High-resolution remote sensing interpretation sample data set construction method

Abstract

The invention provides a high-resolution remote sensing interpretation sample data set construction method. The method comprises the following steps: S101, acquiring high-resolution remote sensing image data by using a satellite remote sensing technology; s102, carrying out preprocessing operation on the high-resolution image data; s103, performing feature extraction on the high-resolution image data after the preprocessing operation; s104, performing image segmentation processing based on the high-resolution image data after feature extraction; s105, performing image interpretation based on the high-resolution image data after image segmentation processing, and constructing a sample data set; s106, performing sample quality inspection, sample quantity inspection and sample precision inspection on the sample data set; and S107, establishing a multi-user sample database based on the PostgreSQL to store and manage the sample data set. According to the method, the high-resolution remote sensing image sample data set is constructed, so that the automatic interpretation precision of the high-resolution image data is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of remote sensing image interpretation, and in particular to a method for constructing a high-resolution remote sensing interpretation sample data set. Background Art

[0002] The traditional remote sensing image survey method is to manually draw various image data from high-resolution remote sensing images. This method requires the mappers to have high professional knowledge and technical capabilities, and is inefficient. Faced with the vast survey area, massive data, and the actual needs of dynamic monitoring, there is an urgent need to develop efficient information extraction methods based on high-resolution remote sensing images. However, the traditional high-resolution remote sensing image automatic interpretation method has weak ability to extract and interpret visual features. It generally requires artificial design of features based on a large amount of prior knowledge, and the completeness of the features cannot be guaranteed. Intelligent methods represented by deep learning can automatically extract complex features from high-resolution remote sensing images and significantly improve the accuracy and efficiency of image interpretation. However, these methods rely on large-scale, high-quality sample data sets for training and optimization. However, the current sample data set construction process still requires a lot of manpower and material resources, and data annotation is cumbersome and time-consuming, which is difficult to meet the rapidly developing remote sensing intelligent interpretation needs. Summary of the invention

[0003] In view of this, an object of the present invention is to provide a method for constructing a high-resolution remote sensing interpretation sample data set to solve or at least partially solve the above-mentioned problems existing in the prior art.

[0004] To achieve the above-mentioned object of the invention, the present invention provides a method for constructing a high-resolution remote sensing interpretation sample data set, the method comprising the following steps:

[0005] S101. Use satellite remote sensing technology to obtain high-resolution remote sensing image data;

[0006] S102, performing preprocessing operations on high-resolution image data;

[0007] S103, extracting features from the high-resolution image data after the preprocessing operation;

[0008] S104, performing image segmentation processing based on the high-resolution image data after feature extraction;

[0009] S105, performing image interpretation based on the high-resolution image data after image segmentation processing, and constructing a sample data set;

[0010] S106, performing sample quality inspection, sample quantity inspection and sample accuracy inspection on the sample data set;

[0011] S107. Establish a multi-user sample database based on PostgreSQL to store and manage the sample data set.

[0012] Furthermore, the preprocessing operation comprises the following steps:

[0013] S31, radiation calibration: converting the brightness gray value of the acquired high-resolution remote sensing image data into a radiation brightness value;

[0014] S32, atmospheric correction: convert the above-mentioned radiation brightness value into the actual reflectivity of the surface;

[0015] S33, geometric correction: eliminate the distortion error of high-resolution remote sensing image data after S33 preprocessing, perform geographic coordinate positioning on the high-resolution remote sensing image data, and obtain the real coordinate information;

[0016] S34, registration and mosaicking: Based on steps S31-S32, image registration and image mosaicking are performed on the high-resolution remote sensing image data.

[0017] Furthermore, feature extraction is performed on the preprocessed high-resolution image data, including:

[0018] Use image mean, image variance and entropy to extract features from high-resolution image data;

[0019] The image average value is used to measure the average brightness of the image data texture;

[0020] The image variance is used to measure the average contrast of the image data texture;

[0021] The entropy is used to describe the measure of randomness of image data.

[0022] Furthermore, the image segmentation process includes the following steps:

[0023] S41, performing image segmentation processing based on the features of the high-resolution image data to obtain a preliminary image segmentation result;

[0024] S42, optimizing the preliminary segmentation result based on the features of the high-resolution image data;

[0025] S43. Based on step S42, perform spot feature extraction on the high-resolution image data after the optimized segmentation result.

[0026] Furthermore, the image interpretation adopts a method combining deep learning and expert interpretation, wherein the deep learning method extracts the boundaries of objects in high-resolution images and generates an automatic object recognition probability map based on object-oriented segmentation;

[0027] The expert interpretation method manually interprets the ground objects with low recognition probability based on the results of the deep learning method, and finally realizes the accurate classification of the boundaries and types of the ground objects.

