classification method under transfer learning provided in the embodiment of the present application Figure 2 ;

[0074] Figure 4 Schematic diagram of the process of remote sensing land cover classification method under transfer learning provided in the embodiment of the present application Figure 3 ;

[0075] Figure 5 A schematic diagram of partition threshold segmentation provided in an embodiment of the present application;

[0076] Figure 6 Schematic diagram of the process of remote sensing land cover classification under transfer learning provided in the embodiment of this application Figure 4 ;

[0077] Figure 7 An image schematic diagram of the remote sensing land cover classification process results under transfer learning provided in an embodiment of the present application;

[0078] Figure 8 A schematic diagram of the structure of a remote sensing land cover classification device under transfer learning provided in an embodiment of the present application;

[0079] Fig. 9 A schematic diagram of the structure of a remote sensing land cover classification device under transfer learning provided in an embodiment of the present application.

[0080] The above drawings have shown clear embodiments of the present application, which will be described in more detail later. These drawings and text descriptions are not intended to limit the scope of the present application in any way, but to illustrate the concept of the present application to those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0081] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application. Instead, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims.

[0082] Since the supervised classification method in the existing technical methods requires manual labeling of different categories of ground object samples, it greatly increases the time and labor costs, and the unsupervised classification method cannot eliminate the interference of the same spectrum but different objects and the same object but different spectra, resulting in reduced accuracy. There are technical problems such as low efficiency of remote sensing land classification processing and poor classification accuracy.

[0083] In response to the above problems, the embodiments of the present application provide a remote sensing land cover classification method, device, equipment and storage medium under transfer learning. First, by obtaining the classification result labels corresponding to the source period images, and automatically generating and supervising the purified samples based on these labels and the spectral slope difference method, there is no need to perform a large amount of manual labeling work, which greatly reduces the time and labor cost of manual sample preparation and labeling. Secondly, by using the source period images, the classification result labels of the source period images and multiple target period images, by calculating and comparing the change intensity of the spectral slope vector, land cover classification products can be quickly generated for multiple target period images. There is no need to re-produce samples and classify them for each target period image, which significantly improves the temporal resolution of the land cover product. At the same time, the spectral slope difference is considered to detect the change of the ground object, and the spectral slope vector is used to purify the initial prediction sample, which can remove noise and misclassified samples, improve the credibility of the sample, and then improve the accuracy of the classification, thereby achieving the effect of improving the temporal resolution, classification efficiency and accuracy of the remote sensing land cover classification product.

[0084] The technical solution of the present application and how the technical solution of the present application solves the above-mentioned technical problems are described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of the present application will be described below in conjunction with the accompanying drawings.

[0085] Figure 1 The present invention provides an application data processing system architecture diagram, wherein the application data processing system is a computer device. Figure 1 As shown, the above architecture includes at least one of a data acquisition device 101 , a processing device 102 and a display device 103 .

[0086] It is understandable that the structure illustrated in the embodiment of the present application does not constitute a specific limitation on the application data processing system architecture. In other feasible implementations of the present application, the above architecture may include more or fewer components than shown in the figure, or combine certain components, or split certain components, or arrange the components differently, which can be determined according to the actual application scenario and is not limited here. Figure 1 The components shown may be implemented in hardware, software, or a combination of software and hardware.

[0087] In a specific implementation process, the data acquisition device 101 may include an input / output interface and may also include a communication interface. The data acquisition device 101 may be connected to the processing device via the input / output interface or the communication interface.

[0088] The processing device 102 can obtain a source period image, a classification result label corresponding to the source period image, and multiple target period images; respectively calculate the spectral slope vector corresponding to the source period image and the spectral slope vector corresponding to each target period image; calculate the change intensity of each target period image and the source period image based on the spectral slope vector corresponding to the source period image and the spectral slope vector corresponding to each target period image; determine an initial prediction sample for supervised classification based on the change intensity and the classification result label; purify the initial prediction sample to obtain a credible sample for supervised classification; and perform supervised classification on multiple target period images based on the credible sample to obtain a target land cover product.

