Forestry Information Identification Method and System Based on Remote Sensing Data

Abstract

This application relates to the field of forestry information recognition technology, and discloses a forestry information recognition method and system based on remote sensing data. The method includes: S1. According to the geographical coordinates of the target range stored in advance, segment the target remote sensing image from the remote sensing image; S2. Distinguish the typical range and the atypical range in the target remote sensing image, and determine different object areas in the typical range; S3. Divide the image elements in the object area whose image element values are greater than the average image element value into the first object sub-area, divide the image elements in the object area whose image element values are less than or equal to the average image element value into the second object sub-area, and divide the image elements with the largest image element value in the object area into the third object sub-area; S4. Use the recognition result with the largest total number of corresponding image elements as the final recognition result of the object area where the first object sub-area is located. This application can improve the recognition accuracy.

Description

Technical Field

[0001] This application relates to the technical field of forestry information recognition, and particularly to a method and system for forestry information recognition based on remote sensing data. Background Art

[0002] Currently, methods for recognizing forestry information based on remote sensing data are being used more and more. Forestry information includes the types of trees in the forest. The automatic recognition of forestry information can provide strong technical support for the scientific management and sustainable utilization of forestry resources.

[0003] Chinese invention application with publication number CN111582176A discloses a visible light remote sensing image dead tree recognition software system and recognition method. The system includes: a tree remote sensing image acquisition module, a central control module, a remote sensing image correction module, a remote sensing image enhancement module, a remote sensing image segmentation module, an image feature extraction module, a dead tree recognition module, a dead tree classification module, a cloud storage module, and a display module. The remote sensing image correction module unifies the coordinate systems of multiple remote sensing images, uses the coordinates of the real positions on the ground as standard points, and corrects the coordinate points in the remote sensing image according to the corresponding ground tree positions. The remote sensing image segmentation module combines the deep learning method with the recognition of ground objects in the tree remote sensing image to achieve the segmentation of ground objects in the remote sensing image. In addition, Chinese invention application with publication number CN108932455A provides a remote sensing image scene recognition method and device. The method includes: extracting the deep features of the remote sensing image based on a pre-trained deep convolutional neural network, extracting the SIFT features of the remote sensing image. The SIFT feature refers to the feature extracted by the scale-invariant feature transform algorithm. According to the SIFT features and the deep features, the scene type of the remote sensing image is determined. Since the SIFT features have scale invariance and rotation invariance, when recognizing the scene type of the remote sensing image, the problem that the deep features are sensitive to rotation transformation or scale transformation of the remote sensing image can be overcome. However, the recognition accuracy of the recognition methods in the above-mentioned invention applications still needs to be improved. Summary of the Invention

[0004] This application extracts the target remote sensing image from the remote sensing image, distinguishes the typical range and the atypical range in the target remote sensing image, also determines different object regions in the typical range, divides the sub-regions of each object region, determines the recognition results corresponding to each image element in the first object sub-region, and takes the recognition result with the largest total number of corresponding image elements as the final recognition result of the object region. This application aims to improve the recognition accuracy of the recognition method for recognizing forestry information.

[0005] This application provides a method for recognizing forestry information based on remote sensing data, mainly including the following steps:

[0006] S1. Take aerial photos of the target area to obtain a remote sensing image, determine the geographical coordinates of each image element in the remote sensing image, and also segment from the remote sensing image a target remote sensing image that only contains the image of the target area according to the geographical coordinates of the target area stored in advance;

[0007] S2. Process the target remote sensing image using a pre-trained classification model to distinguish a typical area and an atypical area in the target remote sensing image, and also determine different object areas in the typical area of the target remote sensing image;

[0008] S3. For each of the object areas in the typical area of the target remote sensing image, calculate the average value of the image element values of all the image elements in the object area to obtain an average image element value, divide the image elements in the object area whose image element values are greater than the average image element value into a first object sub-area, divide the image elements in the object area whose image element values are less than or equal to the average image element value into a second object sub-area, and divide the image element with the largest image element value in the object area into a third object sub-area;

[0009] S4. For each of the first object sub-areas in each of the object areas in the typical area of the target remote sensing image, respectively determine the recognition results corresponding to different image elements in the first object sub-area, and take the recognition result with the largest total number of corresponding image elements as the final recognition result of the object area where the first object sub-area is located.

