A system for intelligent extraction and analysis of ground object information from remote sensing images

Abstract

The invention relates to the technical field of remote sensing image processing, and discloses a remote sensing image ground object information intelligent extraction and analysis system, comprising: selecting M sub-feature information from spectral feature information according to a step-by-step feature analysis method, then obtaining a statistical coefficient corresponding to each sub-feature information, determining optimal feature information from the M sub-feature information based on the statistical coefficient and a preset intelligent optimization algorithm, and finally performing preliminary regional classification on initial ground object information according to the optimal feature information to obtain intermediate ground object information, and correcting the intermediate ground object information according to a feature similarity set to obtain target ground object information. In this way, the invention selects sub-feature information with the best distinguishing ability through the step-by-step feature analysis method, and further optimizes it using the intelligent optimization algorithm, ensuring that the extracted optimal feature information can accurately reflect the spectral differences of different ground object categories, fundamentally improving the accuracy of classification, and making the segmentation line more close to the actual boundary.

Description

Technical Field

[0001] The present invention relates to the technical field of remote sensing image processing, and more specifically, to a remote sensing image ground object information intelligent extraction and analysis system. Background Art

[0002] Remote sensing image change detection is an important technology used to observe surface changes. This technology has been widely used in many key fields such as global land use monitoring, ecological environment monitoring, natural disaster emergency response, and natural resource census. Remote sensing images are the core data source for systematic and dynamic description of land surface evolution. Their wide spatial coverage and large changes in spatial resolution and spectral resolution have greatly promoted the development of applications such as environment and land use monitoring. In the existing technology, there has been a method for processing remote sensing images based on spectral information, which can effectively analyze surface dynamics such as vegetation changes.

[0003] For example, the patent application with publication number CN107564016A provides a multi-band remote sensing image segmentation and labeling method that integrates the spectral information of ground objects. The patent calculates the apparent reflectance of the top of the atmosphere of the multi-band remote sensing image, extracts the spectral features and classifies the ground objects to obtain a classification result map. Then, the J-value image is segmented using the region growing method, and the classification category and proportion of the ground objects in each segmented area are counted. The patent application with publication number CN109191482A provides an image merging and segmentation method based on regional adaptive spectral angle thresholds. The patent uses spectral angle as an important and only input parameter for global gradient calculation, generation through segmentation, initial judgment of global ground object similarity, and generation of regional adaptive spectral angle thresholds.

[0004] Although existing technologies have performed segmentation processing on remote sensing images based on spectral information, in forest management scenarios, especially when monitoring the boundaries between different woods, the accuracy and precision of vegetation image classification vary due to different spectral features, and existing technologies are unable to select the optimal spectral features. In addition, for areas with blurred boundaries in vegetation images, the ability for accurate segmentation is weak, resulting in unsatisfactory results in monitoring vegetation changes.

[0005] In view of this, the present invention proposes a remote sensing image ground object information intelligent extraction and analysis system to solve the above problems. Summary of the invention

[0006] In order to overcome the above-mentioned defects of the prior art, the present invention provides a remote sensing image ground object information intelligent extraction and analysis system.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] In the first aspect, a remote sensing image ground object information intelligent extraction and analysis system is provided, comprising:

[0009] Information screening module: used to obtain initial ground object information and corresponding spectral feature information, and screen out M sub-feature information from the spectral feature information according to a step-by-step feature analysis method, wherein the initial ground object information is remote sensing image information;

[0010] Information optimization module: used to traverse the M sub-feature information, obtain the statistical coefficient corresponding to each sub-feature information, and determine the optimal feature information from the M sub-feature information based on the statistical coefficient and the preset intelligent optimization algorithm;

[0011] Information adjustment module: used to perform preliminary regional classification on the initial feature information according to the optimal feature information, obtain intermediate feature information, obtain the feature similarity set corresponding to the intermediate feature information, correct the intermediate feature information according to the feature similarity set, and obtain the target feature information.

[0012] Furthermore, the method of selecting M sub-feature information from the spectral feature information according to the step-by-step feature analysis method includes:

[0013] Randomly select N elements from the spectral feature information, combine the N elements into initial feature information, calculate the internal deviation matrix corresponding to the initial feature information, calculate the internal determinant corresponding to the internal deviation matrix, take the reciprocal of the internal determinant as the statistical coefficient, and judge whether the statistical coefficient is less than the preset coefficient threshold. If it is less, take the initial feature information as sub-feature information; if it is not less, discard the initial feature information, and repeat the above steps until M sub-feature information is screened out.