[0089] The display device 103 may also be a touch display screen or a screen of a terminal device, which is used to receive user instructions while displaying the above content to achieve interaction with the user.

[0090] It should be understood that the above-mentioned processing device can be implemented by a processor reading instructions in a memory and executing the instructions, or it can be implemented by a chip circuit.

[0091] In addition, the network architecture and business scenarios described in the embodiments of the present application are intended to more clearly illustrate the technical solutions of the embodiments of the present application, and do not constitute a limitation on the technical solutions provided in the embodiments of the present application. Ordinary technicians in this field can know that with the evolution of network architecture and the emergence of new business scenarios, the technical solutions provided in the embodiments of the present application are also applicable to similar technical problems.

[0092] Figure 2 Schematic diagram of the process of remote sensing land cover classification method under transfer learning provided in the embodiment of the present application Figure 1 ,like Figure 2 As shown, the remote sensing land cover classification method under transfer learning provided in this embodiment includes:

[0093] S201, obtaining source period images, classification result labels corresponding to the source period images, and multiple target period images;

[0094] In this embodiment, the source period image refers to a high-resolution remote sensing image of a reference period for comparison, and the target period image refers to a high-resolution remote sensing image acquired at a target time point.

[0095] Obtain the multispectral remote sensing source period images of the study area, the classification result labels corresponding to the source period images, and multiple target period images.

[0096] S202, respectively calculating the spectral slope vector corresponding to the source period image and the spectral slope vector corresponding to each target period image;

[0097] The source period image and each target period image are spectrally analyzed, and the spectral slope vector corresponding to the source period image and the spectral slope vector corresponding to each target period image are calculated respectively.

[0098] S203, calculating the change intensity of each target period image and the source period image according to the spectral slope vector corresponding to the source period image and the spectral slope vector corresponding to each target period image;

[0099] By comparing the spectral slope vector corresponding to the source period image obtained in the above steps with the spectral slope vector corresponding to each target period image, the change intensity of each target period image and the source period image is further calculated. The calculation formula is as follows:

[0100]

[0101] in, is the change intensity of the corresponding pixels of the source image and each target image, is the spectral slope vector of the source image, is the spectral slope vector of each target period image, It is the ratio of the difference in spectral reflectance to the difference in the corresponding band.

[0102] S204, determining an initial prediction sample for supervised classification according to the change intensity and the classification result label;

[0103] For the source image and each target image, the intensity of the change ,The partition threshold method is used to obtain the change intensity threshold of different ground objects, and spatial analysis is used to obtain the unchanged area and the initial sample between the source image and the target image of the study area.

[0104] Specifically, according to the existing classification result label C1 corresponding to the source image (assuming there are C categories in total), the change intensity Partitioning is performed, and then the maximum inter-class variance method (Otsu method) is used to set the threshold for the C change intensity images after partitioning to obtain C threshold values ​​t 1, t2,…,t c In ArcGIS, threshold segmentation is used to obtain the changed and unchanged areas. The unchanged area and the classification result label C1 are used as input for overlay analysis to obtain the initial sample L required for supervised classification of the target image. i .

[0105] S205, purifying the initial prediction samples to obtain reliable samples for supervised classification;

[0106] According to the clustering characteristics of the spatial distribution of objects, the Markov random field model and the information entropy theory, the initial prediction samples are purified to further obtain reliable samples that can be used for supervised classification.

[0107] S206. Based on the credible samples, supervised classification is performed on the images of multiple target periods to obtain target land cover products.

[0108] Based on the credible samples, the Maximum Likelihood Classification (MLC) method is used to perform supervised classification on images of multiple target periods to obtain the classified target land cover products.