[0010] As a preferred technical solution of the present application, in S4, determining the recognition results corresponding to the image elements in the first object sub-area includes: generating a characteristic curve from the image elements, the abscissa corresponding to the characteristic curve represents the increase in wavelength from left to right, the ordinate corresponding to the characteristic curve represents the increase in image element value from bottom to top, and comparing the characteristic curve with different pre-stored standard characteristic curves to determine the recognition result corresponding to the image element.

[0011] As a preferred technical solution of the present application, in the typical area of the target remote sensing image, determining different object areas includes the following steps:

[0012] S211. Perform smoothing processing on the typical area of the target remote sensing image to obtain the smoothed typical area of the target remote sensing image;

[0013] S212. For each image element in the smoothed typical area of the target remote sensing image, subtract a preset first value from the image element value of the image element to obtain the new typical area of the target remote sensing image;

[0014] S213. Determine an image element from the typical range of the new target remote sensing image, and determine whether the image element value of the image element is less than the maximum image element value corresponding to a preset number of surrounding image elements. If the answer is no, jump to S214. If the answer is yes, continue to determine whether the maximum image element value is greater than the previous image element value of the image element. If the answer is no, update the image element value of the image element with the maximum image element value. If the answer is yes, update the image element value of the image element with the previous image element value;

[0015] S214. Determine whether all the image elements in the typical range of the new target remote sensing image have been processed. If the answer is yes, end all steps and obtain the final typical range of the target remote sensing image. If the answer is no, jump to S213.

[0016] As a preferred technical solution of the present application, in the typical range of the target remote sensing image, to determine different object regions, the following steps are further included:

[0017] S221. For each image element in the typical range of the final target remote sensing image, increase the image element value of the image element by a preset second value;

[0018] S222. Determine an image element from the typical range of the final target remote sensing image, and determine whether the image element value of the image element is greater than the minimum image element value corresponding to a preset number of surrounding image elements. If the answer is no, jump to S223. If the answer is yes, continue to determine whether the minimum image element value is greater than or equal to the previous image element value of the image element. If the answer is yes, update the image element value of the image element with the minimum image element value. If the answer is no, update the image element value of the image element with the previous image element value;

[0019] S223. Determine whether all the image elements in the typical range of the final target remote sensing image have been processed. If the answer is yes, determine different object regions and end all steps. If the answer is no, jump to S222.

[0020] As a preferred technical solution of the present application, after the step S3, the following steps are further included: for each object area in the typical range of the target remote sensing image, generating a plurality of feature matrices of the object area, respectively generating a plurality of first eigenvalues and a plurality of second eigenvalues based on the plurality of feature matrices, using the plurality of first eigenvalues to form a first eigenvector, using the plurality of second eigenvalues to form a second eigenvector, and calculating the similarity score between the object area and different standard object areas respectively by using the formula Γ = ω*A + η*B, where Γ is the similarity score, A and B are the first similarity score and the second similarity score between the first eigenvector and the second eigenvector and the standard first eigenvector and the standard second eigenvector respectively, ω and η are the weights of the first similarity score and the second similarity score respectively, and taking the recognition result corresponding to the standard object area corresponding to the maximum similarity score as the final recognition result of the object area.

[0021] As a preferred technical solution of the present application, the first eigenvalue is generated based on the feature matrix, which is realized by the following formula:

[0022]

[0023] where Π is the first eigenvalue, n is the number of rows of the feature matrix, m is the number of columns of the feature matrix, and δ (n,m) is the element value at the position of the nth row and the mth column of the feature matrix.

[0024] As a preferred technical solution of the present application, the second eigenvalue is generated based on the feature matrix, which is realized by the following formula:

[0025] Ω = ΣΣ((n - m) 2 + 1) -1 *δ (n,m) )

[0026] where Ω is the second eigenvalue, n is the number of rows of the feature matrix, m is the number of columns of the feature matrix, and δ (n,m) is the element value at the position of the nth row and the mth column of the feature matrix.

