CN114255406A

CN114255406A - Method for identifying non-penetration surface of remote sensing shadow measurement restoration image

Info

Publication number: CN114255406A
Application number: CN202111593175.4A
Authority: CN
Inventors: 陈思思
Original assignee: Individual
Current assignee: Individual
Priority date: 2021-12-23
Filing date: 2021-12-23
Publication date: 2022-03-29

Abstract

The method for identifying the non-permeable surface of the high-resolution remote sensing image is designed around the measurement restoration, the land interpretation and the refined urban non-permeable surface identification of the shadow area on the high-resolution remote sensing image: aiming at the problem that the shadow generated by a high-rise building and a crown on an image causes the identification of an urban non-infiltration surface to have uncertainty, the method for measuring and repairing the shadow of the high-resolution remote sensing image facing the non-infiltration surface is provided, PCA (principal component analysis) transformation and HIS (high-intensity-localization) transformation are introduced to obtain various spectral characteristics, so that the identification and measurement of a shadow area are realized, and the repair of the urban shadow area is realized on the basis of blue-band component inhibition and HSI (high-speed coherent integration) space repair; secondly, in order to better obtain the urban non-permeability surface recognition result and analyze the influence of the settings of different segmentation factors and parameters on the non-permeability surface recognition, a multi-decision tree combined classifier is adopted to realize the object-oriented non-permeability surface refined recognition, the non-permeability surface recognition accuracy is 96.38%, and powerful guarantee is provided for the research and application of the non-permeability surface.

Description

Method for identifying non-penetration surface of remote sensing shadow measurement restoration image

Technical Field

The application relates to a method for identifying a non-penetration surface of a remote sensing shadow image, in particular to a method for identifying a non-penetration surface of a remote sensing shadow metering repair image, and belongs to the technical field of remote sensing image non-penetration surface identification.

Background

The non-permeable surface is an area which cannot be penetrated by water, the non-permeable surface is mainly constructed by manpower and comprises roads, buildings, parking lots and the like, the permeable surface opposite to the non-permeable surface comprises greenbelts, lakes, rivers, bare soil and the like, and with the rapid development of the current urbanization, a large area of permeable surface is replaced by the non-permeable surface. The increase of non-permeable surface in the city can make the precipitation be difficult to permeate into the earth's surface, leads to regional infiltration water and soil moisture to reduce together for groundwater exchange capacity and soil water storage function weaken by a wide margin, and the rainfall will only more be able to more with the mode of runoff and pipeline inward-remittance river network, thereby lead to the frequency increase that flood and urban waterlogging take place. In addition, the non-permeable surface has strong solar radiation absorption capacity, and after a large amount of solar radiation is absorbed, the absorbed radiation is emitted out, so that the city is continuously heated, the microclimate of the city is changed, and a heat island effect is easily formed.

The remote sensing image processing technology enables a large-range non-penetration surface to be obtained gradually, and due to the fact that the types of ground objects contained in the non-penetration surface are very complex, the non-penetration surface identification method based on spectral features has the phenomena of same object different spectrums and same foreign object spectrums. The non-permeable surface is identified by using the medium-low resolution image, and the non-permeable surface is easily confused with soil, water and shadow areas. Although the problem of assimilation pixel elements can be solved by the sub-pixel level non-penetration surface identification method, the method depends on the ratio of the non-penetration surfaces occupied by the pixel elements, and the problem of overestimation or underestimation of the non-penetration surfaces is easily caused. The current high-resolution image is developed rapidly, so that the acquisition of regional non-permeability surface information by using a high-resolution remote sensing image is more possible. Due to the characteristic of high spatial resolution, high-precision regional non-permeability surface information can be obtained, and classification precision can be improved by selecting a high-resolution remote sensing image as a data source for urban non-permeability surface identification, so that the method is significant. In addition, the water impermeability is one of important indexes in the urbanization process and is also a main initiating factor of urban flood, waterlogging and heat island effect. Therefore, the method has important application value for identifying the non-penetration surface with high precision.

The work of identifying non-permeability surfaces is mainly based on artificial statistics investigation at the earliest, and although the accuracy of the non-permeability surface information obtained through manual interpretation is still enough, the data source is limited in various aspects, so that the non-permeability surface information can be only used in a small area. The remote sensing image has the earth observation capability, can be used for non-penetration surface extraction, and is the only method capable of automatically acquiring the information of the non-penetration surface in a large area.

The quantitative relation exists between the characteristics expressed based on the non-penetration surface and the related information of the non-penetration surface, so that the non-penetration surface characteristics can be adopted for identifying and extracting the non-penetration surface. The more common features are spectral, spatial, geometric and temporal, and these feature differences can be used to distinguish between non-permeable surfaces and other features.

Remote sensing images adopted by the non-penetration surface identification can be divided into high-resolution remote sensing images, medium-low resolution remote sensing images and low-resolution remote sensing images. The images with different resolutions are suitable for different classification scales, are divided into two types of scales based on the pixel and scales based on the sub-pixel, and are identified as binary classification based on the pixel scale.

The method adopts a low-medium resolution remote sensing image suitable for a large-area to identify the non-permeable surface, the identification precision is slightly low, and the method adopts a macroscopic viewing angle. When a high-resolution image is used, high-precision watertight rate information can be provided for a user, and the method adopts a finer visual angle. Due to the requirement for high-precision non-penetration surface identification and acquisition, the high-resolution image has the problems of unavoidable shadow region information loss and the like, and the problem has a large negative effect on the identification precision of the non-penetration surface of the region.

Therefore, to accurately acquire regional non-permeability information, firstly, the shadow needs to be classified, and for the detection, firstly, the shadow region needs to be detected, and the complete removal of the shadow on the image cannot be realized, but the corresponding shadow region can be repaired based on the brightness and color information in the shadow region, so that the condition of shadow information loss is improved. The existing restoration methods in the prior art include a histogram matching method, a linear correlation correction method, an exponential correction method, a homomorphic filtering method, a normalization processing method and the like, which all have a certain positive effect on image enhancement of a shadow region, but no effective processing method which is well recognized by people exists at present. The purpose of the shadow area restoration is to improve the visual effect of the shadow area, and meanwhile, the spectral information in the non-shadow area is kept unchanged as much as possible.

However, the problem of information loss of shadow areas is likely to be encountered by adopting the high-resolution remote sensing image, the method is to adopt the high-resolution remote sensing image to identify the non-penetration surface, firstly adopt multi-feature components to carry out shadow measurement, repair the shadow measurement result based on RGB and HIS space so as to improve the problem of shadow information loss in the high-resolution image, then classify the shadow measurement result by using a remote sensing image classification method based on a multi-decision tree joint classifier and facing an object, obtain the urban non-penetration surface result and improve the identification precision of the non-penetration surface.