[0014] Furthermore, the method of calculating the internal deviation matrix corresponding to the initial feature information includes:

[0015] WSX= ;

[0016] Where WSX is the internal deviation matrix, is the total number of elements in the initial feature information, For the Category eigenvalues, For the Category The average value of the features, For the Category eigenvalues, No. Category The average value of the features.

[0017] Furthermore, the intelligent optimization algorithm is a whale optimization algorithm. The method for determining the optimal feature information from the M sub-feature information based on the statistical coefficient and the preset intelligent optimization algorithm includes:

[0018] Initialization unit: used to define the initial number of whales as Y and the maximum number of iterations , each initial whale individual is represented as a sub-feature information, and the random value rand is a random number sampled from a uniform distribution of [0, 1];

[0019] Fitness calculation unit: used to obtain the statistical coefficients corresponding to the sub-feature information and use the statistical coefficients as the target fitness values ​​of the initial whale individuals;

[0020] Strategy simulation unit: used to implement the shrinking and encircling strategy when the random value rand is less than 0.5, and to implement the spiral hunting strategy when the random value rand is greater than or equal to 0.5;

[0021] Loop output unit: used to repeat the above strategy simulation unit and continuously update the sub-feature information until the maximum number of iterations is reached When , the currently optimal sub-feature information is output, and the currently optimal sub-feature information is used as the optimal feature information.

[0022] Furthermore, the method of performing the shrinking and surrounding strategy includes:

[0023] ;

[0024] in, For the next generation of whale positions, is the current whale position, To pursue the target location, is the shrinkage coefficient of the shrinkage and encirclement strategy, is the control factor of the shrinking and encircling strategy, is the current iteration number.

[0025] Further, the method of performing the spiral hunting strategy includes:

[0026] = ;

[0027] in, is the spiral contraction coefficient of the spiral hunting strategy, A random value in the range [-1, 1].

[0028] Furthermore, the method of performing preliminary regional classification on the initial feature information according to the optimal feature information to obtain the intermediate feature information includes:

[0029] Input the optimal feature information and the initial object information into the pre-built regional classification model to obtain the intermediate object information;

[0030] The construction method of the regional classification model includes:

[0031] Obtain G groups of training data, where G is a positive integer greater than 1, and the training data includes historical optimal feature information, historical initial feature information, and historical intermediate feature information;

[0032] The historical optimal feature information, the historical initial feature information and the historical intermediate feature information are used as sample sets, the sample sets are divided into training sets and test sets, and a classifier is constructed;

[0033] The historical optimal feature information and historical initial feature information in the training set are used as input data, and the historical intermediate feature information in the training set is used as output data to train the classifier and obtain an initial classifier;

[0034] The initial classifier is tested using the test set, and the classifier that meets the preset accuracy is output as the regional classification model.

[0035] Furthermore, the method for obtaining the feature similarity set corresponding to the intermediate feature information includes:

[0036] Determine the dividing line in the intermediate feature information, obtain the maximum feature point in the dividing line, determine the minimum bounding box corresponding to the dividing line according to the maximum feature point, calculate the spectral feature similarity between each pixel point and the adjacent pixel points in the minimum bounding box, and integrate the spectral feature similarities corresponding to all pixels to obtain a feature similarity set.

[0037] Compared with the prior art, the present invention has the following beneficial effects:

[0038] The present invention firstly selects M sub-feature information from the spectral feature information according to the step-by-step feature analysis method, then obtains the statistical coefficient corresponding to each sub-feature information, determines the optimal feature information from the M sub-feature information based on the statistical coefficient and the preset intelligent optimization algorithm, and finally performs preliminary regional classification on the initial object information according to the optimal feature information to obtain intermediate object information, obtains the feature similarity set corresponding to the intermediate object information, corrects the intermediate object information according to the feature similarity set, and obtains the target object information. In this way, the present invention selects the sub-feature information with the best distinguishing ability through the step-by-step feature analysis method, and further optimizes it using the intelligent optimization algorithm to ensure that the extracted optimal feature information can accurately reflect the spectral differences of different object categories, thereby fundamentally improving the accuracy of classification. Then, by adjusting the segmentation line of the initial object information and correcting the feature similarity set, the problem of blurred boundaries between different forests and low classification accuracy in the prior art is effectively solved, so that the segmentation line is more in line with the actual boundary. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 It is a structural schematic diagram of a remote sensing image ground object information intelligent extraction and analysis system in the present invention;