[0027] The present application also provides a forestry information recognition system based on remote sensing data, which mainly includes the following modules:

[0028] A preprocessing module, which is used to take aerial photos of the target range to obtain a remote sensing image, determine the geographical coordinates of each image element in the remote sensing image, and also segment out the target remote sensing image containing the image of the target range from the remote sensing image according to the geographical coordinates of the target range stored in advance;

[0029] A classification module, configured to process a target remote sensing image using a pre-trained classification model to distinguish a typical range and an atypical range in the target remote sensing image, and further determine different object regions in the typical range of the target remote sensing image;

[0030] A partitioning module, configured to calculate, for each object region in the typical range of the target remote sensing image, an average value of the image element values of all the image elements in the object region to obtain an average image element value, partition the image elements in the object region whose image element values are greater than the average image element value into a first object sub-region, partition the image elements in the object region whose image element values are less than or equal to the average image element value into a second object sub-region, and partition the image element with the largest image element value in the object region into a third object sub-region;

[0031] An identification module, configured to respectively determine, for each first object sub-region in each object region in the typical range of the target remote sensing image, identification results corresponding to different image elements in the first object sub-region, and use the identification result with the largest total number of corresponding image elements as the final identification result of the object region where the first object sub-region is located.

[0032] The present application has at least the following beneficial effects:

[0033] In the technical solution provided by the present application, first, according to the geographical coordinates of the target range stored in advance, a target remote sensing image containing only the image of the target range is segmented from the remote sensing image. Secondly, the target remote sensing image is processed using a classification model to distinguish a typical range and an atypical range in the target remote sensing image, and different object regions are also determined in the typical range. Thirdly, the image elements in the object region whose image element values are greater than the average image element value are partitioned into a first object sub-region, the image elements in the object region whose image element values are less than or equal to the average image element value are partitioned into a second object sub-region, and the image element with the largest image element value in the object region is partitioned into a third object sub-region. Finally, the identification results corresponding to different image elements in the first object sub-region are determined, and the identification result with the largest total number of corresponding image elements is used as the final identification result of the object region where the first object sub-region is located. Through the above method, the present application can not only quickly and simply determine the identification result of the object region as forestry information, but also improve the identification accuracy of the identification method for identifying forestry information. Description of the Drawings

[0034] To more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the accompanying drawings required for the description of the embodiments. Obviously, the accompanying drawings in the following description are some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can also be obtained based on these drawings.

[0035] Figure 1 It is a flowchart of the forestry information recognition method based on remote sensing data in the embodiments of the present application;

[0036] Figure 2 It is a structural diagram of the forestry information recognition system based on remote sensing data in the embodiments of the present application. Specific embodiments

[0037] The embodiments of the present application provide a forestry information recognition method and system based on remote sensing data. Terms such as "first", "second", "third", "fourth", etc. (if any) in the specification, claims and above-mentioned accompanying drawings of the present application are used to distinguish similar objects, and do not have to be used to describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances so that the embodiments described here can be implemented in an order different from that shown or described here. In addition, the terms "comprising" or "having" and any variation thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device including a series of steps or units does not have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0038] For ease of understanding, please refer to Figure 1 , the forestry information recognition method based on remote sensing data in the embodiments of the present application mainly includes the following steps:

[0039] S1. Take an aerial photograph of the target range to obtain a remote sensing image, determine the geographical coordinates of each image element in the remote sensing image, and also segment out the target remote sensing image that only contains the image of the target range from the remote sensing image according to the geographical coordinates of the target range stored in advance;

[0040] S2. Process the target remote sensing image with a pre-trained classification model to distinguish the typical range and the non-typical range in the target remote sensing image, and also determine different object areas in the typical range of the target remote sensing image;

[0041] S3. For each object region in the typical range of the target remote sensing image, calculate the average value of the image element values of all the image elements in the object region to obtain the average image element value. Divide the image elements in the object region whose image element values are greater than the average image element value into the first object sub-region, divide the image elements in the object region whose image element values are less than or equal to the average image element value into the second object sub-region, and divide the image element with the largest image element value in the object region into the third object sub-region;

[0042] S4. For the first object sub-region in each object region in the typical range of the target remote sensing image, respectively determine the recognition results corresponding to different image elements in the first object sub-region, and use the recognition result with the largest total number of corresponding image elements as the final recognition result of the object region where the first object sub-region is located.