In summary, the identification of the non-penetration surface of the remote sensing image in the prior art has obvious defects, and the main defects and difficulties thereof include:

firstly, the identification work of the non-permeable surface is mainly manual statistics and investigation at the earliest time, a data source is limited in all aspects, the non-permeable surface can only be used in a small-range area, the work repeatability is strong, the speed is slow, the workload is heavy, the real-time performance is poor, and the data is difficult to update, so that the automation level is urgently required to be improved; the method is characterized in that a low-medium resolution remote sensing image is adopted to be suitable for a large-area to identify the non-permeable surface, the identification precision is low, only a macroscopic view angle can be achieved, the high-precision non-permeable surface identification and acquisition requirements cannot be met, the high-resolution image has the problems of unavoidable shadow area information loss and the like, the problem can cause large negative effects on the area non-permeable surface identification precision, the shadow on the image cannot be completely removed in the prior art, the shadow information loss condition cannot be improved, and a method which is acknowledged by people and is used for effectively repairing the shadow area does not exist. The visual effect of the shadow area cannot be improved, the spectral information in the non-shadow area cannot be ensured to be unchanged, and the requirement for accurately identifying the non-penetration surface of the remote sensing image cannot be met;

secondly, because the types of the ground objects contained in the non-permeable surface are very complex, the non-permeable surface is identified by using a medium-low resolution image, the non-permeable surface is easy to be confused with soil, water and shadow areas, although the problem of assimilation pixel can be solved by using a sub-pixel level non-permeable surface identification method, the problem of over-high estimation or under-low estimation of the non-permeable surface is easily caused by the method depending on the ratio of the non-permeable surface occupied by each pixel, the high-resolution remote sensing image is used for acquiring the regional non-permeable surface information to obtain the regional non-permeable surface information with higher precision, however, the problem of shadow area information loss is likely to be encountered by using the high-resolution remote sensing image, the prior art cannot accurately identify and measure the shadow of the remote sensing image, cannot effectively repair the shadow identification and measurement result so as to improve the problem of shadow information loss in the high-resolution image, and further cannot obtain the accurate result of the urban non-permeable surface by using an automatic classification and identification method, the precision of the non-penetration surface identification precision is low, and even the actual utilization value is lost;

thirdly, due to the influence of shadows generated by relief of terrain, tall buildings and crown shielding, the detection of a shadow region on a high-resolution remote sensing image and the discrimination of land types in the shadow region are difficult, the confusion between the shadows and water, plants and dark land objects is serious, the information in the shadow region is lost, the fine recognition of subsequent urban non-permeable surfaces is seriously influenced, and great difficulty is brought to the fine urban non-permeable surface recognition;

fourthly, even if the shadow area is repaired, certain spectral difference still exists between the shadow area and the non-shadow area, so that the shadow area and the non-shadow area need to be classified respectively, but the accuracy and the stability of the non-osmosis surface identification and classification method in the prior art cannot meet the requirements, the generalization error of a classifier cannot be converged, the over-fitting problem cannot be avoided, the processing capability of the remote sensing image data set with the missing processing characteristics is poor, multiple image characteristics cannot be selected based on the importance of the remote sensing image data set, and the artificial intervention is more; the anti-noise and shadow processing capabilities are poor, the prior art cannot process in parallel, the operation efficiency is low, and finally the feasibility and the precision of the remote sensing image non-penetration surface identification method are not guaranteed.

Disclosure of Invention

The method and the device fully consider the shadow shielding problem caused by high buildings and plant canopies in the remote sensing image, and firstly identify and measure the shadow area on the high-resolution remote sensing image; secondly, considering the problem of information loss of the shadow area, performing shadow repair on the shadow area; finally, after the shadow part of the high-resolution remote sensing image is repaired, an image object is obtained by adopting a multi-factor scale waveband segmentation method, various image characteristics are recognized, the urban non-permeable surface is recognized based on a multi-decision tree joint classifier, the problem of information loss of the building shadow part of the high-resolution remote sensing image is effectively solved, the recognition precision of the urban non-permeable surface is improved, the precision, the efficiency and the real-time performance can meet the application requirements of the current non-permeable surface, the method can be used for a large-range area, the data can be updated timely, the automation level is high, the recognition precision is higher and the visual angle range is wider compared with the medium-low-resolution remote sensing image, and the obtained urban impermeable layer data can completely meet the research and application of the aspects of impermeable layer, urban greenbelt, water area, urban heat island, waterlogging, water quality area source pollution and the like.

In order to achieve the technical effects, the technical scheme adopted by the application is as follows:

a remote sensing shadow measurement restoration image non-penetration surface identification method includes the steps of conducting preprocessing, shadow measurement and restoration on a high-resolution remote sensing initial image of a non-penetration surface identification area, then conducting object-oriented classification on fused images based on multi-decision tree combination, solving the problem that shadow influences information loss in the non-penetration surface identification process, obtaining an area high-precision non-penetration surface identification result, and conducting precision evaluation analysis:

step 1, preprocessing a high-resolution remote sensing image before non-penetration surface identification: the method comprises multi-factor scale band segmentation and spectrum different segmentation; after registration and orthorectification of the panchromatic image and the multispectral image, fusing the multispectral image with low spatial resolution and the panchromatic image with high spatial resolution to obtain the multispectral image with high spatial resolution, and cutting to obtain a processed region;

step 2, measuring the shadow of the object-oriented fusion multi-feature: based on the spectral characteristics of the shadow in each color space, an object-oriented shadow metering method combining multiple characteristic components is provided, and the first principal component of the shadow in PCA (principal component analysis) conversion, the I component in HIS (hue intensity distribution) conversion and the V component and c in HSV (hue intensity distribution) conversion are comprehensively adopted₁c₂c₃C3 component in the color space is used for shadow measurement, plants, water bodies and dark ground objects are distinguished from shadows, and NDVI, NDWI index and region texture features are fused to accurately measure shadow regions;

and 3, shadow repairing based on non-penetration surface identification: the method comprises the steps of controlling components in a blue waveband, setting a shadow homogeneous region, adjusting a HIS space and repairing a near infrared waveband, and performing corresponding shadow repairing and adjusting by adopting spectral characteristics of the HIS space, the shadow and the homogeneous region thereof based on blue waveband control repairing;

and 4, identifying the non-permeability surface based on multi-decision tree combination: the method comprises the steps of firstly voting by combining a series of non-permeable surface decision trees, then returning the previous voting result to the final non-permeable surface prediction result by adopting an integrated non-permeable surface classifier, randomly dividing each node of an individual decision tree in the integrated classifier, enabling each decision tree to depend on independent sampling and have random vector values which are distributed in the same way as all decision trees in the integrated classifier, and finally classifying each object based on the accumulated voting number of each decision tree when images are classified to obtain the final non-permeable surface recognition result.