[0028] Furthermore, the sample quality inspection includes radiation quality inspection, geometric quality inspection, and information quality inspection;

[0029] The sample quantity test includes overall quantity test, category quantity test and scale sample quantity test;

[0030] The sample accuracy test includes position accuracy test, shape accuracy test, category accuracy test and overall accuracy test.

[0031] Furthermore, the overall accuracy test includes:

[0032]

[0033] Among them, OA is the overall accuracy, N is the total number of categories, i is the category, TP is i is the number of correctly classified categories i, C i is the true number of category i.

[0034] Furthermore, the multi-user sample database is used for multi-user registration and login, multi-user upload and download, multi-user query and retrieval, multi-user evaluation and feedback, and multi-user update and maintenance.

[0035] Compared with the prior art, the present invention has the following beneficial effects:

[0036] The present invention improves the quality and availability of high-resolution image data by preprocessing the high-resolution image data; extracts features from the preprocessed high-resolution image data and performs image segmentation based on the features of the high-resolution image data, thereby accurately identifying the features of the high-resolution image data; interprets the high-resolution image data after the image segmentation process, thereby achieving accurate classification of the high-resolution image data and improving the accuracy of automatic interpretation of the high-resolution image data, and constructing a sample data set on this basis can save a lot of manpower; verifies the sample data set to ensure the accuracy and availability of the sample, and finally establishes a multi-user sample database to facilitate the storage and management of the sample data set. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only preferred embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0038] Figure 1 A schematic diagram of a process for constructing a high-resolution remote sensing interpretation sample data set provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0039] The principles and features of the present invention are described below in conjunction with the accompanying drawings. The enumerated embodiments are only used to explain the present invention and are not used to limit the scope of the present invention.

[0040] This embodiment provides a method for constructing a high-resolution remote sensing interpretation sample data set, the method comprising the following steps:

[0041] like Figure 1 As shown, the present invention provides a method for constructing a high-resolution remote sensing interpretation sample data set, the method comprising the following steps:

[0042] S101. Use satellite remote sensing technology to obtain high-resolution remote sensing image data.

[0043] S102, performing a preprocessing operation on the high-resolution image data, wherein the preprocessing operation includes the following steps:

[0044] S31, radiation calibration: convert the brightness gray value of the acquired high-resolution remote sensing image data into a radiation brightness value, or convert the recorded DN value (pixel brightness value of high-resolution remote sensing image data) into an apparent reflectance. The conversion between DN value and radiation brightness value is:

[0045] L λ = k·DN+c

[0046] Among them, L λ is the radiance value of band λ, k and c are gain and offset respectively. The conversion between radiance value and apparent reflectance is:

[0047]

[0048] Among them, ρ λ is the atmospheric apparent reflectivity, d is the average distance between the sun and the earth, ESUN λ is the solar irradiance and θ is the solar altitude angle.

[0049] S32. Atmospheric correction: The process of converting the above-mentioned radiation brightness values ​​into the actual surface reflectivity, eliminating the radiation error caused by atmospheric images, and reflecting the actual surface reflectivity of the surface.

[0050] S33, geometric correction: eliminate the distortion error of high-resolution remote sensing image data after S33 preprocessing, perform geographic coordinate positioning on the high-resolution remote sensing image data, and obtain the real coordinate information;

[0051] S34, registration and mosaicking: Based on steps S31-S32, image registration and image mosaicking are performed on the high-resolution remote sensing image data. The image registration is expressed as follows:

[0052] I2(x,y)=g(I1(f(x,y)))

[0053] Among them, I1(x, y) and I2(x, y) are the brightness values ​​(or other measurement values) of the two images respectively, g is the one-dimensional brightness or other measurement value transformation, and f is the two-dimensional space coordinate transformation (such as affine transformation); the image mosaic takes the maximum brightness value of the two images to be mosaicked as the brightness value of the pixel point in the overlapping area, and the overlapping part of the two images of x and y is from the Zth row to the x+Z-1th row, which is expressed as follows:

[0054] g(i,j)=max[g v (i+Z-1,j),g y (i,j)]

[0055] i=(1,2,...,x)

[0056] Among them, g(i,j) is the brightness value of the overlap area after brightness adjustment, i is the number of rows in the overlap area, j is the number of columns in the overlap area, and g v (i,j) and g y (i, j) represents the brightness values ​​of the v and y images respectively, and the overlapping part of the v image is from row 1 to row x.