[0043] Specifically, in order to provide a simple, fast, and highly accurate forestry information recognition method, the present application provides the above steps S1 to S4. First, a photographic device is used to photograph the target area from the air to obtain a remote sensing image, which is specifically a multi-dimensional remote sensing image. The remote sensing image not only includes the image of the target area but also other images. Therefore, the geographic registration technology is used to determine the geographic coordinates corresponding to each image element in the remote sensing image. Also, according to the pre-stored geographic coordinates of the target area, a target remote sensing image that only contains the image of the target area is segmented from the remote sensing image. Secondly, the present application pre-trains a classification model. The classification model can process the target remote sensing image and distinguish the typical area and the atypical area in the target remote sensing image. The typical area corresponds to the forest area, and the atypical area does not correspond to the forest area, which can be a road area, a river area, etc. After that, different object areas are determined in the typical area of the target remote sensing image. The specific determination process will be described below. An object area corresponds to a tree, and an object area refers to the image of the corresponding tree that can be observed from the air. Thirdly, for each object area in the typical area of the target remote sensing image, the average value of the image element values of all the image elements in the object area is calculated to obtain the average image element value. The image element value can be the brightness value. Subsequently, the object area is divided into sub-areas. The image elements with image element values greater than the average image element value in the object area are divided into the first object sub-area, the image elements with image element values less than or equal to the average image element value in the object area are divided into the second object sub-area, and the image element with the maximum image element value in the object area is divided into the third object sub-area. In fact, the first object sub-area refers to the image of the sunny part of the corresponding tree that can be observed from the air, the second object sub-area refers to the image of the shady part of the corresponding tree that can be observed from the air, and the third object sub-area refers to the image of the top of the corresponding tree that can be observed from the air. Finally, for each first object sub-area in each object area in the typical area of the target remote sensing image, the recognition results corresponding to different image elements in the first object sub-area are respectively determined. The specific determination method will be described below. The recognition result with the largest total number of corresponding image elements is regarded as the final recognition result of the object area where the first object sub-area is located. The recognition result and the final recognition result both refer to the tree species. Determining the final recognition result based on the first object sub-area of the object area in the above method can achieve higher recognition accuracy compared to determining the final recognition result based on the entire object area.

[0044] Further, in S4, determining the recognition result corresponding to the image element in the first object sub-region includes: generating a characteristic curve from the image element, where the abscissa corresponding to the characteristic curve represents the increase in wavelength from left to right, and the ordinate corresponding to the characteristic curve represents the increase in the image element value from bottom to top, and comparing the characteristic curve with different pre-stored standard characteristic curves to determine the recognition result corresponding to the image element.

[0045] Specifically, here is how to determine the recognition result corresponding to the image element in the first object sub-region. First, generate a characteristic curve from the image element. The abscissa corresponding to the characteristic curve represents the increase in wavelength from left to right, successively including blue light, green light, red light, and near-infrared light. The ordinate corresponding to the characteristic curve represents the increase in the image element value, and the image element value can be the brightness value. Then, compare the characteristic curve of the image element with different pre-stored standard characteristic curves. The standard characteristic curves are characteristic curves generated from known image elements of the corresponding tree species by the same method, so that the tree species corresponding to the standard characteristic curve most similar to the characteristic curve can be used as the recognition result. Through the above method, the tree species corresponding to the image element can be determined simply and quickly.

[0046] Further, in the typical range of the target remote sensing image, determining different object regions includes the following steps:

[0047] S211. Perform smoothing processing on the typical range of the target remote sensing image to obtain the smoothed typical range of the target remote sensing image;

[0048] S212. For each image element in the typical range of the smoothed target remote sensing image, subtract a preset first value from the image element value of the image element to obtain a new typical range of the target remote sensing image;

[0049] S213. Determine an image element from the new typical range of the target remote sensing image, and judge whether the image element value of the image element is less than the maximum image element value corresponding to a preset number of surrounding image elements. If not, jump to S214. If so, continue to judge whether the maximum image element value is greater than the past image element value of the image element. If not, update the image element value of the image element with the maximum image element value. If so, update the image element value of the image element with the past image element value;

[0050] S214. Judge whether all the image elements in the new typical range of the target remote sensing image have been processed. If so, end all steps to obtain the final typical range of the target remote sensing image. If not, jump to S213.