A remote sensing shadow measurement restoration image non-penetration surface identification method is further designed based on a high-resolution remote sensing image, around measurement restoration of a shadow area on the high-resolution remote sensing image, interpretation of land types in the shadow area and refined urban non-penetration surface identification, and mainly comprises the following steps:

firstly, aiming at the problem that the identification of an urban impervious surface has uncertainty due to shadows generated by high-rise buildings and tree crowns on a high-resolution remote sensing image, a method for measuring and repairing the shadows of the high-resolution remote sensing image facing the impervious surface is provided, PCA (principal component analysis) transformation and HIS (high-intensity-localization) transformation are introduced to obtain various spectral characteristics based on the specific spectral characteristics of the remote sensing shadows, so that the identification and measurement of shadow areas are realized, and the repair of the urban shadow areas is realized based on blue-band component inhibition and HSI (high-speed information) space repair;

secondly, finely identifying the urban non-permeability surface by adopting an object-oriented method, wherein the object-oriented method is influenced by the segmentation parameters and has a large influence on the identification of the final non-permeability surface, analyzing the influence of the settings of different segmentation factors and parameters on the identification of the non-permeability surface, and finally identifying the urban non-permeability surface information on the high-resolution image by adopting a multi-decision tree joint classifier;

firstly, identifying and metering shadow areas on a high-resolution remote sensing image; secondly, considering the problem of information loss of the shadow area, performing shadow repair on the shadow area; and finally, after the shadow part of the high-resolution remote sensing image is repaired, an image object is obtained by adopting a multi-factor scale wave band segmentation method, various image characteristics are identified, the urban non-permeable surface is identified based on a multi-decision tree joint classifier, the problem of information loss of the building shadow part of the high-resolution remote sensing image is effectively solved, and the identification precision of the urban non-permeable surface is improved.

The method for identifying the non-penetration surface of the remote sensing shadow measurement restored image comprises the following steps of: by combining small segmentation objects, the method ensures that the average heterogeneity between the objects is minimum, and realizes the multi-factor scale band segmentation of the initial image by adopting region combination under the condition that the homogeneity requirement among pixels inside the objects is maximum;

the multi-factor scale band segmentation parameters set by the application comprise:

(1) the scale parameters are as follows: determining the maximum heterogeneity allowed by the object generated in the image segmentation process, wherein if the maximum heterogeneity value is larger, the size of the generated image object is larger, and if the maximum heterogeneity value is smaller, the generated image object is smaller;

(2) image band weight: standardizing the weight occupied by each wave band participating in the division;

(3) setting a heterogeneity factor: the heterogeneity factor is composed of spectral heterogeneity and shape heterogeneity, the sum of the two is 1, the shape heterogeneity is composed of compactness and smoothness, the sum of the two is also 1;

the image segmentation weight of the multi-factor scale band segmentation is set to be 1: 1: 1: 1: 0: 0: 0: 0, wherein 8 wave bands are blue, green, red, near infrared, PCA first principal component, I component, V component, c component₃Component, image segmentation scale set to 5, shape heterogeneity factor set to 0.1, and compactness set to 0.6.

The method for identifying the non-penetration surface of the remote sensing shadow measurement restored image further comprises the following steps of: optimizing on the basis of multi-factor scale band segmentation, and determining whether the brightness difference of adjacent segmented objects meets a set critical value or not by analyzing and judging to merge the target objects or not; firstly, performing multi-factor scale band segmentation on an image, further classifying and merging a plurality of objects in the multi-factor scale band segmented image by judging that a critical value meets a condition based on a multi-factor scale band segmentation result, merging target objects with small brightness value difference by adopting a brightness value index, and reducing the number of segmented objects;

the spectral dissimilarity partition contains two types of parameters:

(1) image band weight: standardizing the weight occupied by each wave band participating in the segmentation, and based on the weight of each image wave band;

(2) maximum spectral dissimilarity index: maximum spectral differences of adjoining segmented objects;

based on design experiments, the maximum spectrum dissimilarity index is set to be 5, object layers of the same ground object are combined, the problem of confusion of different ground objects is solved, and over-segmentation is improved.

The method for identifying the non-penetration surface of the remote sensing shadow measurement restored image further comprises the following steps of object fusion multi-feature oriented shadow measurement: respectively adopting principal component transformation, HIS transformation, HSV transformation and c_lc₂c₃Processing the remote sensing image by color transformation to obtain shadowsAccording to the characteristic components of each space, performing rough identification on the shadow by adopting a rule set facing an object according to the shadow characteristics, and performing fine identification on plant confusion, water body confusion and confusion of dark land features and shadows by combining NDVI, NDWI and texture characteristics; the principal component of PCA in the method is converted into a first principal component, and shadow measurement is roughly carried out on the first principal component of the PCA to be used as an initial value of precise measurement; the I component in the HSI transformation is obviously different from the characteristics of other ground features based on the illumination intensity of a shadow area, the shadow is distinguished by adopting the average value and the standard deviation of an object of the I component, but other types of characteristic components need to be introduced for eliminating the confusion problem of the dark ground features; v component in HSV conversion is removed, water body and dark ground object are removed, interference to shadow is weakened, but dark plants are easily measured as shadow area; c. C_lc₂c₃C in colour conversion₃Component, distinguishing shadow and non-shadow areas, but c₃The components are easy to be mistakenly divided into shadow areas for partial highlight areas and have instability, and the shadow metering errors are large when the components are singly used as the shadow metering errors, so that the components are cooperatively identified by combining other components; in addition, in the establishment of a rule set based on multi-feature component identification, the problem of confusion between shadows and plants is further weakened by adopting an NDVI index, the problem of confusion between water and plants is further weakened by adopting an NDWI index, the problem of confusion between dark ground objects and shadows is further weakened by adopting texture features, and finally a reliable shadow identification result is obtained.

