Machine vision-based landform contour intelligent acquisition system

Abstract

The invention relates to the technical field of image processing, in particular to a landform contour intelligent acquisition system based on machine vision, which comprises a processor and a memory, and the processor executes a computer program stored in the memory to realize the following steps: acquiring a simple landform image and a candidate complex landform image, the method comprises the steps of obtaining a candidate complex landform image, dividing the candidate complex landform image to obtain each divided sub-block on the candidate complex landform image, obtaining a structure evaluation index value of each divided sub-block according to a gradient direction histogram of each divided sub-block, and obtaining a simple sub-block and a complex sub-block on the candidate complex landform image according to the structure evaluation index value. And selecting a lightweight contour acquisition model to process the simple sub-blocks and the simple landform image, and selecting a non-lightweight contour acquisition model to process the complex sub-blocks to obtain a landform contour acquisition result of the target area. Moreover, the method can improve or guarantee the collection precision and collection efficiency of the landform contour.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image processing, and particularly relates to a topographic profile intelligent acquisition system based on machine vision. BACKGROUND

[0002] Topographic profile acquisition is a process of obtaining surface shape characteristics through automatic machine vision technology means such as three-dimensional laser scanning, unmanned aerial vehicle surveying and mapping, and remote sensing image processing, and is a basic work of topographic mapping and geographic information acquisition, and has important application value in many fields such as geographic information science, engineering construction and environmental protection.

[0003] In the prior art, when topographic profile acquisition is performed on a region requiring topographic profile acquisition, personnel related to the region requiring topographic profile acquisition generally assess the complexity of the region requiring topographic profile acquisition according to experience, and select a corresponding topographic profile acquisition model based on the assessment result to perform topographic profile acquisition. However, this method has strong subjectivity, and is prone to cause waste of computing resources or low accuracy of topographic profile acquisition. Therefore, how to perform topographic profile acquisition on a region requiring topographic profile acquisition to ensure or improve the acquisition accuracy and efficiency becomes a problem to be solved. SUMMARY

[0004] To solve the above problems, the present application provides a topographic profile intelligent acquisition system based on machine vision, and the technical solution adopted is as follows: An embodiment of the present application provides a topographic profile intelligent acquisition system based on machine vision, comprising a processor and a memory, and the processor executes a computer program stored in the memory to realize the following steps: obtain a sequence of topographic images of a target region; obtain a global topographic complexity index value of each topographic image in the sequence of topographic images according to a gradient amplitude standard deviation and a texture entropy value of the topographic image, classify the topographic image according to the global topographic complexity index value, and obtain a simple topographic image and a candidate complex topographic image; divide the candidate complex topographic image according to a gradient amplitude accumulation result in a row direction and a gradient amplitude accumulation result in a column direction of the candidate complex topographic image, and obtain each division sub-block on the candidate complex topographic image; obtain a structure evaluation index value of each division sub-block according to a gradient direction histogram of the division sub-block, and obtain a simple sub-block and a complex sub-block on the candidate complex topographic image according to the structure evaluation index value; The simple sub-blocks and the simple terrain images are processed by selecting a lightweight profile collection model, and the complex sub-blocks are processed by selecting a non-lightweight profile collection model, to obtain the terrain profile collection result of the target region.

[0005] Beneficial effects: The present application firstly acquires a terrain image sequence of a target region; then obtains a global terrain complexity index value of each terrain image according to the gradient amplitude standard deviation and the texture entropy value of each terrain image in the terrain image sequence, and classifies the terrain images according to the global terrain complexity index value to obtain simple terrain images and candidate complex terrain images; then divides the candidate complex terrain images according to the gradient amplitude accumulation result in the row direction and the gradient amplitude accumulation result in the column direction to obtain each divided sub-block on the candidate complex terrain images, and obtains a structure evaluation index value of each divided sub-block according to the gradient direction histogram of each divided sub-block, and obtains simple sub-blocks and complex sub-blocks on the candidate complex terrain images according to the structure evaluation index value; finally, the simple sub-blocks and the simple terrain images are processed by selecting a lightweight profile collection model, and the complex sub-blocks are processed by selecting a non-lightweight profile collection model, to obtain the terrain profile collection result of the target region. Moreover, the present application classifies the terrain images according to the global terrain complexity index value and classifies the divided sub-blocks according to the structure evaluation index value, and selects different models for profile collection according to the classification results, which can improve or guarantee the collection accuracy and efficiency of the terrain profile. BRIEF DESCRIPTION OF DRAWINGS

[0006] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, and the advantages thereof, below, a brief introduction will be given to the drawings needed in the embodiments or the prior art description. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can be obtained without creative effort based on these drawings.

[0007] Figure 1 The flowchart of the present application, a terrain profile intelligent collection method based on machine vision. DETAILED DESCRIPTION

[0008] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, not all the embodiments. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art belong to the scope of protection of the embodiments of the present application.