[0057] S103, extracting features from the high-resolution image data after the preprocessing operation, specifically including:

[0058] After completing the preprocessing operation, the image mean, image variance, and entropy are used to extract features from the high-resolution image data, as shown below:

[0059] (1) Image average: used to measure the average brightness of image data texture, specifically expressed as follows:

[0060]

[0061] Among them, m is the image average, L is the total number of gray levels, z i is the i-th gray level, p(z i ) is z i The probability of gray levels in the image data;

[0062] (2) Image variance: It is used to measure the average contrast of image data texture, which is expressed as follows:

[0063]

[0064] Among them, σ is the image variance;

[0065] (3) Entropy: A measure used to describe the randomness of image data. The larger the value, the greater the randomness and the more information it contains. Conversely, the smaller the value, the greater the certainty and the less information it contains. The specific expression is as follows:

[0066]

[0067] Where e is entropy.

[0068] S104, performing image segmentation processing based on the high-resolution image data after feature extraction, specifically including:

[0069] S41. Perform image segmentation processing based on the features of the high-resolution image data to obtain a preliminary image segmentation result. The high-resolution image data after the image segmentation processing contains at least 2 objects.

[0070] S42. Optimizing the preliminary segmentation result based on the features of the high-resolution image data. The optimized image segmentation processing is expressed as follows:

[0071] F=g+MI·α

[0072] Among them, F is the optimized image segmentation processing, g is the smooth boundary, M is the hole filling coefficient, I is the support vector machine, and α is the patch clustering condition.

[0073] S43, based on step S42, the high-resolution image data after the optimization segmentation result is subjected to spot feature extraction, and the abnormal information in the spots is identified according to the feature changes of different spots, which is expressed as follows:

[0074]

[0075] Among them, Q is the feature information of the spot, β is the filtering parameter, and S is the number of pixels.

[0076] S105, performing image interpretation based on the high-resolution image data after image segmentation processing and constructing a sample data set, specifically including:

[0077] The image interpretation adopts a combination of deep learning and expert interpretation. The deep learning method extracts the boundaries of high-resolution image objects based on object-oriented segmentation, generates an automatic object recognition probability map, and automatically maps the deep learning object recognition results to high-resolution image data. The expert interpretation method is to manually interpret the objects with a low recognition probability based on the results of the deep learning method. Ultimately, accurate classification of object boundaries and types is achieved. The combination of deep learning and expert interpretation can automatically learn the features of high-resolution image data and improve the accuracy of automatic interpretation of high-resolution image data. Finally, a data set is constructed based on the accurate classification results of object boundaries and types.

[0078] The method of combining deep learning and expert interpretation includes:

[0079] (1) Automatic segmentation of high-resolution image data based on object-oriented approach;

[0080] (2) Identify the objects on the segmentation results based on the deep learning method and generate an automatic recognition probability map;

[0081] (3) Through expert interpretation methods, human visual interpretation and identification of ground objects with low recognition probability can be performed to obtain more accurate land type information;

[0082] (4) Edit and extract the final interpretation results based on land type information.

[0083] S106. Perform sample quality inspection, sample quantity inspection and sample accuracy inspection on the sample data set, specifically including:

[0084] (1) Sample quality inspection includes radiation quality inspection, geometric quality inspection, and information quality inspection;

[0085] Radiation quality inspection: Check whether the high-resolution image data has interference factors such as clouds, fog, shadows, clutter, and whether pre-processing operations such as radiation calibration and atmospheric correction have been performed;

[0086] Geometric quality inspection: Check whether the high-resolution image data has distortion, misalignment, overlap and other errors, and whether pre-processing operations such as geometric correction, registration and mosaicking have been performed;

[0087] Information quality inspection: Check whether the high-resolution image data has sufficient resolution, number of bands, dynamic range and other parameters.

[0088] (2) Sample quantity test includes overall quantity test, category quantity test, and scale sample quantity test;

[0089] Overall quantity test: check whether the total number of samples reaches the expected size;

[0090] Category quantity test: Check whether the number of samples in each category is balanced and whether it can reflect the diversity and differences of high-resolution image data;

[0091] Scale sample number test: Check whether the sample numbers at different scales are reasonable and whether they can reflect the hierarchy and details of high-resolution image data.

[0092] (3) Sample accuracy test includes position accuracy test, shape accuracy test, category accuracy test and overall accuracy test;

[0093] Location accuracy test: Check whether the sample is spatially consistent with the actual situation and whether it is consistent with other data sources (such as maps, GPS, etc.);

[0094] Shape accuracy test: Check whether the sample is consistent with the actual situation in terms of morphology and whether it is consistent with other data sources (such as texture, morphology, etc.);

[0095] Category accuracy test: Check whether the sample is consistent with other data sources (such as expert knowledge, rule base, etc.).