A remote sensing shadow measurement repairing image non-penetration surface identification method further comprises the following main steps of:

the method comprises the following steps: performing principal component analysis on the fused image to obtain a first principal component;

step two: respectively converting the fused images into HIS, HSV and c_lc₂c₃Color space, respectively identifying I, V, c₃A feature component;

step three: based on different dimensions, the first principal component, the I component, the V component and the c are combined₃The total 4 characteristic components are normalized to [0-255]Interval, then combining with original multispectral image wave band to obtain multispectral image with 8 spectrumBlue, green, red, near infrared, PCA first principal component, I component, V component, c component₃A component;

step four: and (3) dividing the multi-factor scale wave band, and respectively setting the weight values as 1: 1: 1: 1: 0: 0: 0: 0, considering that the imperfect weights of the feature components to the original image information are all set to be 0, and reducing the redundancy of the information; the shape heterogeneity factor was set to 0.1: the tightness was set to 0.6;

step five: spectrum is divided differently, objects with relatively close brightness values are further merged, the phenomenon of over-division in multi-factor scale band division is improved, and the weight is set to be 1: 1: 1: 1: 0: 0: 0: 0, setting the maximum spectrum difference value to be 5;

step six: and identifying the shadow by adopting a multilevel rule set combination to obtain a shadow metering result.

The method for identifying the non-penetration surface of the remote sensing shadow measurement restored image further comprises the following steps of controlling components in a blue wave band: the method for carrying out fusion control on the blue band component in the RGB band of the remote sensing image comprises the following steps:

d and e are repair factors, e is set to be 1, d is set to be 0.6, and the purpose of controlling the blue band component is achieved, B (x, y) is the blue band component of the shadow area of the initial image, and B (x, y)' is the blue band component of the shadow area after control fusion.

The method for identifying the non-penetration surface of the remote sensing shadow metering restored image further comprises the following steps of: adjusting the brightness and color of a shadow area by adopting an HIS space, and improving the approximation degree between the shadow and the non-shadow;

in the HIS space, repairing three components H, S, I and a near-infrared wave band by adopting non-shadow regions around a shadow region as a reference system, assuming that a local region range of a remote sensing image is stable based on a homogeneous region, judging the non-shadow regions in a certain range around the shadow region as the homogeneous region of the shadow region through a rule system, wherein the statistical information of the non-shadow regions is consistent;

the ground feature types that the remote sensing image contains are numerous and complicated, and it is comparatively difficult to confirm the homogeneous region of shadow automatically, and the homogeneous region of shadow confirms the rule in this application and includes:

rule one is as follows: homogeneous regions contain no shadows;

rule two: the homogeneous region is adjacent to the shadow and is obtained according to the illumination projection direction;

rule three: an appropriate distance threshold is obtained to obtain a reasonably sized contiguous non-shaded region.

The identification method of the non-penetration surface of the remote sensing shadow measurement restored image, further, HIS space adjustment and near infrared band restoration: converting an image RGB wave band into an HIS color space, and performing repair correction on H, S, I, 3 components of a shadow region based on the mean value and standard deviation of a homogeneous non-shadow region corresponding to shadow, wherein the specific repair is based on formula 2:

and repairing and correcting the near infrared component Nir of the shadow region based on the mean value and the standard deviation of the homogeneous non-shadow region corresponding to the shadow, wherein the concrete repairing is based on a formula 3: :

wherein H ', S', I 'and NIR' are H, I, S and a near infrared band after repair respectively, n and g are a mean value and a standard deviation of a shadow region respectively, n 'and g' are a mean value and a standard deviation of a non-shadow region with the same shadow correspondingly, D, E, S, A is a repair intensity factor of each component, and a reference value region of the repair intensity factor is [0.6,1 ];

based on remote sensing shadow and non-infiltration surface HIS space characteristics, the shadow and the non-shadow have large difference on a brightness value I, the repair intensity coefficients are different, when the shadow identified by the non-infiltration surface is repaired, D ═ 0.8,1, [0.6,0.8], S ═ 0.8,1], A ═ 0.6,0.9, and homogeneous region distance critical values are all 50.

The method for identifying the non-penetration surface of the remote sensing shadow measurement restored image further comprises the following non-penetration surface identification steps based on multi-decision tree combination:

the first step is as follows: masking the high-resolution remote sensing image after shadow restoration by adopting a remote sensing image shadow measurement result;

the second step is that: carrying out multi-factor scale band segmentation and spectrum different segmentation on the non-shadow area of the repaired high-resolution remote sensing image, wherein the image segmentation weight is set as 1: 1: 1: 1: 0: 0: 0: 0, wherein 8 wave bands are blue, green, red, near infrared, PCA first principal component, I component, V component, c component₃Component, image segmentation scale is set to 5, shape heterogeneity factor is set to 0.1, compactness is set to 0.6, and maximum spectral dissimilarity index is set to 5;

the third step: carrying out multi-factor scale band segmentation and spectrum different segmentation on the shadow area of the repaired high-resolution remote sensing image, wherein the image segmentation weight is set as 1: 1: 1: 1: 0: 0: 0: 0, wherein 8 wave bands are blue, green, red, near infrared, PCA first principal component, I component, V component, c component₃Component, image segmentation scale is set to 5, shape heterogeneity factor is set to 0.1, compactness is set to 0.6, and maximum spectral dissimilarity index is set to 2;

the fourth step: selecting a training sample from the initial image, wherein the training sample comprises various characteristics of the high-resolution remote sensing image;

the fifth step: training and learning the multi-decision tree combined classifier by adopting various image characteristics in non-shadow and shadow samples respectively to obtain two multi-decision tree combined classifiers;

and a sixth step: classifying the non-shadow and shadow unclassified samples respectively by adopting corresponding multi-decision tree joint classifiers;

the seventh step: and (4) carrying out precision evaluation based on the remote sensing shadow measurement repairing image non-penetration surface identification result.

Compared with the prior art, the innovation points and advantages of the application are as follows:

firstly, with the help of a high-resolution remote sensing image, a set of high-resolution remote sensing image non-penetration surface identification method is designed around measurement restoration of a shadow area on the high-resolution remote sensing image, interpretation of land types in the shadow area and refined urban non-penetration surface identification: firstly, aiming at the problem that the identification of urban non-infiltration surfaces has uncertainty due to shadows generated by high-rise buildings and tree crowns on a high-resolution remote sensing image, a method for metering and repairing the shadows of the high-resolution remote sensing image facing the non-infiltration surfaces is provided, PCA (principal component analysis) transformation and HIS (high-intensity-localization) transformation are introduced to obtain various spectral characteristics based on the specific spectral characteristics of the remote sensing shadows, so that the identification and metering of shadow areas are realized, and the repair of the urban shadow areas is realized based on blue-band component inhibition and HSI (high-speed integrated information) space repair; secondly, the urban non-permeability surface is finely identified by adopting an object-oriented method, in order to better obtain an urban non-permeability surface identification result, the influence of the settings of different segmentation factors and parameters on the non-permeability surface identification is analyzed, and finally, urban non-permeability surface information on a high-resolution image is identified by adopting a multi-decision tree combined classifier, and an experimental result shows that the integral non-permeability surface identification accuracy is 96.38%, so that a powerful guarantee is provided for the research and application of the non-permeability surface;