[0009] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present application belongs.

[0010] The embodiment provides a machine vision-based intelligent topographic profile collection system, which comprises a processor and a memory. The processor executes a computer program stored in the memory to implement a machine vision-based intelligent topographic profile collection method. Figure 1 As shown in the figure, the machine vision-based intelligent topographic profile collection method comprises the following steps. In step S001, a topographic image sequence of a target region is acquired.

[0011] Since the current relevant personnel have strong subjectivity in evaluating the complexity or simplicity of the region requiring topographic profile collection according to experience, the reliability or accuracy of the evaluation result is low. When the reliability or accuracy of the evaluation result is low, the precision or efficiency of the subsequent topographic profile collection of the region requiring topographic profile collection according to the evaluation result is low. Low precision means that there is a significant difference in precision, completeness and smoothness between the collection result and the real terrain. Low efficiency means that more computing resources and time are consumed. Low precision may lead to engineering design deviation, misjudgment of geological disasters and huge engineering risks. In addition, the prior art usually selects a single topographic profile collection method to collect the topographic profile of the entire region requiring topographic profile collection. This single topographic profile collection method for collecting the topographic profile of the entire region requiring topographic profile collection also causes low precision or efficiency. For example, if the global macro topographic complexity of the region requiring topographic profile collection is high, but there is a region with complex performance but simple structure in the region requiring topographic profile collection. At this time, the evaluation result of the global macro topographic complexity of the region requiring topographic profile collection usually selects a non-lightweight profile collection model to collect the topographic profile of the entire region requiring topographic profile collection. However, the region with complex performance but simple structure can use a lightweight profile collection model to ensure the collection precision. Therefore, the prior art selects a single topographic profile collection method to collect the topographic profile of the entire region requiring topographic profile collection, which causes waste of computing resources and low collection efficiency. The calculation amount of the non-lightweight profile collection model is much larger than that of the lightweight profile collection model. Therefore, in order to ensure or improve the collection precision and efficiency of the topographic profile, the embodiment provides a machine vision-based intelligent topographic profile collection method.

[0012] The geomorphic profile intelligent collection method in the embodiment needs to first collect images of a region requiring geomorphic profile collection to obtain an image sequence. After obtaining the image sequence, the images in the image sequence are classified into two categories. The purpose of the two-classification is to identify relatively obvious simple geomorphic images. The collection accuracy of simple geomorphic images is ensured while the collection efficiency is ensured. The calculation amount is reduced to ensure or improve the collection efficiency. After classification, adaptive block analysis is performed on non-simple geomorphic images, that is, candidate complex geomorphic images. The divided blocks are analyzed and judged to be real complex regions or pseudo complex regions. Based on the different judgment results, a matching geomorphic profile collection model is selected to further ensure or improve the collection accuracy and collection efficiency. Moreover, for ease of understanding and description, the geomorphic profile collection process of any region requiring geomorphic profile collection will be described in the following embodiment, and the region requiring geomorphic profile collection is recorded as a target region.

[0013] Therefore, based on the above description, it can be known that the embodiment needs to first collect images of the target region to obtain the geomorphic image sequence of the target region. The specific acquisition process of the geomorphic image sequence of the target region is as follows: First, a high-resolution camera carried by a UAV collects images of the target region according to a preset path. The sequence of the collected images according to the order of collection is recorded as a to-be-processed image sequence. There are partially overlapping regions between the images in the to-be-processed image sequence. After the splicing and fusion of all the images in the to-be-processed image sequence, the complete target region is obtained. The high-precision POS (position, attitude) data of each frame of image is recorded synchronously during collection, so as to facilitate subsequent splicing or fusion. The collected images are basic data for subsequent profile collection. In the embodiment, the implementer needs to set the preset path during collection according to experience and actual conditions, that is, the setting of the collection path is known.

[0014] Due to the fact that in the actual field collection process, the image collection is often seriously affected by factors such as illumination change, atmospheric disturbance, sensor noise and lens distortion, it is necessary to preprocess the acquired image sequence to be processed, and the preprocessed image is recorded as a topographic image, and the time sequence formed by all topographic images is recorded as a topographic image sequence. The preprocessing can avoid serious deviation in subsequent terrain complexity calculation and image or sub-block classification; the preprocessing process includes but is not limited to image correction, contrast enhancement, noise suppression, etc., and each image in the image sequence to be processed is respectively subjected to image correction, contrast enhancement and noise suppression to obtain the topographic image sequence; image correction refers to correcting the lens distortion of the image to be processed by using the camera internal parameters (such as focal length, principal point and distortion coefficient) calibrated in advance, so as to eliminate the image deformation introduced by the optical system itself and ensure the geometric authenticity of the image, thereby providing a guarantee for the subsequent accurate stereo matching; contrast enhancement refers to enhancing the contrast of the corrected image by using adaptive histogram equalization algorithm, so that the terrain details (such as small gullies and textures) are more clearly visible; noise suppression refers to filtering the enhanced image by using a multi-scale guided filtering algorithm based on Gaussian scale space, which can effectively smooth the random noise caused by high ISO or insufficient light while maximizing the preservation of important edge and texture information of the terrain, thereby providing a high-quality data source for subsequent gradient calculation, and the gradient is mainly used for analysis of complex terrain conditions.