[0109] The remote sensing land cover classification method under transfer learning provided in the embodiment of the present application obtains a source period image, a classification result label corresponding to the source period image, and multiple target period images; respectively calculates the spectral slope vector corresponding to the source period image and the spectral slope vector corresponding to each target period image; calculates the change intensity of each target period image and the source period image according to the spectral slope vector corresponding to the source period image and the spectral slope vector corresponding to each target period image; determines the initial prediction sample for supervised classification according to the change intensity and the classification result label; purifies the initial prediction sample to obtain a credible sample for supervised classification; and supervises and classifies multiple target period images according to the credible sample to obtain a target land cover product, thereby first obtaining the classification result label corresponding to the source period image, and based on these labels and the spectral slope The difference method automatically generates and supervises the purification of samples without a lot of manual labeling work, which greatly reduces the time and labor cost of manual sample preparation and labeling. Secondly, by using the source period images, the classification result labels of the source period images and multiple target period images, by calculating and comparing the change intensity of the spectral slope vector, it can quickly generate land cover classification products for multiple target period images. There is no need to re-produce samples and classify for each target period image, which significantly improves the temporal resolution of land cover products. At the same time, the spectral slope difference is considered to detect the change of ground objects, and the spectral slope vector is used to purify the initial prediction samples, which can remove noise and misclassified samples, improve the credibility of the samples, and then improve the accuracy of classification, so as to achieve the effect of improving the temporal resolution, classification efficiency and accuracy of remote sensing land cover classification products.

[0110] Figure 3 Schematic diagram of the process of remote sensing land cover classification method under transfer learning provided in the embodiment of the present application Figure 2 ,like Figure 3 As shown, this embodiment, based on the above embodiment, describes in detail the purification process of the initial prediction sample, and the method includes:

[0111] S301, according to the aggregation characteristics of the spatial distribution of the ground objects, deleting the fragmented labels in the initial prediction sample to obtain a first purified sample;

[0112] Specifically, considering the clustering characteristics of the spatial distribution of objects, assuming that each retained sample consists of at least q pixels, raster analysis is used to remove the initial sample L i The broken tags in the sample are obtained to obtain the first purified sample L after preliminary purification. i 1 .

[0113] S302, determining the average value and variance of the change intensity corresponding to each change intensity partition according to the change intensity and the unchanged area, and determining the change intensity range corresponding to each change intensity partition according to the average value and the variance;

[0114] Specifically, the intensity variation image obtained in step S203 is used The unchanged areas of each category obtained in S204 are integrated and analyzed in ArcGIS to obtain the average value of the change intensity of category C features. and variance , the characteristic statistical method of mean ± N times the variance is used to obtain the range of change intensity of each type of land feature.

[0115] S303, screening the first purified sample according to the change intensity range to retain tags with change intensities within the change intensity range, to obtain a second purified sample;

[0116] Specifically, the sample L is retained by the raster calculator in ArcGIS i 1 The pixel change intensity is included in the label within this range, and the second purified sample L after secondary purification is obtained. i 2 .

[0117] S304, using the second purified sample as a label to perform pre-classification processing on multiple target period images using a support vector machine model to obtain a pre-classification posterior probability and a pre-classification result;

[0118] Specifically, the sample L after secondary purification i 2 Use the support vector machine model (SVM) as a label to pre-classify the target image and obtain the pre-classification posterior probability P and pre-classification result C pre .

[0119] S305, introducing a Markov random field model for the pre-classification result to output a normalized total energy image;

[0120] Specifically, the Markov random field model refers to a binary Markov random field model or a Gaussian Markov random field model. In the Markov random field model, the probability P of a pixel (m, n) belonging to category X is determined by the spectral information of the pixel and the adjacent pixels. The probability obtained by the spectral information and the probability obtained by the texture information are added after taking the logarithm respectively, thereby converting the maximum a posteriori probability problem into a minimum energy problem, as shown in the following formula:

[0121]

[0122] Where Z is the normalization constant, and are the texture energy and spectral energy of pixel (m,n) respectively. The texture energy is calculated based on the classification information of adjacent pixels. For pixel (m,n), the second-order neighborhood system N(m,n) is defined by the following formula:

[0123]

[0124] The texture energy can be obtained through the second-order neighborhood system, which is calculated as follows:

[0125]

[0126] Among them, β is the weight coefficient of texture energy, which is usually between 0.5 and 2. As shown below:

[0127]

[0128] Considering that each pixel point is independent and each class of pixels satisfies Gaussian distribution, a Gaussian density function is established for each class. Then for a pixel with a gray value of s, the probability of it belonging to each class can be obtained, and the logarithm of the probability value is used as the spectral energy value of the pixel. As shown in the following formula:

[0129]

[0130] Among them, x is the gray value of the pixel, μ and σ are the mean and variance of the gray value of each type of ground object. After obtaining the spectral energy value and texture energy value, it is necessary to find the spectral energy and texture energy with the minimum total energy. That is, as shown in the following formula:

[0131]

[0132] The iterated conditional modes (ICM) optimization algorithm is used to minimize the energy function. The iterative steps are as follows:

[0133] Step 1: Introduce the pre-classification result C of the SVM classifierpre As the initial label for classification;

[0134] Step 2: Traverse the pixels of the image and calculate the spectral energy and texture energy of each pixel point when it becomes different categories;

[0135] Step 3: Compare the total energy values ​​of each pixel point when it becomes different categories, and select the state when the total energy value reaches the minimum as the category of the point;

[0136] Repeat steps 2 and 3 for a finite number of iterations until convergence, and output the total energy function value at convergence to obtain the maximum posterior probability P.

[0137] S306. Based on the pre-classification posterior probability, the information entropy theory is introduced to calculate the entropy value graph;

[0138] Specifically, for the pre-classification posterior probability P, first make it clear that for a pixel, the greater the difference in the probability value of belonging to each category, the less likely the pixel is to be misclassified; on the contrary, if the difference in the probability value of belonging to each category is relatively small, the greater the probability of being misclassified. By calculating the Shannon entropy of each pixel, it is more difficult to determine the category of pixels with larger entropy values, while it is relatively easy to determine the category of pixels with smaller entropy values. The definition of Shannon entropy is as follows:

[0139]

[0140] in, is the value of the random variable X The probability of X is greater, the greater the entropy value, the greater the uncertainty. When the probability values ​​of are equal, the entropy value is the largest, as shown below:

[0141]

[0142] Therefore, the principle of information entropy is to make no subjective assumptions when the situation is unknown. In this case, the probability distribution is the most uniform and the uncertainty of the prediction is the smallest, and the entropy value is the largest at this time.

[0143] S307, generating a weighted graph according to the normalized total energy image and the entropy value graph, and determining a credible region based on the weighted graph;

[0144] Specifically, for the normalized total energy image I1 and the entropy map I2, the weighted summation method is used to obtain the weighted map I3, and the weights are set to 0.6 and 0.4 respectively. Then, the Otsu method is used to automatically segment the probability image I3 by threshold to obtain the credible area. It should be clear that the smaller the pixel value of I3, the easier it is to determine the category of the object and the higher the credibility.

[0145] S308 : Perform spatial superposition processing on the credible region and the second purified sample to obtain a credible sample for supervised classification.

[0146] In ArcGIS, the trusted area is compared with the second purified sample L i 2 Perform spatial overlay analysis to finally obtain the reliable sample L i 3 .

[0147] The remote sensing land cover classification method under transfer learning provided in the embodiment of the present application can effectively remove noise and irregular small areas by deleting fragmented labels according to the clustering characteristics of the spatial distribution of the objects, thereby obtaining a purer first purified sample. In addition, the mean value and variance of each change intensity partition are determined according to the change intensity and the unchanged area, and the change intensity range is determined accordingly, which helps to more accurately divide the change characteristics of different areas. At the same time, a support vector machine model is used to pre-classify images of multiple target periods, and a Markov random field model and information entropy theory are introduced to generate normalized total energy images and entropy value maps, respectively, which can further analyze the spatial structure and information distribution of the image, help maintain spatial coherence during the classification process, and reduce classification errors. Finally, the credible areas are determined by weighted graphs, and these areas are spatially superimposed with the second purified samples to obtain credible samples for supervised classification, ensuring that the selected samples are more representative, thereby improving the reliability and classification processing speed of subsequent classification, and achieving the effect of improving the temporal resolution, classification efficiency and accuracy of remote sensing land cover classification products.