[0040] Figure 2 A schematic diagram of a process for intelligent extraction and analysis of ground object information from remote sensing images in the present invention;

[0041] Figure 3 is a schematic diagram of the spectral reflectance of different tree species in the present invention;

[0042] Figure 4 is a schematic diagram of initial ground feature information in the present invention;

[0043] Figure 5 is a schematic diagram of the intermediate feature information of the present invention;

[0044] Figure 6 A schematic diagram of a segmentation line, a minimum bounding box, and a maximum feature point in the present invention;

[0045] Reference numerals:

[0046] 101, dividing line; 102, minimum bounding box; 103, maximum feature point. DETAILED DESCRIPTION

[0047] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0048] Example 1

[0049] See also Figure 1 As shown, this embodiment discloses a remote sensing image ground object information intelligent extraction and analysis system, including:

[0050] Information screening module: used to obtain initial ground object information and corresponding spectral feature information, and screen out M sub-feature information from the spectral feature information according to a step-by-step feature analysis method, wherein the initial ground object information is remote sensing image information;

[0051] In this embodiment, the initial ground object information refers to remote sensing image information, which can be a tree vegetation image obtained by equipment such as drones and aircraft. Specifically, the obtained tree vegetation image contains at least two different types of tree communities. The spectral feature information refers to the reflection data of each band extracted from the remote sensing image, which can characterize the spectral characteristics of different ground object types. By analyzing the spectral feature information, different types of tree communities can be further identified and classified.

[0052] It should be added that the spectral feature information at least includes reflectivity or central wavelength. Reflectivity refers to the ability of the surface of the object to reflect incident light in a specific band, usually expressed as a ratio or percentage, reflecting the reflection intensity of the object to light of different wavelengths. The central wavelength refers to the representative wavelength value of each spectral band in the remote sensing image, usually the frequency with the strongest spectral response in the band, such as Figure 3 As shown, Figure 3 The reflectivity and central wavelength of trees such as Rhizophora mangrove, Caryophyllaceae, Sea lotus, Kandelia candel, Prunus elata and Sonneratia apetala in specific bands are shown. It can be understood that the above-mentioned trees usually belong to a forest ecosystem. Therefore, when forest management is required, the boundaries between different forests must be distinguished. Since the reflectivity and central wavelength corresponding to the above-mentioned trees are different, the reflectivity and central wavelength play an important role in this process, which helps to accurately delineate the boundaries of forests and ensure the accuracy and effectiveness of forest management.

[0053] It should be noted that the stepwise feature analysis method refers to a statistical analysis method, which is mainly used to screen out the most effective features for classification tasks from a large number of features. In the analysis of remote sensing images, the stepwise feature analysis method is used to select the most discriminative spectral feature information to help classify land objects (such as tree species classification).

[0054] The method for selecting M sub-feature information from spectral feature information according to the step-by-step feature analysis method includes:

[0055] Randomly select N elements from the spectral feature information, combine the N elements into initial feature information, calculate the internal deviation matrix corresponding to the initial feature information, calculate the internal determinant corresponding to the internal deviation matrix, take the reciprocal of the internal determinant as the statistical coefficient, and judge whether the statistical coefficient is less than the preset coefficient threshold. If it is less, take the initial feature information as sub-feature information; if it is not less, discard the initial feature information, and repeat the above steps until M sub-feature information is screened out.

[0056] It can be understood that the elements in the spectral feature information represent the reflectance value or other spectral characteristic value corresponding to each spectral band. Taking the reflectance value as an example, for example, the reflectance value corresponding to the red light band of one kind of tree is 0.2, and the reflectance value corresponding to the near-infrared band is 0.6, while the reflectance value corresponding to the red light band of the other kind of tree is 0.28, and the reflectance value corresponding to the near-infrared band is 0.35. Therefore, selecting the reflectance value corresponding to the near-infrared band makes it easier to distinguish the two trees. The spectral bands can be red light, green light, blue light, and near-infrared bands, etc. This embodiment will not go into details.

[0057] The method of calculating the internal deviation matrix corresponding to the initial feature information includes:

[0058] WSX= ;

[0059] Where WSX is the internal deviation matrix, is the total number of elements in the initial feature information, For the Category eigenvalues, For the Category The average value of the features, For the Category eigenvalues, No. Category The average value of the features.