[0015] Therefore, the topographic image sequence of the target region is obtained by the above process.

[0016] Step S002, obtaining a global topographic complexity index value of each topographic image in the topographic image sequence according to the gradient amplitude standard deviation and the texture entropy value of each topographic image, classifying the topographic images according to the global topographic complexity index value, and obtaining simple topographic images and candidate complex topographic images.

[0017] After obtaining the sequence of topographic images, the embodiment evaluates the global terrain complexity based on the standard deviation of gradient amplitude and texture entropy value of the topographic images, and performs binary classification based on the evaluation to obtain simple topographic images and candidate complex topographic images. Subsequently, the candidate complex topographic images are classified again in detail. The binary classification in the embodiment has low computational complexity, and can separate the large-area simple topographic region that is obvious and the candidate complex topographic region that needs further processing, thereby avoiding unnecessary calculation of the subsequent time-consuming fine analysis algorithm on the simple region, and improving the efficiency and accuracy of subsequent topographic contour collection. As described above, the global terrain complexity needs to be evaluated according to the standard deviation of gradient amplitude and texture entropy value of the topographic images before binary classification, that is, the global terrain complexity index value of each topographic image needs to be obtained according to the standard deviation of gradient amplitude and texture entropy value of each topographic image in the sequence of topographic images. The specific process of obtaining the global terrain complexity index value of each topographic image is as follows: for any topographic image A: First, the gradient amplitude and gradient direction of each pixel point on the topographic image A are calculated. The gradient amplitude and gradient direction of the pixel point can be calculated by using gradient operators such as Sobel and Prewitt, and the acquisition process is known. Then, the standard deviation of the gradient amplitude of all pixel points on the topographic image A is calculated, and is denoted as the gradient amplitude standard deviation corresponding to the topographic image A. The gradient amplitude standard deviation is normalized by using a normalization function Norm(), and the normalized gradient amplitude standard deviation is denoted as the normalized gradient amplitude standard deviation corresponding to the topographic image A. Then, the texture entropy value of the topographic image A is calculated, and the texture entropy value of the topographic image A is normalized by using a normalization function Norm(), and the normalized texture entropy value of the topographic image A is denoted as the normalized texture entropy value corresponding to the topographic image A. The texture entropy value of the topographic image A is the entropy value in the gray level co-occurrence matrix feature value of the topographic image A, and the calculation process is known. Finally, the normalized gradient amplitude standard deviation corresponding to the topographic image A and the normalized texture entropy value corresponding to the topographic image A are weighted and summed, and the weighted sum result is denoted as the global terrain complexity index value of the topographic image A. The calculation expression of the global terrain complexity index value of the topographic image A is: wherein, is the global terrain complexity index value of the topographic image A, is the first weight value, is the second weight value, and Norm() is a normalization function, is the gradient amplitude standard deviation corresponding to the topographic image A, is the texture entropy value of the topographic image A; The greater, the more scattered, the more dramatic changes, the more irregular gradient amplitude of the geomorphology image A, the higher the global macro geomorphology complexity of the region corresponding to the geomorphology image A, and the global macro geomorphology complexity is mainly jointly acted by the texture and the edge number; The greater, the more complex or more chaotic texture of the geomorphology image A, the higher the global macro geomorphology complexity of the region corresponding to the geomorphology image A; and The greater, the more complex or more chaotic texture of the geomorphology image A, the higher the global macro geomorphology complexity of the region corresponding to the geomorphology image A; and The greater, the more complex or more chaotic texture of the geomorphology image A, the higher the global macro geomorphology complexity of the region corresponding to the geomorphology image A; and The greater, the more complex or more chaotic texture of the geomorphology image A, the higher the global macro geomorphology complexity of the region corresponding to the geomorphology image A; and The greater, the more complex or more chaotic texture of the geomorphology image A, the higher the global macro geomorphology complexity of the region corresponding to the geomorphology image A; and The smaller, the more complex or more chaotic texture of the geomorphology image A, the higher the global macro geomorphology complexity of the region corresponding to the geomorphology image A; and The smaller, the more complex or more chaotic texture of the geomorphology image A, the higher the global macro geomorphology complexity of the region corresponding to the geomorphology image A; and The smaller, the more complex or more chaotic texture of the geomorphology image A, the higher the global macro geomorphology complexity of the region corresponding to the geomorphology image A; and The smaller, the more complex or more chaotic texture of the geomorphology image A, the higher the global macro geomorphology complexity of the region corresponding to the geomorphology image A; and The smaller, the more complex or more chaotic texture of the geomorphology image A, the higher the global macro geomorphology complexity of the region corresponding to the geomorphology image A; and