[0148] Figure 4 Schematic diagram of the process of remote sensing land cover classification method under transfer learning provided in the embodiment of the present application Figure 3 ,like Figure 4 As shown, this embodiment, based on the above embodiment, describes in detail the process of obtaining the initial prediction sample, including:

[0149] S401, determining a plurality of change intensity partitions according to the classification result label and the change intensity;

[0150] Specifically, Figure 5 A schematic diagram of partition threshold segmentation provided in an embodiment of the present application is shown in FIG. Figure 5 As shown, according to the existing classification result labels corresponding to the source image, the change intensity is partitioned to obtain multiple change intensity partitions.

[0151] It should be noted that Figure 5 It is only an illustrative display result and not an improvement point, and does not affect the protection scope of the embodiments of the present application.

[0152] S402, determining multiple partition thresholds for multiple change intensity partitions;

[0153] Specifically, Otsu's method is used to set thresholds for the multiple intensity variation images after partitioning to obtain multiple corresponding thresholds.

[0154] S403, determining unchanged areas according to multiple partition thresholds and change intensities;

[0155] Specifically, threshold segmentation is used in ArcGIS to obtain the changed and unchanged areas.

[0156] S404: Overlay processing is performed with the unchanged area and the classification result label as output to obtain an initial prediction sample for supervised classification.

[0157] The unchanged area and the classification result label are used as input for overlay analysis to obtain the initial samples required for supervised classification of the target image.

[0158] The remote sensing land cover classification method provided in the embodiment of the present application can more finely divide the areas of different change degrees in the image by determining multiple change intensity partitions according to the classification result label and the change intensity, which is helpful to more accurately analyze and process the image data in the subsequent steps. In addition, the partition threshold is determined for the multiple change intensity partitions to ensure that the change intensity in each partition is within a reasonable range. At the same time, the unchanged area is determined according to the partition threshold and the change intensity, and the stable and unchanging part of the image can be accurately identified. Finally, the unchanged area and the classification result label are superimposed to obtain the initial prediction samples for supervised classification. These samples contain both the information of the unchanged area and the information of the classification result label, so that the quality and reliability are further improved, thereby improving the processing speed and accuracy of the subsequent supervised classification, and achieving the effect of improving the temporal resolution, classification efficiency and accuracy of the remote sensing land cover classification product.

[0159] Figure 6 Schematic diagram of the process of remote sensing land cover classification method under transfer learning provided in the embodiment of the present application Figure 4 ,like Figure 6 As shown, based on the above embodiment, this embodiment provides a supplementary explanation of the calculation process of the spectral slope vector corresponding to the source period image and the spectral slope vector corresponding to each target period image, including:

[0160] S601, respectively calculating the spectral slope corresponding to the source period image and the spectral slope corresponding to each target period image;

[0161] For source images And each target period image , respectively calculate the slope between adjacent bands, the spectral slope It refers to the ratio of the difference in spectral reflectance to the corresponding band difference, and its calculation formula is shown as follows:

[0162]

[0163] in, It represents the difference in reflectivity between adjacent bands. Indicates adjacent bands and The band length difference is and Bands and The reflectivity at and Indicates the corresponding wavelength.

[0164] S602 , calculating a spectral slope vector corresponding to the source period image and a spectral slope vector corresponding to each target period image according to the spectral slope corresponding to the source period image and the spectral slope corresponding to each target period image.

[0165] For source images And each target period image , combining the spectral slopes into a vector , the slope vector is used to describe the shape of the spectral curve, as shown in the following formula:

[0166]

[0167] in, Refers to the number of bands of the image. After the above steps, the source image is obtained. And each target period image The corresponding spectral slope vector and .

[0168] The remote sensing land cover classification method under transfer learning provided in the embodiment of the present application provides key spectral information support for monitoring land object changes, improving classification accuracy and subsequent land object identification and analysis tasks by extracting and comparing spectral features at different time points, thereby improving the processing speed and accuracy of subsequent supervised classification, and achieving the effect of improving the temporal resolution, classification efficiency and accuracy of remote sensing land cover classification products.