secondly, the problem of shadow shielding caused by high buildings and plant canopies in the remote sensing image is fully considered, and the shadow area on the high-resolution remote sensing image is identified and measured; secondly, considering the problem of information loss of the shadow area, performing shadow repair on the shadow area; finally, after the shadow part of the high-resolution remote sensing image is repaired, an image object is obtained by adopting a multi-factor scale waveband segmentation method, various image characteristics are recognized, the urban non-permeable surface is recognized based on a multi-decision tree joint classifier, the problem of information loss of the building shadow part of the high-resolution remote sensing image is effectively solved, the recognition precision of the urban non-permeable surface is improved, the precision, the efficiency and the real-time performance can meet the application requirements of the current non-permeable surface, the method can be used for a large-scale area, the data can be updated timely, the automation level is high, and compared with the low-and-medium-resolution remote sensing image, the recognition precision is higher, the visual angle range is wider, and the obtained urban impermeable layer data can completely meet the research and application of the impermeable layer and the aspects of urban greenbelt, water area, urban heat island, waterlogging, water quality area source pollution and the like;

thirdly, the multi-factor scale band segmentation of the initial image is realized by adopting region combination under the condition that the average heterogeneity between objects is minimum and the homogeneity requirement among pixels in the objects is maximum by combining small segmented objects; spectrum different segmentation is optimized on the basis of multi-factor scale band segmentation, whether the brightness difference of adjacent segmented objects meets a set critical value is judged through analysis, whether target objects are combined or not is determined, the target objects with smaller brightness difference are combined by adopting a brightness value index, and the number of the segmented objects is reduced; the object-oriented multi-feature fusion shadow metering method fully adopts the spectral features of remote sensing shadows, effectively makes up the limitation of low shadow metering precision caused by single feature components, and effectively solves the problem of confusion of water bodies, plants, dark ground objects and shadows; the classification advantages of near-infrared wave bands on water bodies and plants during classification are considered, a near-infrared cooperative repair method is introduced, further repair is carried out in an HIS color space, a high-quality shadow repair result is obtained, support is provided for further non-permeable surface identification, shadow repair efficiency is greatly improved, non-permeable surface characteristics can be identified more quickly, and the method is high in sensitivity, reliability and practicability;

fourthly, the multi-decision tree combined classifier is an integrated non-permeable surface classifier, the classification precision of the non-permeable surface of the multi-decision tree combined classifier depends on the accuracy of each decision tree and the mutual independence between each decision tree, as long as the classification precision of a single decision tree and the mutual independence between a plurality of decision trees are ensured, the classification precision of the integrated non-permeable surface classifier under the same condition can be ensured to be higher than the precision of the single classifier, the synergistic action between a plurality of classifiers ensures that the stability of the classification result is stronger, compared with the single decision tree, because the number of the decision trees in the multi-decision tree combined classifier is numerous, the generalization error of the integrated non-permeable surface classifier is converged, and the advantages of the multi-decision tree combined classifier also comprise: firstly, the over-fitting problem can be well avoided; secondly, the remote sensing image data set with the processing characteristics lost still has good processing capacity; thirdly, aiming at various image characteristics, the selection can be carried out based on the importance of the image characteristics, so that the human intervention is reduced; fourthly, the method has better anti-noise and shadow processing capability; fifthly, the parallel processing can be realized, and the calculation efficiency is high. The non-permeability surface recognition is feasible and efficient by adopting the multi-decision tree combined classifier, and the recognition accuracy reaches 96.38%.

Drawings

FIG. 1 is a diagram of the parameter settings of a multi-factor scale band splitting experiment.

FIG. 2 is a diagram of the result of the A, B, C parameter multi-factor scale band segmentation.

FIG. 3 is a diagram of the partial result of the multi-factor scale band segmentation with parameter D.

Fig. 4 is a schematic diagram of parameter settings of a spectrum dissimilarity segmentation experiment. .

FIG. 5 is a diagram showing the results of the different division of the parameter spectra of a, b and c.

FIG. 6 is a diagram of the main steps of the object-oriented fusion multi-feature shadow metering method.

FIG. 7 is a schematic diagram of the whole process of shadow repair facing non-penetration surface identification.

FIG. 8 is a graph of experimental results of the A-th image of the multi-feature shadow measurement method at various stages.

FIG. 9 is a graph of experimental results of stages of a B-th image of a multi-feature shadow metering method.

FIG. 10 is a flow chart of a non-permeability surface identification method based on multi-decision tree union.

Detailed description of the invention

The technical scheme of the method for identifying the non-penetration surface of the remote sensing shadow metering repair image provided by the application is further described below with reference to the accompanying drawings, so that a person skilled in the art can better understand and implement the application.

The urban impervious layer is one of important indexes for measuring urban ecological environment and sustainable development, and currently, along with the acceleration of an urbanization process, the ratio of the urban impervious layer is gradually increased, so that urban greenbelt and water body area are reduced, a series of negative effects are generated on the urban ecological environment, such as urban heat island, waterlogging, water quality non-point source pollution and the like, and the life of residents and the urban hydrological ecological structure are seriously affected. The traditional manual mapping method is huge in acquisition cost of the city impermeable surface, poor in real-time performance and difficult in data updating. In recent years, with the advantages of macroscopic, rapid and strong periodicity of a satellite remote sensing earth observation technology, the method for acquiring the information of the urban non-permeable surface becomes an application hotspot, and the traditional low-and-medium-resolution remote sensing image mainly takes the spectral information of a pixel as a main part and is mainly used for identifying the non-permeable surface at the global or regional level. The high-spatial-resolution remote sensing image has the advantages of clear ground object boundaries, capability of more clearly expressing ground object targets and the like, and is more favorable for identifying urban fine-scale non-penetration surfaces. However, due to the influence of the shadow caused by the relief of the terrain, tall buildings and crown occlusion, the fine city non-infiltration surface identification is difficult. Therefore, by means of the high-resolution remote sensing image, around metering restoration of a shadow area on the high-resolution remote sensing image, interpretation of land types in the shadow area and refined urban non-permeable surface identification, a set of high-resolution remote image non-permeable surface identification method is designed, and the method mainly comprises the following steps:

the method for measuring and repairing the shadow of the high-resolution remote sensing image facing the non-penetration surface is provided for solving the problem that the measurement of the shadow area and the discrimination of the land type in the shadow area seriously affect the fine identification of the subsequent urban non-penetration surface on the high-resolution remote sensing image, so that the problem that the identification of the urban non-penetration surface has uncertainty due to the shadow generated by a high-rise building and a tree crown on the high-resolution remote sensing image is solved. Experiments show that the method can effectively identify and repair shadow areas shielded by high buildings and crowns on the high-resolution remote sensing images.