[0060] It can be understood that the purpose of calculating the statistical coefficient in this embodiment is to screen useful band information according to the statistical coefficient and complete the final division according to the band information. Therefore, N elements are arbitrarily selected from the spectral feature information, and each element corresponds to a band, such as a near-infrared band, a blue light band, and a green light band. Each band has at least one characteristic value sequence. The above implementation method is described as follows in combination with specific application scenarios:

[0061] For example, in this embodiment, the tree vegetation image of the Rhizophora mangrove tree group and the Anthocarpus serrata tree group is divided, and the spectral bands corresponding to the spectral characteristic information include the red light band, the green light band, the blue light band and the near infrared band. is 4. For the red mangrove tree group, the reflectance and central wavelength corresponding to each spectral band are:

[0062]

[0063] Then for the group of hornberry trees, the reflectance and central wavelength corresponding to each spectral band are:

[0064]

[0065] So if =1, it means that in the red light band, The first feature can be reflectivity, The characteristic can be the central wavelength, then the corresponding It can be 0.24, It can be 630nm. It is understandable that and It is detected and obtained in real time. For example, in a pixel point or pixel area in a tree vegetation image, the corresponding spectral feature information is as follows:

[0066]

[0067] Now, to classify the above pixel points or pixel areas, and distinguish whether they belong to the Rhizophora mangrove tree group or the Anthocarpus hortensis tree group, we need to calculate the corresponding internal deviation matrix. The calculation of the internal deviation matrix is ​​as follows:

[0068] Then for the red light band, the corresponding internal sub-bias matrix is:

[0069] ;

[0070] Then the internal deviation matrix in the above-mentioned is composed of N internal sub-deviation matrices arranged in order. Calculating the internal determinant corresponding to the internal deviation matrix is ​​a prior art, and this embodiment will not go into details about it.

[0071] In this embodiment, the main purpose of calculating the internal deviation matrix is ​​to measure the variability or discreteness within the spectral feature information. Samples of different categories have different performances on the same feature (such as reflectance). In remote sensing images, each spectral band may have different distinguishing capabilities for different types of land objects (such as tree communities). By calculating the internal deviation matrix of each spectral band, the spectral band with the most classification efficiency can be identified, thereby selecting the most effective features for the classification task. It can be understood that in this embodiment, the internal determinant corresponding to the internal deviation matrix is ​​calculated first. The size of the internal determinant is directly related to the classification efficiency. The larger the internal determinant, the greater the variability of the feature values ​​of samples within the same category, which helps to distinguish samples of different categories. Therefore, the feature (band) has a stronger classification ability.

[0072] Information optimization module: used to traverse the M sub-feature information, obtain the statistical coefficient corresponding to each sub-feature information, and determine the optimal feature information from the M sub-feature information based on the statistical coefficient and the preset intelligent optimization algorithm;

[0073] In this embodiment, obtaining the statistical coefficients corresponding to the sub-feature information has been described in the above content, and will not be described in detail again.

[0074] The intelligent optimization algorithm is a whale optimization algorithm. The method for determining the optimal feature information from M sub-feature information based on statistical coefficients and a preset intelligent optimization algorithm includes:

[0075] Initialization unit: used to define the initial number of whales as Y and the maximum number of iterations , each initial whale individual is represented as a sub-feature information, and the random value rand is a random number sampled from a uniform distribution of [0, 1];

[0076] It is worth noting that each initial whale individual represents a sub-feature information. The number of initial whale groups determines the search range of the optimization process. A larger group can cover a wider solution space and increase the possibility of finding the optimal feature information.

[0077] Fitness calculation unit: used to obtain the statistical coefficients corresponding to the sub-feature information and use the statistical coefficients as the target fitness values ​​of the initial whale individuals;

[0078] It can be understood that the statistical coefficient corresponds to the "fitness value" in the whale optimization algorithm, which represents the "survival ability" of an individual in the entire group. In the whale optimization algorithm, each individual (whale) has a fitness value to measure the quality of the individual. The smaller the fitness value, the closer the individual is to the optimal solution.

[0079] Strategy simulation unit: used to implement the shrinking and encircling strategy when the random value rand is less than 0.5, and to implement the spiral hunting strategy when the random value rand is greater than or equal to 0.5;

[0080] Among them, the shrinking and encircling strategy is implemented, and the whale will shrink and encircle around the optimal solution. The spiral hunting strategy is implemented to simulate the whale moving around the target in a spiral path.