[0169] Figure 7 This is a schematic diagram of the image of the remote sensing land cover classification process results under the transfer learning provided in the embodiment of the present application, such as Figure 7 As shown, by using spectral analysis technology, the remote sensing land cover classification process provided by the above embodiments is gradually displayed by image, and the corresponding result schematic diagram can be obtained, wherein: Figure 7 a is the source image, Figure 7b is the target image, Figure 7 c is the source classification product map, Figure 7 d is the intensity variation image. Figure 7 e is the partition threshold segmentation map, Figure 7 f is the first purified sample graph, Figure 7 g is the second purified sample graph, Figure 7 h is the SVM pre-classified image, Figure 7 i is the entropy value image, Figure 7 j is the normalized total energy image, Figure 7 k is the weighted credibility image, Figure 7 l is the credible sample area image, Figure 7 m is a credible sample image, Figure 7 n is the target classification product image.

[0170] Understandably, Figure 7 It is only used as a reference for display effect and is not used as an improvement point, and does not affect the protection scope of the embodiments of the present application.

[0171] Figure 8 This is a schematic diagram of the structure of the remote sensing land cover classification device under transfer learning provided in the embodiment of the present application. The device of this embodiment can be in the form of software and / or hardware. Figure 8 As shown, the remote sensing land cover classification device 800 under transfer learning provided by the embodiment of the present application includes: an acquisition module 801, a first calculation module 802, a second calculation module 803, a first processing module 804, a second processing module 805, and a third processing module 806:

[0172] An acquisition module 801 is used to acquire source period images, classification result labels and multiple target period images;

[0173] The first calculation module 802 is used to calculate the spectral slope vector corresponding to the source period image and the spectral slope vector corresponding to each target period image respectively;

[0174] The second calculation module 803 is used to calculate the change intensity of each target period image and the source period image according to the spectral slope vector corresponding to the source period image and the spectral slope vector corresponding to each target period image;

[0175] The first processing module 804 is used to determine an initial prediction sample for supervised classification according to the change intensity and the classification result label;

[0176] The second processing module 805 is used to purify the initial prediction sample to obtain a reliable sample for supervised classification;

[0177] The third processing module 806 is used to perform supervised classification on multiple target period images based on the credible samples to obtain the target land cover product.

[0178] In a possible implementation manner, the second processing module 805 is further configured to:

[0179] According to the clustering characteristics of the spatial distribution of the objects, the broken labels in the initial prediction sample are deleted to obtain the first purified sample;

[0180] Performing feature analysis on the first purified sample to retain tags with change intensities within a change intensity range, thereby obtaining a second purified sample;

[0181] The Markov random field model and information entropy theory are used to perform computational processing on the second purified sample to obtain a credible sample for supervised classification.

[0182] In a possible implementation manner, the second processing module 805 is further configured to:

[0183] The second purified sample is used as a label to pre-classify the images of multiple target periods using a support vector machine model to obtain a pre-classification posterior probability and a pre-classification result;

[0184] According to the pre-classification results, a Markov random field model is introduced to output the normalized total energy image;

[0185] For the pre-classification posterior probability, the information entropy theory is introduced to calculate the entropy value graph;

[0186] A weighted map is generated according to the normalized total energy image and the entropy map, and a credible area is determined based on the weighted map;

[0187] The trusted region is spatially overlapped with the second purified sample to obtain a trusted sample for supervised classification.

[0188] In a possible implementation, the first processing module 804 is further configured to:

[0189] According to the classification result label and the change intensity, multiple change intensity partitions are determined;

[0190] For multiple change intensity partitions, multiple partition thresholds are determined;

[0191] According to multiple partition thresholds and change intensities, the unchanged areas are determined;

[0192] The unchanged area and the classification result label are used as output for overlay processing to obtain the initial prediction sample for supervised classification.

[0193] In a possible implementation manner, the second processing module 805 is further configured to:

[0194] According to the change intensity and the unchanged area, determine the mean value and variance of the change intensity corresponding to each change intensity partition;

[0195] According to the mean value and variance, determine the change intensity range corresponding to each change intensity partition;

[0196] The first purified sample is screened according to the variation intensity range to retain tags with variation intensities within the variation intensity range, thereby obtaining a second purified sample.