The city non-permeability surface is finely identified by adopting an object-oriented method, and the object-oriented method is influenced by segmentation parameters and has a great influence on the final identification of the non-permeability surface, so that in order to better obtain the city non-permeability surface identification result, the influence of the settings of different segmentation factors and parameters on the non-permeability surface identification is analyzed, and finally, the city non-permeability surface information on a high-resolution image is identified by adopting a multi-decision tree joint classifier;

the method and the device fully consider the shadow shielding problem caused by high buildings and plant canopies in the remote sensing image, and firstly identify and measure the shadow area on the high-resolution remote sensing image; secondly, considering the problem of information loss of the shadow area, performing shadow repair on the shadow area; and finally, after the shadow part of the high-resolution remote sensing image is repaired, an image object is obtained by adopting a multi-factor scale wave band segmentation method, various image characteristics are identified, and the urban non-permeable surface is identified based on a multi-decision tree joint classifier, so that the problem of information loss of the building shadow part of the high-resolution remote sensing image is effectively solved, and the identification precision of the urban non-permeable surface is improved.

Identification process of non-penetration surface of remote sensing shadow measurement repairing image

Preprocessing, shadow metering and repairing a high-resolution remote sensing initial image of a non-penetration surface identification area, then carrying out object-oriented classification on the fused image based on multi-decision tree combination, solving the influence of shadow on information loss in the non-penetration surface identification process, obtaining an area high-precision non-penetration surface identification result, and carrying out precision evaluation analysis:

step 1, preprocessing a high-resolution remote sensing image before non-penetration surface identification: the method comprises multi-factor scale band segmentation and spectrum different segmentation; after registration and orthorectification of the panchromatic image and the multispectral image, fusing the multispectral image with the spatial resolution of 3.2 meters and the panchromatic image with the spatial resolution of 0.8 meter to obtain the multispectral image with the spatial resolution of 0.8 meter, and cutting to obtain a processed area;

step 2, measuring the shadow of the object-oriented fusion multi-feature: based on the spectral characteristics of the shadow in each color space, an object-oriented shadow metering method combining multiple characteristic components is provided, and a first principal component of the shadow in PCA conversion and an I component in HIS conversion are comprehensively adopted,V component and c in HSV conversion₁c₂c₃Shadow measurement is carried out on the c3 component in the color space, plants, water bodies and dark ground objects are distinguished from shadows, NDVI, NDWI indexes and region texture features are fused to accurately measure shadow regions, and the method is reflected by recognition results of images of different scenes, so that the method has better universality and stability;

and 4, identifying the non-permeability surface based on multi-decision tree combination: the method comprises the steps of firstly voting by combining a series of non-permeable surface decision trees, then returning the previous voting result to the final non-permeable surface prediction result by adopting an integrated non-permeable surface classifier, randomly dividing each node of an individual decision tree in the integrated classifier, enabling each decision tree to depend on independent sampling and have random vector values which are distributed in the same way as all decision trees in the integrated classifier, and finally classifying each object based on the accumulated voting number of each decision tree when classifying unknown images to obtain the final non-permeable surface recognition result.

Preprocessing high-resolution remote sensing image before non-penetration surface recognition

After registration and orthorectification of the panchromatic image and the multispectral image, the multispectral image with the spatial resolution of 3.2 meters and the panchromatic image with the spatial resolution of 0.8 meter are fused to obtain the multispectral image with the spatial resolution of 0.8 meter, and the multispectral image is cut to obtain a processed area.

(one) multi-factor scale band segmentation

By combining small segmentation objects, the method ensures that the average heterogeneity between the objects is minimum, and realizes the multi-factor scale band segmentation of the initial image by adopting region combination under the condition that the homogeneity requirement among pixels in the objects is maximum.

The following types of parameters are related to the multi-factor scale band segmentation information:

(1) the scale parameters are as follows: determining the maximum heterogeneity allowed by the object generated in the image segmentation process, wherein the larger the maximum heterogeneity value is, the larger the size of the generated image object is, and the smaller the maximum heterogeneity value is, the smaller the generated image object is.

(2) Image band weight: the weights of all the bands participating in segmentation are normalized, the weights of all the bands are close, but when a certain type of thematic information is identified or a certain band contains more image information, a special band needs to be set with a larger weight, and other bands with smaller information quantity are set with a smaller weight or even set as 0 to not participate in segmentation.

(3) Setting a heterogeneity factor: the heterogeneity factor is composed of spectral heterogeneity and shape heterogeneity, the sum of both is 1, the shape heterogeneity is composed of compactness and smoothness, the sum of both is also 1.

The spectral heterogeneity parameters are set to be larger, meanwhile, reasonable shape heterogeneity parameters are set to ensure the integrity of the image object, so that the image object has good geometric characteristics, the spectral heterogeneity factor is set to be about 0.8, the corresponding shape heterogeneity factor is about 0.2, the smoothness factor is set to be about 0.6, the corresponding compactness factor is about 0.4, the generated image object layer is compared with actual ground objects by multiple attempts, the proportion of assimilation objects in the image object layer is reduced as much as possible, the assimilation pixels in the image object layer are ensured to be the minimum, and the influence of various parameter selections on the remote sensing image is researched according to the graph 1.

The image segmentation weight is set to be 1: 1: 1: 1: 0: 0: 0: 0, wherein 8 wave bands are blue, green, red, near infrared, PCA first principal component, I component, V component, c component₃The component is that according to the A number segmentation parameters such as the graph, roads and trees, buildings and trees, tree shadows and trees, and buildings and trees cannot be well distinguished, which causes errors on subsequent identification, so the adjustment segmentation scale is adopted, the B number segmentation parameters are compared with the A number segmentation parameters, and as a result, plants and roads, buildings and trees are well distinguished, but the shadow and trees are still seriously confused, and the C number segmentation parameters can well distinguish various land featuresThe situation of assimilating pixels in the object layer is rare, so the image segmentation scale is selected to be 5 rather suitable, the segmentation scale is not suitable to be reduced again, and if a smaller segmentation scale is adopted, various ground objects can be well distinguished, but the workload is increased when the ground objects are severely fragmented. A. The graph of the result of the B, C scale division is shown in FIG. 2.

Then, an experiment for setting the heterogeneity factor parameter is performed, and based on the priori knowledge, the shape heterogeneity factor is set to about 0.2, and the compactness is set to about 0.6, so that the segmentation situation when the shape heterogeneity factor is respectively 0.1, 0.2, and 0.3, and the compactness is respectively 0.5, 0.6, and 0.7 is discussed. The result comparison shows that the image segmentation result of the parameter D can not only ensure that the assimilation pixel in the image object layer is minimum, but also truly and objectively reflect the actual ground object coverage condition in the image, as shown in FIG. 3.