[0018] After obtaining the global terrain complexity index value of the landform image, binary classification is performed based on the size of the global terrain complexity index value to obtain a simple landform image and a candidate complex landform image. Specifically, for any landform image A, it is determined whether the global landform complexity index value of the landform image A is greater than a preset initial judgment threshold. If yes, it indicates that the terrain or landform complexity of the region corresponding to the landform image A is high in the global macroscopic view, but this complexity is a global macroscopic high relief high texture, and the internal region may contain a region with complex performance but simple structure, such as regular wheat field and tile roof. Therefore, fine reclassification needs to be performed. At this time, the landform image A is recorded as a candidate complex landform image. The candidate complex landform image is an image that needs to be fine reclassified. If it is determined that the global landform complexity index value of the landform image A is not greater than the preset initial judgment threshold, it indicates that the overall terrain or landform complexity of the region corresponding to the landform image A is low. A lightweight contour collection model can be used to realize high-precision collection, and it is not necessary to perform subsequent fine reclassification. Therefore, at this time, the landform image A is recorded as a simple landform image. In addition, in the present embodiment, the implementer can set the preset initial judgment threshold according to the actual situation, experimental statistics, value range, etc. For example, the preset initial judgment threshold can be set to 0.3. The preset initial judgment threshold can also be set according to historical data. The process of setting the preset initial judgment threshold according to historical data is as follows: first, all historical images that are evaluated as simple landforms by relevant practitioners in the contour collection stage are obtained. The global landform complexity index value of each historical image is calculated, and the mean value of the global landform complexity index values of all the historical images is taken as the preset initial judgment threshold.

[0019] Therefore, the landform image sequence in the landform image sequence is obtained through the above process.

[0020] In step S003, the candidate complex landform image is divided according to the gradient amplitude accumulation result in the row direction and the gradient amplitude accumulation result in the column direction to obtain each divided sub-block on the candidate complex landform image. The structure evaluation index value of each divided sub-block is obtained according to the gradient direction histogram of each divided sub-block. The simple sub-block and the complex sub-block on the candidate complex landform image are obtained according to the structure evaluation index value. A lightweight contour collection model is selected to process the simple sub-block and the simple landform image, and a non-lightweight contour collection model is selected to process the complex sub-block to obtain the landform contour collection result of the target region.

[0021] Since the complexity of the candidate complex landform image obtained above is presented by the joint action of texture and edge quantity, the complexity is high relief and high texture in the global macroscopic view, but it can contain regions with simple structure and complex appearance, such as regular wheat fields, tile roofs, etc., and regions with real internal complexity, such as erosion gullies and rock piles, etc. Since the regions with simple structure have regular characteristics, the use of a lightweight contour collection model can still ensure accuracy, so they need to be divided into simple landform regions or simple sub-blocks, and a lightweight contour collection model is used for subsequent contour collection, so as to achieve the purpose of ensuring collection accuracy while reducing unnecessary calculation amount. But for the real internal complex regions, such as erosion gullies and rock piles, etc., a high-precision contour collection model is needed to ensure the contour collection accuracy.

[0022] Based on the above description, the subsequent embodiment needs to adaptively divide the candidate complex landform image into blocks or grids, and the purpose of the division is to divide the simple structure regions and the complex structure regions on the image as much as possible to ensure or improve the subsequent contour collection efficiency and collection accuracy, and reduce unnecessary calculation amount to ensure or improve the collection efficiency. In addition, the adaptive block division of the candidate complex landform image in this embodiment needs to be based on the concentration or complexity in different directions for block division, and the subsequent embodiment is analyzed and divided from the gradient amplitude comprehensive situation in the row and column directions, that is, the subsequent embodiment will divide the candidate complex landform image according to the gradient amplitude accumulation result in the row direction and the gradient amplitude accumulation result in the column direction of the candidate complex landform image, to obtain each divided sub-block on the candidate complex landform image, and the specific process of obtaining each divided sub-block on the candidate complex landform image is as follows: For any candidate complex landform image Q, first, the row division number parameter and the column division number parameter of the candidate complex landform image Q are obtained according to the gradient amplitude accumulation result in the row direction and the gradient amplitude accumulation result in the column direction of the candidate complex landform image Q, and then the candidate complex landform image Q is divided according to the row division number parameter and the column division number parameter of the candidate complex landform image Q to obtain each divided sub-block on the candidate complex landform image Q. The row division number parameter and the column division number parameter of the candidate complex landform image can measure the gradient amplitude consistency or complexity from the row and column directions, and are also the key parameters for dividing the simple structure regions and the complex structure regions on the image.