[0096] Overall accuracy test: refers to the ratio of the number of pixels correctly detected as shadows / shadow semantics (i.e., land cover information inside the shadow) to the number of categories set overall. It is mainly used to evaluate the overall accuracy of the sample data set based on confusion evidence, and is expressed as follows:

[0097]

[0098] Among them, OA is the overall accuracy, N is the total number of categories, i is the category, TP is i is the number of correctly classified categories i, C i is the true number of category i.

[0099] S107. Establish a multi-user sample database based on PostgreSQL to store and manage the sample data set. The multi-user sample database includes the following functions:

[0100] It supports multi-user registration and login to realize user identity authentication and permission control; it supports multi-user upload and download to realize sample data set sharing and communication among users; it supports multi-user query and retrieval to realize users' rapid query and screening of sample data sets; it supports multi-user evaluation and feedback to realize users' quality evaluation and improvement suggestions for sample data sets; it supports multi-user update and maintenance to realize users' addition, deletion, modification and version control of sample data sets.

[0101] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for constructing a high-resolution remote sensing interpretation sample data set, characterized in that: The method comprises the following steps: S101. Use satellite remote sensing technology to obtain high-resolution remote sensing image data; S102, performing preprocessing operations on high-resolution image data; S103, extracting features from the high-resolution image data after preprocessing; S104, performing image segmentation processing based on the high-resolution image data after feature extraction; S105, performing image interpretation based on the high-resolution image data after image segmentation processing, and constructing a sample data set; S106, performing sample quality inspection, sample quantity inspection and sample accuracy inspection on the sample data set; S107. Establish a multi-user sample database based on PostgreSQL to store and manage the sample data set.

2. The method for constructing a high-resolution remote sensing interpretation sample data set according to claim 1, characterized in that: In step S102, the preprocessing operation includes the following steps: S31, radiation calibration: converting the brightness gray value of the acquired high-resolution remote sensing image data into a radiation brightness value; S32, atmospheric correction: converting the radiation brightness value into actual surface reflectivity; S33, geometric correction: eliminate the distortion error of the high-resolution remote sensing image data after S32 preprocessing, perform geographic coordinate positioning on the high-resolution remote sensing image data, and obtain the real coordinate information; S34, registration and mosaicking: Based on steps S31-S32, image registration and image mosaicking are performed on the high-resolution remote sensing image data.

3. The method for constructing a high-resolution remote sensing interpretation sample data set according to claim 1, characterized in that: Feature extraction is performed on the preprocessed high-resolution image data, including: Use image mean, image variance and entropy to extract features from high-resolution image data; The image average value is used to measure the average brightness of the image data texture; The image variance is used to measure the average contrast of the image data texture; The entropy is used to describe the measure of randomness of image data.

4. The method for constructing a high-resolution remote sensing interpretation sample data set according to claim 1, characterized in that: In step S104, the image segmentation process includes the following steps: S41, performing image segmentation processing based on the features of the high-resolution image data to obtain a preliminary image segmentation result; S42, optimizing the preliminary segmentation result based on the features of the high-resolution image data; S43. Based on step S42, perform spot feature extraction on the high-resolution image data after the optimized segmentation result.

5. The method for constructing a high-resolution remote sensing interpretation sample data set according to claim 1, characterized in that: The image interpretation adopts a method combining deep learning and expert interpretation. The deep learning method extracts the boundaries of high-resolution image objects on the basis of object-oriented segmentation and generates an automatic object recognition probability map; The expert interpretation method manually interprets the ground objects with low recognition probability based on the results of the deep learning method, and finally realizes the accurate classification of the boundaries and types of the ground objects.

6. The method for constructing a high-resolution remote sensing interpretation sample data set according to claim 1, characterized in that: The sample quality inspection includes radiation quality inspection, geometric quality inspection, and information quality inspection; The sample quantity test includes overall quantity test, category quantity test and scale sample quantity test; The sample accuracy test includes position accuracy test, shape accuracy test, category accuracy test and overall accuracy test.

7. The method for constructing a high-resolution remote sensing interpretation sample data set according to claim 6, characterized in that: The overall accuracy test includes: Among them, OA is the overall accuracy, N is the total number of categories, i is the category, TP is i is the number of correctly classified categories i, C i is the true number of category i.

8. The method for constructing a high-resolution remote sensing interpretation sample data set according to claim 1, characterized in that: In step S107, the multi-user sample database is used for multi-user registration and login, multi-user upload and download, multi-user query and retrieval, multi-user evaluation and feedback, and multi-user update and maintenance.

Images