[0197] In a possible implementation manner, the second calculation module 803 is further configured to:

[0198] Calculate the spectral slope corresponding to the source period image and the spectral slope corresponding to each target period image respectively;

[0199] According to the spectral slope corresponding to the source period image and the spectral slope corresponding to each target period image, the spectral slope vector corresponding to the source period image and the spectral slope vector corresponding to each target period image are calculated.

[0200] The remote sensing land cover classification device under transfer learning provided in this embodiment can execute the method provided in the above method embodiment. Its implementation principle and technical effect are similar, and this embodiment will not be repeated here.

[0201] Fig. 9 This is a schematic diagram of the structure of a remote sensing land cover classification device under transfer learning provided in an embodiment of the present application. Fig. 9 As shown, the electronic device 900 provided in this embodiment includes: at least one processor 901 and a memory 902. Optionally, the device 900 also includes a communication component 903. The processor 901, the memory 902 and the communication component 903 are connected via a bus.

[0202] In a specific implementation process, at least one processor 901 executes the computer execution instructions stored in the memory 902, so that at least one processor 901 executes the above method.

[0203] The specific implementation process of the processor 901 can be found in the above method embodiment, and its implementation principle and technical effect are similar, so this embodiment will not be repeated here.

[0204] In the above embodiments, it should be understood that the processor can be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), etc. A general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in the invention can be directly implemented as a hardware processor, or can be implemented by a combination of hardware and software modules in the processor.

[0205] The memory may include a high-speed memory (Random Access Memory, RAM), and may also include a non-volatile memory (NVM), such as at least one disk storage.

[0206] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, the bus in the drawings of this application is not limited to only one bus or one type of bus.

[0207] The present application also provides a computer program product, including a computer program, which implements the above method when executed by a processor.

[0208] The present application also provides a computer-readable storage medium, in which computer-executable instructions are stored. When a processor executes the computer-executable instructions, the above method is implemented.

[0209] The above-mentioned readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk. The readable storage medium can be any available medium that can be accessed by a general or special-purpose computer.

[0210] An exemplary readable storage medium is coupled to a processor so that the processor can read information from the readable storage medium and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can be located in an application specific integrated circuit (Application Specific Integrated Circuits, referred to as: ASIC). Of course, the processor and the readable storage medium can also exist in the device as discrete components.

[0211] The division of units is only a logical function division, and there may be other divisions in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interface, device or unit, which can be electrical, mechanical or other forms.

[0212] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0213] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0214] If the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods of each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0215] Those skilled in the art can understand that all or part of the steps of implementing the above-mentioned method embodiments can be completed by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, the steps of the above-mentioned method embodiments are executed; and the aforementioned storage medium includes: ROM, RAM, disk or optical disk and other media that can store program codes.

[0216] Finally, it should be noted that those skilled in the art will readily conceive of other embodiments of the present invention after considering the specification and practicing the invention disclosed herein. The present invention is intended to cover any variations, uses or adaptations of the present invention, which follow the general principles of the present invention and include common knowledge or customary technical means in the art not disclosed by the present invention, are not limited to the precise structure described above and shown in the drawings, and may be modified and changed in various ways without departing from the scope thereof. The scope of the present invention is limited only by the appended claims.

Claims

1. A remote sensing land cover classification method under transfer learning, characterized in that: include: Obtaining source period images, classification result labels corresponding to the source period images, and multiple target period images; Respectively calculating the spectral slope vector corresponding to the source period image and the spectral slope vector corresponding to each of the target period images; Calculating the change intensity of each target period image and the source period image according to the spectral slope vector corresponding to the source period image and the spectral slope vector corresponding to each target period image; Determining an initial prediction sample for supervised classification according to the change intensity and the classification result label; Performing two purification processes and credibility judgment on the initial prediction samples to obtain credible samples for supervised classification; According to the credible samples, supervised classification is performed on the images of the plurality of target periods to obtain a target land cover product; The initial prediction sample is purified twice and the credibility is judged to obtain a credible sample for supervised classification, including: According to the clustering characteristics of the spatial distribution of the objects, the broken labels in the initial prediction sample are deleted to obtain a first purified sample; Performing feature analysis on the first purified sample to retain tags with change intensities within a change intensity range, thereby obtaining a second purified sample; Using the second purified sample as a label, pre-classifying the plurality of target period images using a support vector machine model to obtain a pre-classification posterior probability and a pre-classification result; According to the pre-classification result, a Markov random field model is introduced to output a normalized total energy image; According to the pre-classification posterior probability, the information entropy theory is introduced to calculate the entropy value graph; Generating a weighted graph according to the normalized total energy image and the entropy value graph, and determining a credible region based on the weighted graph; The credible region and the second purified sample are spatially overlapped to obtain a credible sample for supervised classification.