[0023] In this embodiment, the specific process of dividing the candidate complex landform image Q according to the row division number parameter and the column division number parameter of the candidate complex landform image Q to obtain each divided sub-block on the candidate complex landform image Q is as follows: Firstly, the comprehensive gradient amplitudes corresponding to each row and each column in the candidate complex topography image Q are obtained, and the comprehensive gradient amplitude corresponding to the i-th row in any candidate complex topography image is the sum of the gradient amplitudes of all pixel points on the i-th row in the candidate complex topography image, and the comprehensive gradient amplitude corresponding to the j-th column is the sum of the gradient amplitudes of all pixel points on the j-th column in the candidate complex topography image; then the vector composed of the comprehensive gradient amplitudes corresponding to all rows in the candidate complex topography image Q is denoted as the row vector of the candidate complex topography image Q, and the vector composed of the comprehensive gradient amplitudes corresponding to all columns in the candidate complex topography image Q is denoted as the column vector of the candidate complex topography image Q, the value of the a-th element in the row vector of the candidate complex topography image Q is the comprehensive gradient amplitude corresponding to the a-th row in the candidate complex topography image Q, and the value of the b-th element in the column vector of the candidate complex topography image Q is the comprehensive gradient amplitude corresponding to the b-th column in the candidate complex topography image Q; then the row division number parameter of the candidate complex topography image Q is obtained according to the row vector of the candidate complex topography image Q, and the column division number parameter of the candidate complex topography image Q is obtained according to the row and column vectors of the candidate complex topography image Q; then the candidate complex topography image Q is divided according to the row division number parameter and the column division number parameter of the candidate complex topography image Q, to obtain each division sub-block on the candidate complex topography image Q; and after the candidate complex topography image Q is divided, reclassification is performed based on the division result, and the corresponding collection model is selected according to the classification result, which is the key to ensure or improve the collection efficiency and collection accuracy of the embodiment.

[0024] In the embodiment, the specific process of obtaining the row division number parameter of the candidate complex topography image Q and obtaining the column division number parameter of the candidate complex topography image Q is as follows: Firstly, the square value of the accumulated result of all element values in the row vector of the candidate complex topography image Q is calculated, and denoted as a first comprehensive value, the accumulated result of the squares of all element values in the row vector of the candidate complex topography image Q is calculated, and denoted as a second comprehensive value, and the ratio of the first comprehensive value to the second comprehensive value is calculated, and denoted as the row distribution representation value of the candidate complex topography image Q, and the expression of the row distribution representation value of the candidate complex topography image Q is , H is the total number of elements in the row vector of the candidate complex topography image Q, is the value of the a-th element in the row vector of the candidate complex landform image Q, and the greater the row distribution representation value of the candidate complex landform image Q, the more it indicates that the structural consistency of the gradient amplitude of the candidate complex landform image Q in the row direction is low, the more the change mode, the more complex the distribution, and the more the independent change mode in the vector. The lower the structural consistency of the gradient amplitude in the row direction indicates that the internal appearance is complex and the structure is disordered from the row direction. In order to divide the structural simple region and the structural complex region on the image as much as possible, the greater the degree of segmentation in the row direction should be. The smaller the row distribution representation value of the candidate complex landform image Q, the higher the structural consistency of the gradient amplitude of the candidate complex landform image Q in the row direction or the fewer the change mode, the less complex the distribution, and the higher the structural consistency of the gradient amplitude in the row direction indicates that the internal appearance is complex but the structure is ordered from the row direction. At this time, the smaller degree of segmentation can divide the structural simple region and the structural complex region on the image as much as possible. The greater the degree of segmentation indicates that the number or times of division is more.

[0025] Then the square value of the cumulative result of all element values in the column vector of the candidate complex landform image Q is calculated, and is recorded as the third comprehensive value. The cumulative result of the square of all element values in the column vector of the candidate complex landform image Q is calculated, and is recorded as the fourth comprehensive value. The ratio of the third comprehensive value to the fourth comprehensive value is calculated, and is recorded as the column distribution representation value of the candidate complex landform image Q. The expression of the column distribution representation value of the candidate complex landform image Q is , W is the total number of elements in the column vector of the candidate complex landform image Q, is the value of the b-th element in the column vector of the candidate complex landform image Q, and the greater the column distribution representation value of the candidate complex landform image Q, the lower the consistency or concentration of the gradient amplitude of the candidate complex landform image Q in the column direction, the more the change mode, the more complex the distribution. The greater the column distribution representation value of the candidate complex landform image Q, the lower the structural consistency of the gradient amplitude of the candidate complex landform image Q in the column direction, the more the change mode, the more complex the distribution, and the more the independent change mode in the vector. The lower the structural consistency of the gradient amplitude in the column direction indicates that the internal appearance is complex and the structure is disordered from the row direction. In order to divide the structural simple region and the structural complex region on the image as much as possible, the greater the degree of segmentation in the row direction should be. The smaller the row distribution representation value of the candidate complex landform image Q, the higher the structural consistency of the gradient amplitude of the candidate complex landform image Q in the row direction or the fewer the change mode, the less complex the distribution, and the higher the structural consistency of the gradient amplitude in the row direction indicates that the internal appearance is complex but the structure is ordered from the row direction. At this time, the smaller degree of segmentation can divide the structural simple region and the structural complex region on the image as much as possible. The greater the degree of segmentation indicates that the number or times of division is more.