[0081] Loop output unit: used to repeat the above strategy simulation unit and continuously update the sub-feature information until the maximum number of iterations is reached When , the currently optimal sub-feature information is output, and the currently optimal sub-feature information is used as the optimal feature information.

[0082] In this embodiment, in each iteration, the whale group is continuously updated until the maximum number of iterations is reached. After each iteration, the algorithm will update the weight factor combination of individual whales and calculate the new fitness value. Through repeated iterations, the weight factor combination of individual whales is gradually improved each time, so that the whale group gradually approaches the optimal solution.

[0083] Methods for implementing the shrinking and encircling strategy include:

[0084] ;

[0085] in, For the next generation of whale positions, is the current whale position, To pursue the target location, is the shrinkage coefficient of the shrinkage and encirclement strategy, is the control factor of the shrinking and encircling strategy, is the current iteration number.

[0086] This embodiment allows individual whales to gradually shrink when approaching the optimal solution, accurately adjust the weight factor, and perform local optimization. Through this strategy, individual whales gradually approach the current optimal solution, ensuring a detailed search in the solution space.

[0087] Methods for conducting a spiral hunting strategy include:

[0088] = ;

[0089] in, is the spiral contraction coefficient of the spiral hunting strategy, A random value in the range [-1, 1].

[0090] This embodiment first screens out the sub-feature information with the most classification ability in the spectral feature information according to the step-by-step feature analysis method, and then uses the intelligent optimization algorithm to further optimize these sub-feature information to obtain the optimal feature information. The combination of the two forms a closed-loop process from global screening to local fine optimization, thereby significantly improving the accuracy and efficiency of the classification task and ensuring that the extraction and analysis results of the initial ground object information are more reliable and accurate.

[0091] Information adjustment module: used to perform preliminary regional classification on the initial feature information according to the optimal feature information, obtain intermediate feature information, obtain a feature similarity set corresponding to the intermediate feature information, correct the intermediate feature information according to the feature similarity set, and obtain target feature information;

[0092] The method of performing preliminary regional classification on the initial feature information according to the optimal feature information to obtain the intermediate feature information includes:

[0093] Input the optimal feature information and the initial object information into the pre-built regional classification model to obtain the intermediate object information;

[0094] The construction method of the regional classification model includes:

[0095] Obtain G groups of training data, where G is a positive integer greater than 1, and the training data includes historical optimal feature information, historical initial feature information, and historical intermediate feature information;

[0096] The historical optimal feature information, the historical initial feature information and the historical intermediate feature information are used as sample sets, the sample sets are divided into training sets and test sets, and a classifier is constructed;

[0097] The historical optimal feature information and historical initial feature information in the training set are used as input data, and the historical intermediate feature information in the training set is used as output data to train the classifier and obtain an initial classifier;

[0098] The initial classifier is tested using the test set, and a classifier that meets the preset accuracy is output as the regional classification model. The classifier can be a deep neural network model.

[0099] It should be added that, from the above content, it can be known that the initial feature information refers to remote sensing image data, which can be tree and vegetation images obtained by equipment such as drones and aircraft. Similarly, the intermediate feature information is also tree and vegetation images, but unlike the initial feature information, the intermediate feature information is classified and divided into boundaries, such as Figure 4 and Figure 5 As shown, Figure 4 The initial feature information is shown. Figure 5 The intermediate feature information is shown.

[0100] The method for obtaining the feature similarity set corresponding to the intermediate feature information includes:

[0101] Determine the dividing line in the intermediate feature information, obtain the maximum feature point in the dividing line, determine the minimum bounding box corresponding to the dividing line according to the maximum feature point, calculate the spectral feature similarity between each pixel point in the minimum bounding box and the adjacent pixel points, and arrange the spectral feature similarities corresponding to all pixels in sequence to obtain a feature similarity set.

[0102] like Figure 6 As shown, Figure 6 A dividing line 101, a minimum bounding box 102, and a maximum feature point 103 are shown. It can be understood that the maximum feature point can be the point on both sides of the dividing line that is farthest from the center line. The center line refers to the line connecting the two end points of the dividing line. Then the minimum bounding box refers to a rectangular area with the smallest area defined around the dividing line and its maximum feature point in the intermediate feature information, and the rectangular area can completely surround the dividing line.