Therefore, the image segmentation weight of the multi-factor scale band segmentation is set to 1: 1: 1: 1: 0: 0: 0: 0, wherein 8 wave bands are blue, green, red, near infrared, PCA first principal component, I component, V component, c component₃Component, image segmentation scale set to 5, shape heterogeneity factor set to 0.1, and compactness set to 0.6.

(II) spectral dissimilarity segmentation

Optimizing on the basis of multi-factor scale band segmentation, and determining whether the brightness difference of adjacent segmented objects meets a set critical value through analysis so as to determine whether to merge the target objects; firstly, multi-factor scale band segmentation is carried out on an image, based on a multi-factor scale band segmentation result, a plurality of objects in the multi-factor scale band segmentation image are further classified and combined by judging that a critical value meets a condition, and a target object with small brightness value difference is combined by adopting a brightness value index, so that the number of segmented objects is reduced.

The spectral dissimilarity partition contains two types of parameters:

(2) maximum spectral dissimilarity index: for selecting a suitable spectrum dissimilarity segmentation parameter, the scale experiment is as shown in fig. 4, and the weight 1: 1: 1: 1: 0: 0: 0: 0 and the experimental parameters of the maximum spectral dissimilarity index are respectively 2, 5, 10, 25 and 50.

As shown in fig. 5, when the maximum spectral dissimilarity index is set to 2, many ground objects are still fragmented, objects cannot be combined well, especially in a shadow region, shape information of various ground objects cannot be highlighted, and when the maximum spectral difference value is set to 10, object layers of different classes are also classified into one class, so that the maximum spectral dissimilarity index is not large, and the situation that the spectrum dissimilarity segmentation results of the number d and the number e are combined into one class of ground objects is more serious, so that a parameter experiment of the number b is performed, the experiment better combines the object layers of the same class of ground objects, and the problem of confusion of different classes of ground objects is avoided, so that the phenomenon of over-segmentation is improved to a certain extent.

Therefore, the maximum spectral dissimilarity index is set to 5.

Object-oriented fusion multi-feature shadow metering

Shadow classification on remote sensing images

The shadow is widely distributed in the high-resolution remote sensing image data, is a basic component of the ground features of the high-resolution remote sensing image, and is mainly caused by that an object with elevation blocks sunlight so that partial area in the sun illumination direction cannot receive sun illumination light and only partial sun scattering light can cause, and therefore the shadow area has a lower gray value.

The space position of the shadow generated by the shielding of sunlight by the ground object is divided into a main shadow and a projection, the main shadow is generated because the position of the shielding object opposite to the light source is not irradiated by the sunlight, the projection is formed because the sunlight irradiation light is shielded by the ground object with a certain elevation, and the shadow is formed on the surface of the other object, so the spectral brightness values of the main shadow and the projection are different, the brightness value of the shadow in the image is derived from the indirect sunlight reflectivity of the ground object at the periphery, the main shadow receives more indirect sunlight light sources, the brightness value of the main shadow is greater than the brightness value of the projection, the projection is divided into a whole shadow and a semi shadow, the whole shadow is a background area in which the sunlight is completely shielded, and the semi shadow is a background area in which the light is partially shielded. The remote sensing image shadow is a projection area, including a total shadow area and a penumbra area, and is generated by covering a building shadow and a plant.

(II) remote sensing spectral characteristics of shadows

The spectral characteristics of the high-resolution remote sensing shadow area relative to the non-shadow ground object mainly stand out the following five points:

(1) under the same imaging condition, the brightness value of the shadow area of the image is far smaller than that of the normal non-shadow area;

(2) the color characteristics of the ground objects in the shadow area are stable and are not influenced and changed by the shadow;

(3) the shaded area has low frequency characteristics;

(4) shadow region texture feature invariance;

(5) the shadow area is highly saturated.

(III) shadow metering method for object-oriented fusion of multiple features

Respectively adopting principal component transformation, HIS transformation, HSV transformation and c_lc₂c₃Carrying out color transformation on the remote sensing image to obtain characteristic components of the shadow in each space, carrying out rough identification on the shadow by adopting a rule set facing an object according to shadow characteristics, and carrying out fine identification on plant confusion, water body confusion and confusion of dark land plants and the shadow by combining NDVI, NDWI and texture characteristics; the principal component of PCA in the method is converted into a first principal component, and shadow measurement is roughly carried out on the first principal component of the PCA to be used as an initial value of precise measurement; the I component in the HSI transformation is obviously different from the characteristics of other ground features based on the illumination intensity of a shadow area, the shadow is distinguished by adopting the average value and the standard deviation of an object of the I component, but other types of characteristic components need to be introduced for eliminating the confusion problem of the dark ground features; v component in HSV conversion can better eliminate water body and dark ground objects, reduce interference to shadow, but easily measure dark plants as shadow areas; c. C_lc₂c₃C in colour conversion₃Component, effective to distinguish between shadow and non-shadow areas, but c₃The components are easily mistakenly divided into shadow areas for partial highlight areasInstability exists, the error is large when the shadow is used alone as a shadow metering error, and the shadow is combined with other components for cooperative identification; in addition, in the establishment of a rule set based on multi-feature component identification, the problem of confusion between shadows and plants is further weakened by adopting an NDVI index, the problem of confusion between water and plants is further weakened by adopting an NDWI index, the problem of confusion between dark ground objects and shadows is further weakened by adopting texture features, and finally a reliable shadow identification result is obtained.

As shown in fig. 6, the main steps of the object-oriented fusion multi-feature shadow metering method include:

step three: based on different dimensions, the first principal component, the I component, the V component and the c are combined₃The total 4 characteristic components are normalized to [0-255]Combining the interval with original multispectral image band to obtain multispectral image with 8 spectra, which are blue, green, red, near infrared, PCA first principal component, I component, V component, c component₃A component;

The method fully adopts the spectral characteristics of the remote sensing shadow, effectively makes up the limitation of low shadow metering precision caused by single characteristic component, and effectively solves the problem of confusion of water, plants, dark ground objects and the shadow.