[0026] The row distribution representation value of the candidate complex landform image Q is then rounded up, and the rounding result is taken as the row division number parameter of the candidate complex landform image Q. The column distribution representation value of the candidate complex landform image Q is rounded up, and the rounding result is taken as the column division number parameter of the candidate complex landform image Q. The rounding purpose is to ensure that the subsequent divided sub-blocks have consistent areas. In addition, in order to prevent the divided sub-blocks from being too small or to prevent the grid from being too fine, the embodiment needs to set a division upper limit. For example, the embodiment can require that the row division number parameter of the candidate complex landform image Q does not exceed the upward rounding value of 10% of the length of the wide side of the candidate complex landform image Q. The embodiment can require that the column division number parameter of the candidate complex landform image Q does not exceed the upward rounding value of 10% of the length of the long side of the candidate complex landform image Q.

[0027] In the embodiment, the candidate complex landform image Q is divided according to the row division number parameter and the column division number parameter of the candidate complex landform image Q, and the specific obtaining process of each division sub-block on the candidate complex landform image Q is as follows: The row division number parameter of the candidate complex landform image Q is denoted as m, the column division number parameter of the candidate complex landform image Q is denoted as n, the candidate complex landform image Q is uniformly divided into n x m sub-blocks, and each is denoted as a division sub-block. The length of the edge on the division sub-block of the candidate complex landform image Q that is parallel to the long side of the candidate complex landform image Q is , the length of the edge on the division sub-block of the candidate complex landform image Q that is parallel to the wide side of the candidate complex landform image Q is , N is the length of the long side of the candidate complex landform image Q, and M is the length of the wide side of the candidate complex landform image Q. That is, the wide side of the candidate complex landform image Q is uniformly divided by m-1 straight lines parallel to the long side of the candidate complex landform image Q, and the long side of the candidate complex landform image Q is uniformly divided by n-1 straight lines parallel to the wide side of the candidate complex landform image Q. Each grid obtained by the division is a division sub-block on the candidate complex landform image Q, and the long side of the image is parallel to the horizontal direction.

[0028] After the division is completed, the gradient direction of each division sub-block obtained by the division is evaluated for simplicity or complexity, and the simple sub-blocks and complex sub-blocks on the candidate complex landform image are obtained based on the structure evaluation result. That is, the embodiment will first obtain the structure evaluation index value of each division sub-block on the candidate complex landform image according to the gradient direction histogram of each division sub-block on the candidate complex landform image, and then obtain the simple sub-blocks and complex sub-blocks on the candidate complex landform image according to the structure evaluation index value of each division sub-block on the candidate complex landform image.

[0029] In the embodiment, the specific process of obtaining the structure evaluation index value of each divided sub-block on the candidate complex landform image according to the gradient direction histogram of each divided sub-block on the candidate complex landform image is as follows: For any divided sub-block on any candidate complex landform image, a normalized gradient direction histogram of the divided sub-block is constructed, and the information entropy of the normalized gradient direction histogram of the divided sub-block is calculated. The information entropy of the normalized gradient direction histogram of the divided sub-block is negatively correlated and mapped, and the mapping result is recorded as the structure evaluation index value of the divided sub-block. The negative correlation mapping means that the information entropy is added by 1 and then the reciprocal is taken, and the addition of 1 is to prevent the denominator from being 0. The construction process of the normalized gradient direction histogram and the calculation process of the information entropy of the normalized gradient direction histogram are both known. The greater the structure evaluation index value of the divided sub-block is, the more concentrated the edge direction of the divided sub-block is, the more ordered or regular the structure of the divided sub-block is, and the more likely the divided sub-block is a simple structure region or a pseudo-complex region. The smaller the structure evaluation index value of the divided sub-block is, the more disordered or irregular the structure of the divided sub-block is, and the more likely the divided sub-block is a real complex structure region.