[0103] It should be added that the calculation of the spectral feature similarity between each pixel point and the adjacent pixel points in the minimum bounding box may be to calculate the cosine similarity between each pixel point and the adjacent pixel points, and use the cosine similarity as the spectral feature similarity, exemplarily:

[0104] The reflectivities corresponding to one pixel are 0.2, 0.15, 0.1 and 0.4, and the reflectivities corresponding to the adjacent pixels are 0.28, 0.2, 0.15 and 0.35, respectively. Then, the cosine similarity between the pixel and the adjacent pixels can be calculated by calculating the cosine value between 0.2 and 0.28, the cosine value between 0.15 and 0.2, etc., and then finding the average value as the cosine similarity. The same is true for the central wavelength, which will not be described in detail in this embodiment.

[0105] In this embodiment, the purpose of calculating the spectral feature similarity is to adjust the dividing line. The spectral feature refers to the reflection characteristics of the ground object to light of different bands, which is expressed in the form of reflectivity or central wavelength of a specific band. Different types of ground objects (such as trees, vegetation, water bodies, etc.) have different reflection behaviors to light due to different materials, structures, and chemical compositions. The two sides of the dividing line usually represent the boundaries of two or more types of ground objects. By calculating the spectral feature similarity, the similarity or difference of these ground objects in the spectral dimension can be quantified, so as to determine whether the dividing line accurately reflects the actual boundary of the ground object type. It can be understood that when the spectral feature similarity is greater, it indicates that the spectral characteristics of the ground objects on both sides of the dividing line are less different, and the probability of belonging to the same ground object type is greater. The smaller the spectral feature similarity, it indicates that the spectral characteristics of the ground objects on both sides of the dividing line are more different, and the probability of belonging to the same ground object type is smaller.

[0106] It can be understood that the process of correcting the intermediate land object information based on the feature similarity set is specifically achieved by adjusting the dividing line in the intermediate land object information. The target land object information refers to the remote sensing image information generated after the dividing line adjustment based on the feature similarity set, which can more accurately reflect the actual distribution of land object categories.

[0107] For example, when the similarity of the spectral characteristics of a certain pixel is lower than the preset similarity threshold, it indicates that the spectral characteristics of the pixel and the adjacent pixels are greatly different. At this time, it can be inferred that the pixel may not belong to the area of ​​the current feature category. Therefore, it is necessary to readjust the position of the segmentation line so that the segmentation line can more accurately divide the boundaries of different feature categories and classify the pixel into a feature category area that is more consistent with its spectral characteristics. The specific adjustment method can be:

[0108] Mark the pixels below the preset similarity threshold, regard these low-similarity pixels as "outliers", mark their distribution positions as the basis for adjusting the segmentation line, and recalculate the position of the segmentation line according to the position and distribution characteristics of the outliers to ensure that the segmentation line passes through the area with the largest difference in spectral characteristics (i.e., category boundaries) and maximizes the consistency of the spectral characteristics of the ground object categories on both sides.

[0109] In this embodiment, M sub-feature information is first screened out from the spectral feature information according to the step-by-step feature analysis method, and then the statistical coefficient corresponding to each sub-feature information is obtained. Based on the statistical coefficient and the preset intelligent optimization algorithm, the optimal feature information is determined from the M sub-feature information. Finally, the initial land object information is preliminarily regionally classified according to the optimal feature information to obtain intermediate land object information, and the feature similarity set corresponding to the intermediate land object information is obtained. The intermediate land object information is corrected according to the feature similarity set to obtain the target land object information. In this way, this embodiment selects the sub-feature information with the best distinguishing ability through the step-by-step feature analysis method, and further optimizes it using the intelligent optimization algorithm to ensure that the extracted optimal feature information can accurately reflect the spectral differences of different land object categories, thereby fundamentally improving the accuracy of classification. Then, by adjusting the segmentation line of the initial land object information and correcting the feature similarity set, the problem of blurred boundaries between different forests and low classification accuracy in the prior art is effectively solved, so that the segmentation line is more in line with the actual boundary.

[0110] Example 2

[0111] See also Figure 2 As shown, based on the same inventive concept, this embodiment discloses a method for intelligent extraction and analysis of ground object information in remote sensing images. For details not provided in this embodiment, please refer to the description of the relevant parts in Embodiment 1. The method includes:

[0112] S10: obtaining initial ground object information and corresponding spectral feature information, and selecting M sub-feature information from the spectral feature information according to a step-by-step feature analysis method, wherein the initial ground object information is remote sensing image information;

[0113] In this embodiment, the initial ground object information refers to remote sensing image data, which can be tree vegetation images acquired by equipment such as drones and aircraft. Specifically, the acquired tree vegetation images contain at least two different types of tree communities. The spectral feature information refers to the reflectance data of each band extracted from the remote sensing image, which can characterize the spectral characteristics of different ground object types. By analyzing the spectral feature information, different types of tree communities can be further identified and classified.