[0051] Specifically, the process of determining different object regions in the typical range of the target remote sensing image is introduced here. First, the typical range of the target remote sensing image is smoothed, and a Gaussian filter can be specifically used to obtain the smoothed typical range of the target remote sensing image. Secondly, for each image element in the typical range of the smoothed target remote sensing image, the image element value of the image element is subtracted by a preset first value to obtain a new typical range of the target remote sensing image. The first value is set according to the actual application situation. Thirdly, an image element is determined from the new typical range of the target remote sensing image, which can be determined randomly or according to a certain rule. It is judged whether the image element value of the image element is less than the maximum image element value corresponding to a preset number of surrounding image elements. The number is also set according to the actual application situation. If not, jump to the last step. If so, continue to judge whether the maximum image element value is greater than the past image element value of the image element. If not, update the image element value of the image element with the maximum image element value. If so, update the image element value of the image element with the past image element value. Finally, it is judged whether all the image elements in the new typical range of the target remote sensing image have undergone the above processing. If so, all steps are ended, and at the same time, the final typical range of the target remote sensing image is obtained. If not, jump to the previous step, re-determine an image element that has not been determined, and then continue to perform subsequent processing.

[0052] Furthermore, in the typical range of the target remote sensing image, determining different object regions further includes the following steps:

[0053] S221. For each image element in the typical range of the final target remote sensing image, increase the image element value of the image element by a preset second value;

[0054] S222. Determine an image element from the typical range of the final target remote sensing image, and judge whether the image element value of the image element is greater than the minimum image element value corresponding to a preset number of surrounding image elements. In the case of no, jump to S223. In the case of yes, continue to judge whether the minimum image element value is greater than or equal to the past image element value of the image element. In the case of yes, update the image element value of the image element with the minimum image element value. In the case of no, update the image element value of the image element with the past image element value;

[0055] S223. Judge whether all the image elements in the typical range of the final target remote sensing image have been processed. In the case of yes, determine different object regions and end all steps. In the case of no, jump to S222.

[0056] Specifically, following S214, the process of determining different object regions in the typical range of the target remote sensing image is continued here. First, for each image element in the typical range of the final target remote sensing image, increase the image element value of the image element by a preset second value, and the second value is set according to the actual application situation. Secondly, determine an image element from the typical range of the final target remote sensing image, which can be determined randomly or according to a certain rule, and judge whether the image element value of the image element is greater than the minimum image element value corresponding to a preset number of surrounding image elements. The number here is the same as the above number. If not, proceed to the next step. If so, continue to judge whether the minimum image element value is greater than or equal to the past image element value of the image element. If so, update the image element value of the image element with the minimum image element value. If not, update the image element value of the image element with the past image element value. Finally, judge whether all the image elements in the typical range of the final target remote sensing image have undergone the above processing. If so, detect different object regions from the typical range of the final target remote sensing image, and then end all steps. Specifically, existing technologies such as edge detection algorithms can be used. If not, jump to the previous step, re-determine an image element that has not been determined before, and then continue to perform subsequent processing.

[0057] Through the above method, the boundary part of the object region becomes clearer, so that the object region can be more easily detected from the typical range of the final target remote sensing image.

[0058] Furthermore, after S3, it also includes: for each object region in the typical range of the target remote sensing image, generate several feature matrices of the object region, generate several first eigenvalues and several second eigenvalues respectively based on the several feature matrices, use the several first eigenvalues to form a first eigenvector, use the several second eigenvalues to form a second eigenvector, and use the formula Γ = ω*A + η*B to calculate the similarity score between the object region and different standard object regions respectively, where Γ is the similarity score, A and B are the first similarity score and the second similarity score between the first eigenvector and the second eigenvector and the standard first eigenvector and the standard second eigenvector respectively, ω and η are the weights of the first similarity score and the second similarity score respectively, and regard the recognition result corresponding to the standard object region corresponding to the maximum similarity score as the final recognition result of the object region;