[0030] In the embodiment, the specific process of obtaining the simple sub-block and the complex sub-block on the candidate complex landform image according to the structure evaluation index value of each divided sub-block on the candidate complex landform image is as follows: For any divided sub-block on any candidate complex landform image, it is judged whether the structure evaluation index value of the divided sub-block is less than a preset target judgment threshold. If yes, it indicates that the divided sub-block is a real complex structure region, and a high-precision acquisition model is required to ensure the acquisition accuracy. Therefore, the divided sub-block is recorded as a complex sub-block. If it is judged that the structure evaluation index value of the divided sub-block is not less than the preset target judgment threshold, it indicates that the divided sub-block is a pseudo-complex structure region or a simple structure region. At this time, a light-weight acquisition model can also ensure the acquisition accuracy. Therefore, in order to reduce the calculation amount and ensure the acquisition efficiency, the divided sub-block is recorded as a simple sub-block at this time. In the embodiment, the implementer can set the preset target judgment threshold according to the actual situation, experimental statistics, value range, etc. For example, the preset target judgment threshold can be set to 0.6. The preset target judgment threshold can also be set according to historical data, and the process of setting the preset target judgment threshold according to historical data is as follows: first, all divided sub-blocks evaluated as simple sub-blocks by relevant practitioners are obtained on the historical images collected in the historical contour acquisition stage. The mean value of the structure evaluation index values of the simple sub-blocks on the obtained historical images can be taken as the preset target judgment threshold.

[0031] Therefore, the embodiment obtains the simple sub-blocks and the complex sub-blocks on the simple terrain image and the candidate complex terrain image based on the above process, and based on the above analysis, in order to improve or ensure the collection efficiency and the collection accuracy, the simple sub-blocks and the simple terrain image should be processed by using the lightweight contour collection model, the complex sub-blocks should be processed by using the non-lightweight contour collection model, then the simple sub-blocks on all the candidate complex terrain images and all the simple terrain images are processed by using the lightweight contour collection model, the complex sub-blocks on all the candidate complex terrain images are processed by using the non-lightweight contour collection model, the results output by the processing are spliced and fused according to the original coordinates, and finally a complete high-precision terrain contour data corresponding to the target region is generated. The non-lightweight contour collection model belongs to a high-precision three-dimensional reconstruction model. The non-lightweight contour collection model usually refers to a complex model architecture used for high-precision contour extraction in computer vision or three-dimensional modeling. The calculation complexity of the non-lightweight contour collection model is high, and the lightweight contour collection model refers to a low-power and high-efficiency contour detection algorithm architecture designed for edge computing devices. The core feature of the lightweight contour collection model is to realize real-time contour extraction capability through model compression and algorithm optimization. The calculation complexity of the lightweight contour collection model is low. In addition, the process of processing the image or the sub-block by using the lightweight contour collection model or the non-lightweight contour collection model and outputting data is known.

[0032] Up to now, the embodiment completes the intelligent collection of the terrain contour of the target region, and the intelligent collection method of the terrain contour provided by the embodiment can ensure or improve the terrain contour collection accuracy and the collection efficiency.

[0033] In summary, the embodiment first obtains the terrain image sequence of the target region, then obtains the global terrain complexity index value of each terrain image according to the gradient amplitude standard deviation and the texture entropy value of each terrain image in the terrain image sequence, classifies the terrain images according to the global terrain complexity index value, obtains the simple terrain image and the candidate complex terrain image, then divides the candidate complex terrain image according to the gradient amplitude accumulation result in the row direction and the gradient amplitude accumulation result in the column direction, obtains each division sub-block on the candidate complex terrain image, obtains the structure evaluation index value of each division sub-block according to the gradient direction histogram of each division sub-block, obtains the simple sub-block and the complex sub-block on the candidate complex terrain image according to the structure evaluation index value, and finally selects the lightweight contour collection model to process the simple sub-block and the simple terrain image, selects the non-lightweight contour collection model to process the complex sub-block, and obtains the terrain contour collection result of the target region. According to the global terrain complexity index value and the structure evaluation index value, the embodiment classifies the terrain images and the division sub-blocks, and selects different models for contour collection according to the classification results, which can improve or ensure the collection accuracy and the collection efficiency of the terrain contour.

[0034] The above-described embodiments are only used to illustrate the technical solutions of the present application, but not limit the present application; although the present application has been described in detail with reference to the foregoing embodiments, those ordinarily skilled in the art should understand: the technical solutions recorded in the foregoing embodiments can still be modified, or some technical features can be replaced equivalently; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be included in the protection scope of the present application.