2. The method according to claim 1, characterized in that Determining an initial prediction sample for supervised classification according to the change intensity and the classification result label includes: Determining a plurality of change intensity partitions according to the classification result label and the change intensity; Determining a plurality of partition thresholds for the plurality of change intensity partitions; Determining unchanged areas according to the multiple partition thresholds and the change intensity; The unchanged region and the classification result label are used as outputs for superposition processing to obtain an initial prediction sample for supervised classification.

3. The method according to claim 2, characterized in that The step of performing feature analysis on the first purified sample to retain tags with variation intensities within a variation intensity range to obtain a second purified sample comprises: Determine, according to the change intensity and the unchanged area, a mean value and a variance of the change intensity corresponding to each of the change intensity partitions; Determine, according to the mean value and the variance, a change intensity range corresponding to each change intensity partition; The first purified sample is screened according to the variation intensity range to retain tags with variation intensities within the variation intensity range, thereby obtaining a second purified sample.

4. The method according to any one of claims 1 to 3, characterized in that: The respectively calculating the spectral slope vector corresponding to the source period image and the spectral slope vector corresponding to each of the target period images comprises: Calculating the spectral slope corresponding to the image in the source period and the spectral slope corresponding to each image in the target period respectively; According to the spectral slope corresponding to the source period image and the spectral slope corresponding to each of the target period images, the spectral slope vector corresponding to the source period image and the spectral slope vector corresponding to each of the target period images are calculated.

5. A remote sensing land cover classification device under transfer learning, characterized in that: include: An acquisition module, used to acquire a source period image, a classification result label corresponding to the source period image, and a plurality of target period images; A first calculation module, used for respectively calculating the spectral slope vector corresponding to the source period image and the spectral slope vector corresponding to each of the target period images; A second calculation module is used to calculate the change intensity of each target period image and the source period image according to the spectral slope vector corresponding to the source period image and the spectral slope vector corresponding to each target period image; A first processing module, used for determining an initial prediction sample for supervised classification according to the change intensity and the classification result label; A second processing module is used to perform two purification processes and credibility judgment on the initial prediction sample to obtain a credible sample for supervised classification; A third processing module is used to perform supervised classification on the multiple target period images according to the credible samples to obtain a target land cover product; The second processing module is specifically used to delete the broken labels in the initial prediction sample according to the clustering characteristics of the spatial distribution of the ground objects to obtain a first purified sample; perform feature analysis on the first purified sample to retain the labels whose change intensity is within the change intensity range to obtain a second purified sample; Using the second purified sample as a label, pre-classifying the plurality of target period images using a support vector machine model to obtain a pre-classification posterior probability and a pre-classification result; For the pre-classification result, a Markov random field model is introduced to output a normalized total energy image; for the pre-classification posterior probability, information entropy theory is introduced to calculate an entropy value map; a weighted map is generated according to the normalized total energy image and the entropy value map, and a credible area is determined based on the weighted map; The credible region and the second purified sample are spatially overlapped to obtain a credible sample for supervised classification.

6. A remote sensing land cover classification device under transfer learning, characterized in that: include: at least one processor; as well as a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the remote sensing land cover classification method under transfer learning as described in any one of claims 1 to 4.

7. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the remote sensing land cover classification method under transfer learning as described in any one of claims 1 to 4.

8. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the remote sensing land cover classification method under transfer learning as described in any one of claims 1 to 4 is implemented.

Images