[0114] The method for selecting M sub-feature information from spectral feature information according to the step-by-step feature analysis method includes:

[0115] Randomly select N elements from the spectral feature information, combine the N elements into initial feature information, calculate the internal deviation matrix corresponding to the initial feature information, calculate the internal determinant corresponding to the internal deviation matrix, take the reciprocal of the internal determinant as the statistical coefficient, and judge whether the statistical coefficient is less than the preset coefficient threshold. If it is less, take the initial feature information as sub-feature information; if it is not less, discard the initial feature information, and repeat the above steps until M sub-feature information is screened out.

[0116] The method of calculating the internal deviation matrix corresponding to the initial feature information includes:

[0117] WSX= ;

[0118] Where WSX is the internal deviation matrix, is the total number of elements in the initial feature information, For the Category eigenvalues, For the Category The average value of the features, For the Category eigenvalues, No. Category The average value of the features.

[0119] S20: traverse the M sub-feature information, obtain the statistical coefficient corresponding to each sub-feature information, and determine the optimal feature information from the M sub-feature information based on the statistical coefficient and a preset intelligent optimization algorithm;

[0120] S30: performing preliminary regional classification on the initial feature information according to the optimal feature information to obtain intermediate feature information, obtaining a feature similarity set corresponding to the intermediate feature information, and correcting the intermediate feature information according to the feature similarity set to obtain target feature information;

[0121] The method of performing preliminary regional classification on the initial feature information according to the optimal feature information to obtain the intermediate feature information includes:

[0122] Input the optimal feature information and the initial object information into the pre-built regional classification model to obtain the intermediate object information;

[0123] The construction method of the regional classification model includes:

[0124] Obtain G groups of training data, where G is a positive integer greater than 1, and the training data includes historical optimal feature information, historical initial feature information, and historical intermediate feature information;

[0125] The historical optimal feature information, the historical initial feature information and the historical intermediate feature information are used as sample sets, the sample sets are divided into training sets and test sets, and a classifier is constructed;

[0126] The historical optimal feature information and historical initial feature information in the training set are used as input data, and the historical intermediate feature information in the training set is used as output data to train the classifier and obtain an initial classifier;

[0127] The initial classifier is tested using the test set, and the classifier that meets the preset accuracy is output as the regional classification model.

[0128] The method for obtaining the feature similarity set corresponding to the intermediate feature information includes:

[0129] Determine the dividing line in the intermediate feature information, obtain the maximum feature point in the dividing line, determine the minimum bounding box corresponding to the dividing line according to the maximum feature point, calculate the spectral feature similarity between each pixel point and the adjacent pixel points in the minimum bounding box, and integrate the spectral feature similarities corresponding to all pixels to obtain a feature similarity set.

[0130] The above embodiments may be implemented in whole or in part by software, hardware, firmware or any other combination thereof. When implemented by software, the above embodiments may be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the process or function described in the embodiment of the present invention is generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network or other programmable device. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions may be transmitted from one website, computer, server or data center to another website, computer, server or data center via a wired network or a wireless network. The computer-readable storage medium may be any available medium that can be accessed by a computer or a data storage device such as a server or data center that includes one or more available media sets. The available medium may be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD) or a semiconductor medium. The semiconductor medium may be a solid-state hard disk.

[0131] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed in the present invention can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0132] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0133] In the several embodiments provided by the present invention, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of the units is only one, and there may be other division methods 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 through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0134] 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.

[0135] 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.