Claims

1.A machine vision-based intelligent topographic profile acquisition system, comprising a processor and a memory, characterized in that, The processor executes the computer program stored in the memory to implement the following steps: Obtain a topographic image sequence of a target region; According to the gradient amplitude standard deviation and the texture entropy value of each topographic image in the topographic image sequence, obtain a global topographic complexity index value of the topographic image, classify the topographic image according to the global topographic complexity index value, and obtain a simple topographic image and a candidate complex topographic image; According to the gradient amplitude accumulation result of the candidate complex topographic image in the row direction and the gradient amplitude accumulation result in the column direction, divide the candidate complex topographic image to obtain each division sub-block on the candidate complex topographic image; According to the gradient direction histogram of each division sub-block, obtain a structure evaluation index value of the division sub-block, and according to the structure evaluation index value, obtain a simple sub-block and a complex sub-block on the candidate complex topographic image; Select a lightweight contour acquisition model to process the simple sub-block and the simple topographic image, and select a non-lightweight contour acquisition model to process the complex sub-block, to obtain a topographic contour acquisition result of the target region. 2.The machine vision-based topographic profile intelligent acquisition system of claim 1, wherein, The global topographic complexity index value of the topographic image is a weighted sum result of the normalization result of the gradient amplitude standard deviation of all pixel points on the topographic image and the normalization result of the texture entropy value of the topographic image. 3.The machine vision based geomorphologic profile intelligent acquisition system of claim 1, wherein, The method for obtaining the simple topographic image and the candidate complex topographic image comprises: The topographic image with a global topographic complexity index value greater than a preset initial judgment threshold in the topographic image sequence is recorded as a candidate complex topographic image, and the topographic image with a global topographic complexity index value not greater than the preset initial judgment threshold is recorded as a simple topographic image. 4.The machine vision-based topographic profile intelligent acquisition system of claim 1, wherein, The method for obtaining each division sub-block on the candidate complex topographic image comprises: According to the gradient amplitude accumulation result of the candidate complex topographic image in the row direction and the gradient amplitude accumulation result in the column direction, obtain a row division quantity parameter and a column division quantity parameter of the candidate complex topographic image, and divide the candidate complex topographic image according to the row division quantity parameter and the column division quantity parameter to obtain each division sub-block on the candidate complex topographic image. 5.The machine vision based topographic profile intelligent acquisition system of claim 4, wherein, The method for obtaining the row division quantity parameter and the column division quantity parameter comprises: A vector composed of the comprehensive gradient amplitudes corresponding to all rows in the candidate complex topographic image is recorded as a row vector of the candidate complex topographic image, and the row division quantity parameter of the candidate complex topographic image is obtained according to the row vector of the candidate complex topographic image; a vector composed of the comprehensive gradient amplitudes corresponding to all columns in the candidate complex topographic image is recorded as a column vector of the candidate complex topographic image, and the column division quantity parameter of the candidate complex topographic image is obtained according to the column vector of the candidate complex topographic image. 6.The machine vision based topographic profile intelligent acquisition system of claim 5, wherein, The row division quantity parameter of the candidate complex landform image is an integer result of a ratio of a square value of an accumulated result of all elements in the row vector to an accumulated result of square values of all elements in the row vector, and the column division quantity parameter of the candidate complex landform image is an integer result of a ratio of a square value of an accumulated result of all elements in the column vector to an accumulated result of square values of all elements in the row vector. 7.The machine vision based topographic profile intelligent acquisition system of claim 5, wherein, The integrated gradient amplitude corresponding to the i-th row in any candidate complex landform image is a sum of gradient amplitudes of all pixel points on the i-th row in the candidate complex landform image, and the integrated gradient amplitude corresponding to the j-th column is a sum of gradient amplitudes of all pixel points on the j-th column in the candidate complex landform image. 8.The machine vision based topographic profile intelligent acquisition system of claim 4, wherein, The method for dividing the candidate complex landform image according to the row division quantity parameter and the column division quantity parameter to obtain each division sub-block on the candidate complex landform image comprises: The row division number parameter of the candidate complex landform image is denoted as m, the column division number parameter of the candidate complex landform image is denoted as n, the candidate complex landform image is uniformly divided into n x m sub-blocks, and each is denoted as a division sub-block on the candidate complex landform image. The length of the edge on the division sub-block parallel to the long side of the candidate complex landform image is , the length of the edge on the division sub-block parallel to the width side of the candidate complex landform image is , N is the length of the long side of the candidate complex landform image, and M is the length of the width side of the candidate complex landform image. 9.The machine vision based topographic profile intelligent acquisition system of claim 1, wherein, The structure evaluation index value of the division sub-block is a result of negative correlation mapping of an information entropy of a normalized gradient direction histogram of the division sub-block. 10.The machine vision based topographic profile intelligent acquisition system of claim 1, wherein, The method for obtaining simple sub-blocks and complex sub-blocks on the candidate complex landform image comprises: The division sub-blocks with a structure evaluation index value less than a preset target judgment threshold are all recorded as complex sub-blocks, and the division sub-blocks with a structure evaluation index value not less than the preset target judgment threshold are all recorded as simple sub-blocks.

Images