[0059] Furthermore, generating the first eigenvalue based on the feature matrix is realized by the following formula 1:

[0060]

[0061] where, Π is the first eigenvalue, n is the number of rows of the feature matrix, m is the number of columns of the feature matrix, and δ (n,m) is the element value at the position of the n-th row and m-th column of the feature matrix;

[0062] Further, a second eigenvalue is generated based on the feature matrix, which is implemented by the following formula 2:

[0063]

[0064] where, Ω is the second eigenvalue, n is the number of rows of the feature matrix, m is the number of columns of the feature matrix, and δ (n,m) is the element value at the position of the n-th row and m-th column of the feature matrix.

[0065] Specifically, it has been introduced above how to determine the recognition result corresponding to the image element in the first object sub-region, so that in the above S4, the recognition result with the largest total number of corresponding image elements is used as the final recognition result of the object region. However, the recognition accuracy of this method still needs to be improved. Therefore, after the above S3, a new method for determining the final recognition result of the object region can also be proposed to replace the above S4.

[0066] The new method includes: for each object region in the typical range of the target remote sensing image, several feature matrices of the object region are generated. For the sake of easy understanding, the feature matrix is illustrated by an example. Suppose the pixel matrix of the object region is a 4-row and 4-column matrix, and each element in the pixel matrix corresponds to an image element in the object region. The element value of each element in the pixel matrix is the image element value of the corresponding image element in the object region. The image element value can be a brightness value. The total element values corresponding to the pixel matrix include 0, 1, 2, 3. Taking the generation of one feature matrix as an example, for each element in the pixel matrix, the corresponding paired element is determined to form an element pair. The positional relationship between the element and the paired element is that the row number of the paired element is the same as the row number of the element, and the column number of the paired element is 1 more than the column number of the element. That is to say, different element pairs all correspond to combinations of two element values. Since the total element values corresponding to the pixel matrix are 4 types, the number of combinations of two element values is 16. A 4-row and 4-column feature matrix is generated, and the row number and column number of the matrix both start from 0, specifically from 0 to 3, and respectively correspond to the element values from 0 to 3. For example, the element value at the position of row 1 and column 0 in the feature matrix refers to the result value obtained by dividing the number of element pairs composed of the element with the element value of 1 and the paired element with the element value of 0 by the total number of all element pairs. And so on for the element values at other positions in the feature matrix. Compared with this feature matrix, other feature matrices only have different positional relationships between the elements and the paired elements.

[0067] After generating several feature matrices of the object region, several first eigenvalues and several second eigenvalues are generated respectively based on the several feature matrices. Specifically, the first eigenvalues are generated using the above formula 1, and the second eigenvalues are generated using the above formula 2. Thus, several first eigenvalues can be used to form a first eigenvector, and several second eigenvalues can be used to form a second eigenvector. The similarity degree scores between the object region and different standard object regions are calculated respectively through the formula Γ = ω * A + η * B, where Γ is the similarity degree score, and A and B are the first similarity degree score and the second similarity degree score between the first eigenvector and the second eigenvector and the standard first eigenvector and the standard second eigenvector respectively. The first similarity degree score or the second similarity degree score can be obtained by first calculating the distance between two vectors and then calculating the reciprocal of the distance. The standard first eigenvector and the standard second eigenvector are generated according to the corresponding known standard object regions of tree species by the same method. ω and η are the weights of the first similarity degree score and the second similarity degree score respectively, and are both set according to the actual application situation. Furthermore, the recognition result corresponding to the standard object region corresponding to the maximum similarity degree score is regarded as the final recognition result of the object region. Through the above method, the recognition accuracy can be further improved.