[0136] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

[0137] Finally: 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 principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A remote sensing image ground object information intelligent extraction and analysis system, characterized in that: include: Information screening module: used to obtain initial ground object information and corresponding spectral feature information, and screen out M sub-feature information from the spectral feature information according to a step-by-step feature analysis method, wherein the initial ground object information is remote sensing image information; Information optimization module: used to traverse the M sub-feature information, obtain the statistical coefficient corresponding to each sub-feature information, and determine the optimal feature information from the M sub-feature information based on the statistical coefficient and the preset intelligent optimization algorithm; Information adjustment module: used to perform preliminary regional classification on the initial feature information according to the optimal feature information, obtain intermediate feature information, obtain a feature similarity set corresponding to the intermediate feature information, correct the intermediate feature information according to the feature similarity set, and obtain target feature information; The method for selecting M sub-feature information from spectral feature information according to the step-by-step feature analysis method comprises: Randomly select N elements from the spectral feature information, combine the N elements into initial feature information, calculate the internal deviation matrix corresponding to the initial feature information, calculate the internal determinant corresponding to the internal deviation matrix, use the reciprocal of the internal determinant as the statistical coefficient, and judge whether the statistical coefficient is less than a preset coefficient threshold. If it is less than, use the initial feature information as sub-feature information; if it is not less than, remove the initial feature information, and repeat the above steps until M sub-feature information is screened out; The method for calculating the internal deviation matrix corresponding to the initial feature information includes: Where WSX is the internal deviation matrix, N is the total number of elements in the initial feature information, and x q (i) is the i-th eigenvalue of the q-th category, is the average value of the ith feature of the qth category, x q (j) is the j-th eigenvalue of the q-th category, The average value of the j-th feature of the q-th class.

2. The remote sensing image ground object information intelligent extraction and analysis system according to claim 1 is characterized in that: The intelligent optimization algorithm is a whale optimization algorithm. The method for determining the optimal feature information from M sub-feature information based on statistical coefficients and a preset intelligent optimization algorithm includes: Initialization unit: used to define the initial number of whales as Y and the maximum number of iterations as T max , each initial whale individual is represented as a sub-feature information, and the random value rand is a random number sampled from a uniform distribution of [0, 1]; Fitness calculation unit: used to obtain the statistical coefficients corresponding to the sub-feature information and use the statistical coefficients as the target fitness values ​​of the initial whale individuals; Strategy simulation unit: used to implement the shrinking and encircling strategy when the random value rand is less than 0.5, and to implement the spiral hunting strategy when the random value rand is greater than or equal to 0.5; Loop output unit: used to repeat the above strategy simulation unit and continuously update the sub-feature information until the maximum number of iterations T is reached max When , the currently optimal sub-feature information is output, and the currently optimal sub-feature information is used as the optimal feature information.

3. The remote sensing image ground object information intelligent extraction and analysis system according to claim 2 is characterized in that: The method for performing the shrinking and surrounding strategy comprises: Among them, X(t+1) is the next generation whale position, X(t) is the current whale position, and X * (t) is the position of the target, α is the shrinkage coefficient of the shrinkage strategy, C is the control factor of the shrinkage strategy, T current is the current iteration number.

4. The remote sensing image ground object information intelligent extraction and analysis system according to claim 3 is characterized in that: The method for performing the spiral hunting strategy comprises: X(t+1)=|X*(t)-X(t)|×e bL ×Cos(2πL)+X*(t); Where b is the spiral contraction coefficient of the spiral hunting strategy, and L is a random value in the range of [-1, 1].

5. The remote sensing image ground object information intelligent extraction and analysis system according to claim 4 is characterized in that: The method of performing preliminary regional classification on the initial object information according to the optimal feature information to obtain the intermediate object information comprises: Input the optimal feature information and the initial object information into the pre-built regional classification model to obtain the intermediate object information; The construction method of the regional classification model includes: Obtain G groups of training data, where G is a positive integer greater than 1, and the training data includes historical optimal feature information, historical initial feature information, and historical intermediate feature information; The historical optimal feature information, the historical initial feature information and the historical intermediate feature information are used as sample sets, the sample sets are divided into training sets and test sets, and a classifier is constructed; The historical optimal feature information and historical initial feature information in the training set are used as input data, and the historical intermediate feature information in the training set is used as output data to train the classifier and obtain an initial classifier; The initial classifier is tested using the test set, and the classifier that meets the preset accuracy is output as the regional classification model.

6. The remote sensing image ground object information intelligent extraction and analysis system according to claim 5, characterized in that: The method for obtaining the feature similarity set corresponding to the intermediate feature information includes: Determine the dividing line in the intermediate feature information, obtain the maximum feature point in the dividing line, determine the minimum bounding box corresponding to the dividing line according to the maximum feature point, calculate the spectral feature similarity between each pixel point and the adjacent pixel points in the minimum bounding box, and integrate the spectral feature similarities corresponding to all pixels to obtain a feature similarity set.

Images