[0068] The above describes the forestry information recognition method based on remote sensing data in the embodiments of the present application. Next, the forestry information recognition system based on remote sensing data in the embodiments of the present application will be described. Please refer to Figure 2 , the forestry information recognition system based on remote sensing data in the embodiments of the present application includes the following modules:

[0069] The preprocessing module is used to take an aerial photograph of the target range to obtain a remote sensing image, and determine the geographical coordinates of each image element in the remote sensing image. Also, according to the geographical coordinates of the target range stored in advance, a target remote sensing image that only contains the image of the target range is segmented from the remote sensing image;

[0070] The classification module is used to process the target remote sensing image using a pre-trained classification model to distinguish the typical range and the atypical range in the target remote sensing image, and also determine different object regions in the typical range of the target remote sensing image;

[0071] The partitioning module is used to calculate the average value of the image element values of all the image elements in the object region to obtain the average image element value for each object region in the typical range of the target remote sensing image. The image elements with image element values greater than the average image element value in the object region are partitioned into the first object sub-region, the image elements with image element values less than or equal to the average image element value in the object region are partitioned into the second object sub-region, and the image element with the largest image element value in the object region is partitioned into the third object sub-region;

[0072] An identification module is configured to respectively determine, for each first object sub-region in each object region within a typical range of a target remote sensing image, identification results corresponding to different image elements in the first object sub-region, and use the identification result with the largest total number of corresponding image elements as the final identification result of the object region where the first object sub-region is located.

[0073] In summary, in the technical solution provided in this application, first, according to the geographical coordinates of the target range stored in advance, a target remote sensing image that only contains the image of the target range is segmented from the remote sensing image. Second, the classification model is used to process the target remote sensing image to distinguish the typical range and the atypical range in the target remote sensing image, and different object regions are also determined in the typical range. Third, the image elements with image element values greater than the average image element value in the object region are divided into the first object sub-region, the image elements with image element values less than or equal to the average image element value in the object region are divided into the second object sub-region, and the image element with the largest image element value in the object region is divided into the third object sub-region. Finally, the identification results corresponding to different image elements in the first object sub-region are determined, and the identification result with the largest total number of corresponding image elements is used as the final identification result of the object region where the first object sub-region is located. Through the above method, this application can not only quickly and simply determine the identification result of the object region as forestry information, but also improve the identification accuracy of the identification method for identifying forestry information.

[0074] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the above-described system and modules can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated herein.

[0075] As described above, the above embodiments are only used to illustrate the technical solutions of this application and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments or perform equivalent replacements for some of the technical features. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A forestry information identification method based on remote sensing data, characterized in that: The steps include: S1, taking aerial photography of a target range to obtain a remote sensing image, determining the geographic coordinates of each image element in the remote sensing image, and segmenting a target remote sensing image containing only an image of the target range from the remote sensing image according to the geographic coordinates of the target range stored in advance; S2. Processing the target remote sensing image using a classification model trained in advance to distinguish a typical range and an atypical range in the target remote sensing image, and also determining different object areas in the typical range of the target remote sensing image; S3, for each of the object regions in the typical range of the target remote sensing image, calculating the average of the image element values ​​of all the image elements in the object region to obtain an average image element value, dividing the image elements in the object region whose image element values ​​are greater than the average image element value into a first object sub-region, dividing the image elements in the object region whose image element values ​​are less than or equal to the average image element value into a second object sub-region, and dividing the image elements in the object region whose image element values ​​are the largest into a third object sub-region; S4, for each of the first object sub-regions in the object region in the typical range of the target remote sensing image, respectively determining the recognition results corresponding to different image elements in the first object sub-region, and taking the recognition result with the largest total number of corresponding image elements as the final recognition result of the object region where the first object sub-region is located; Among them, the typical range corresponds to the forest range, and the image element value is the brightness value.

2. The forestry information identification method based on remote sensing data according to claim 1 is characterized in that: In S4, determining the recognition result corresponding to the image element in the first object sub-area includes: generating a characteristic curve from the image element, the horizontal axis corresponding to the characteristic curve represents the increase of wavelength from left to right, the vertical axis corresponding to the characteristic curve represents the increase of image element value from bottom to top, and comparing the characteristic curve with different standard characteristic curves stored in advance to determine the recognition result corresponding to the image element.

3. The forestry information identification method based on remote sensing data according to claim 1 is characterized in that: Determining different object regions in a typical range of the target remote sensing image comprises the following steps: S211, performing smoothing processing on the typical range of the target remote sensing image to obtain the typical range of the target remote sensing image after smoothing processing; S212, for each image element in the typical range of the target remote sensing image after smoothing, subtract a preset first value from the image element value of the image element to obtain a new typical range of the target remote sensing image; S213, determining an image element from the typical range of the new target remote sensing image, and judging whether the image element value of the image element is less than the maximum image element value corresponding to the surrounding preset number of image elements, if not, jumping to S214, if yes, continuing to judge whether the maximum image element value is greater than the past image element value of the image element, if not, using the maximum image element value to update the image element value of the image element, if yes, using the past image element value to update the image element value of the image element; S214, judging whether all the image elements in the typical range of the new target remote sensing image have been processed, if so, ending all the steps to obtain the final typical range of the target remote sensing image, if not, jumping to S213.

4. The forestry information identification method based on remote sensing data according to claim 3 is characterized in that: Determining different object regions in a typical range of the target remote sensing image also includes the following steps: S221, for each image element in the typical range of the final target remote sensing image, increasing the image element value of the image element by a preset second value; S222, determining an image element from the typical range of the final target remote sensing image, judging whether the image element value of the image element is greater than the minimum image element value corresponding to the preset number of surrounding image elements, if not, jumping to S223, if yes, continuing to judge whether the minimum image element value is greater than or equal to the past image element value of the image element, if yes, using the minimum image element value to update the image element value of the image element, if not, using the past image element value to update the image element value of the image element; S223, determine whether all image elements in the typical range of the final target remote sensing image have been processed. If yes, determine the different object areas and end all steps. If not, jump to S222.

5. The forestry information identification method based on remote sensing data according to claim 1 is characterized in that: After S3, it also includes: for each object area in the typical range of the target remote sensing image, generating several feature matrices of the object area, generating several first eigenvalues ​​and several second eigenvalues ​​based on the several feature matrices respectively, using several first eigenvalues ​​to form a first eigenvector, using several second eigenvalues ​​to form a second eigenvector, using formulas to respectively calculate the sameness score between the object area and different standard object areas, wherein is the sameness score, and are respectively the first similarity score and the second similarity score between the first eigenvector and the second eigenvector and the standard first eigenvector and the standard second eigenvector, and are respectively the weight of the first similarity score and the weight of the second similarity score, and the recognition result corresponding to the standard object area corresponding to the largest sameness score is regarded as the final recognition result of the object area.

6. The forestry information identification method based on remote sensing data according to claim 5 is characterized in that: Generate the first eigenvalue based on the characteristic matrix, which is achieved through the following formula: ； in, is the first eigenvalue, is the number of rows of the feature matrix, is the number of columns of the feature matrix, is the feature matrix Line The value of the element at the column position.

7. The forestry information identification method based on remote sensing data according to claim 5 is characterized in that: The second eigenvalue is generated based on the characteristic matrix, which is achieved through the following formula: ； in, is the second eigenvalue, is the number of rows of the feature matrix, is the number of columns of the feature matrix, is the feature matrix Line The value of the element at the column position.

8. A forestry information identification system based on remote sensing data, used to implement the method according to any one of claims 1 to 7, characterized in that: Includes the following modules: A preprocessing module is used to perform high-altitude photography on the target range to obtain a remote sensing image, determine the geographic coordinates of each image element in the remote sensing image, and segment a target remote sensing image containing an image of the target range from the remote sensing image according to the geographic coordinates of the target range stored in advance; A classification module is used to process the target remote sensing image using a classification model trained in advance to distinguish a typical range and an atypical range in the target remote sensing image, and also to determine different object areas in the typical range of the target remote sensing image, wherein the typical range corresponds to a forest range; a division module, for calculating, for each object region in a typical range of the target remote sensing image, an average of image element values ​​of all image elements in the object region to obtain an average image element value, dividing image elements in the object region whose image element values ​​are greater than the average image element value into a first object sub-region, dividing image elements in the object region whose image element values ​​are less than or equal to the average image element value into a second object sub-region, and dividing image elements in the object region whose image element values ​​are the largest into a third object sub-region, wherein the image element values ​​are brightness values; The recognition module is used to determine the recognition results corresponding to different image elements in the first object sub-region in each object region in a typical range of the target remote sensing image, and take the recognition result with the largest total number of corresponding image elements as the final recognition result of the object region where the first object sub-region is located.

Images