(IV) shadow measurement Experimental analysis

In order to verify the universality and the accuracy of the object-oriented multi-feature fusion shadow metering method, two GF2 images with different ground object coverage situations are selected, wherein the shadow complexity situations are different. According to the difference of the shadow distribution conditions, the two images can be mainly divided into: the B image has a complex shadow image with more water interference, and the A image is a simple shadow image without water interference. In the image B, in a dense urban area, tall buildings and bungalows are densely staggered, dark ground objects are more, the plant coverage density is higher, and the shadow metering result shows that the method can better distinguish the plants from the shadows and the dark ground objects from the shadows. A large amount of water, plants and a small amount of dark ground objects are arranged in the image A, and the experimental result shows that a large amount of water is distinguished from the shadow, so that the situation of water and shadow confusion can be better dealt with.

Shadow repair based on non-permeability surface identification

The reduction of the information quantity in the ground shadow area is lacked, so that the interpretation obstacle of the subsequent non-permeable surface classification is caused. The shadow has great influence on the classification of the non-permeable surface, and how to eliminate the influence caused by the lack of shadow region information is a difficult problem of remote sensing image processing.

Because the blue band control shadow restoration method based on the RGB space is independently adopted and only the blue band is controlled, the saturation and brightness value of the shadow are not effectively restored, and the classification advantages of the near-infrared band on water and plants during classification are considered, a near-infrared cooperative restoration method is introduced on the basis, and further restoration is selected in the HIS color space to obtain the final shadow restoration result, so that support is provided for further non-osmotic surface identification.

As shown in fig. 7, the entire flow chart of shadow repair facing non-penetration surface identification includes: blue wave band control component, shadow homogeneous area setting, HIS space adjustment and near infrared wave band repair.

Blue band control component

The method for carrying out fusion control on the blue band component in the RGB band of the remote sensing image comprises the following steps:

(II) setting shadow homogeneous region

Only the blue band component is controlled, and other spectral characteristics of the shadow, such as saturation, are not comprehensively considered, so that the shadow repair cannot be effectively carried out. Therefore, the brightness and the color of the shadow area are adjusted by adopting the HIS space, and the approximation degree between the shadow and the non-shadow is improved.

According to the method, the H, S, I three components and the near-infrared waveband are repaired in the HIS space, the non-shadow area around the shadow area is used as a reference system, the local area range of the remote sensing image is stable on the basis of the homogeneous area, the non-shadow area around the shadow area in a certain range is judged to be the homogeneous area of the shadow area through a rule system, and the statistical information of the non-shadow area and the homogeneous area is consistent.

rule one is as follows: homogeneous regions contain no shadows;

(III) HIS spatial modulation and near-infrared band repair

Converting an image RGB wave band into an HIS color space, and performing repair correction on H, S, I, 3 components of a shadow region based on the mean value and standard deviation of a homogeneous non-shadow region corresponding to shadow, wherein the specific repair is based on formula 2:

wherein, H ', S', I 'and NIR' are H, I, S and a near infrared band after repair respectively, n and g are mean values and standard deviations of shadow areas respectively, n 'and g' are mean values and standard deviations of non-shadow areas with the same shadow quality correspondingly, D, E, S, A is repair intensity factors of each component, and the reference value area of the repair intensity factors is [0.6,1 ].

(IV) Experimental analysis and Effect evaluation

Based on the shadow repairing method facing to the non-penetration surface identification, the information repairing problem of the HIS color space is considered, and a shadow repairing image is obtained. The results are analyzed visually and evaluated in effect, and experiments prove that the method can effectively repair the real information of the shadow area.

And respectively repairing the shadow of the two images by adopting the shadow repairing method, and analyzing the experimental result.

As shown in fig. 8, in the a-th image, the contrast between the shadow area and the non-shadow area after shadow repair is reduced, so that the shadow area information is richer, which affects quality improvement, but the shadow repair effect in the shadow area with a more complex homogeneous area is slightly poor. In (c) and (d), shadow repair is good, so that shadow region information is richer: in the step (f), although the shadow candidate homogeneous areas are buildings, roads and playgrounds respectively, the attribute difference is large, but the shadow repairing effect is still good.

As shown in fig. 9, in the B-th image, the brightness value of the repaired shadow area is significantly improved, and for example, in (B) and (d), especially near a tall building, the information is well repaired, and the color tone of the shadow area is closer to that of the non-shadow area, the repair effect of the shadow area including part of (d) with higher difficulty is ideal, and the contrast between the shadow and the non-shadow is significantly reduced. The shadow area information is effectively recovered, the interpretation capability of the remote sensing image is enhanced, and a foundation is provided for the shadow problem of non-penetration surface identification.

Fourth, non-penetration surface identification based on multi-decision tree combination

The method comprises the steps of firstly voting by combining a series of non-permeable surface decision trees, then returning the previous voting result to the final non-permeable surface prediction result by adopting an integrated non-permeable surface classifier, randomly dividing each node of an individual decision tree in the integrated classifier, enabling each decision tree to depend on independent sampling and have random vector values which are distributed in the same way as all decision trees in the integrated classifier, and finally classifying each object based on the accumulated voting number of each decision tree when classifying unknown images to obtain the final non-permeable surface recognition result.

Because the multi-decision-tree combined classifier is an integrated non-penetration surface classifier, the classification precision of the non-penetration surface of the multi-decision-tree combined classifier depends on the accuracy of each decision-tree and the mutual independence among the decision-trees, the classification precision of the integrated non-penetration surface classifier under the same condition can be ensured to be higher than that of a single classifier as long as the shadow classification precision of the single decision-tree and the mutual independence among a plurality of decision-trees are ensured, and the stability of the classification result is stronger due to the synergistic action among a plurality of classifiers. Compared with a single decision tree, the number of decision trees in the multi-decision tree combination is large, so that the generalization error of the integrated non-permeability surface classifier is converged, and the problem of overfitting can be well avoided, and the method has the following advantages: firstly, the over-fitting problem can be well avoided; secondly, the remote sensing image data set with the processing characteristics lost still has good processing capacity; thirdly, aiming at various image characteristics, the selection can be carried out based on the importance of the image characteristics, so that the human intervention is reduced; fourthly, the method has better anti-noise and shadow processing capability; fifthly, the parallel processing can be realized, and the calculation efficiency is high. Therefore, the non-penetration surface identification of the high-resolution remote sensing image is feasible by adopting the multi-decision tree combined classifier, and the identification accuracy rate reaches 96.38%.

Non-permeability surface identification method based on multi-decision tree combination

The non-permeability surface identification step based on multi-decision tree combination mainly comprises the following steps:

(II) non-permeability surface identification experiment analysis and precision evaluation

And (3) adopting non-penetration surface identification based on multi-decision tree combination, and classifying the non-shadow area and the shadow repairing area respectively in consideration of the fact that although the information is rich after the shadow repairing, the non-shadow area has spectral difference with the non-shadow area. As shown in fig. 10.

On the basis of the original method, the application purpose of classification is considered, and the image near-infrared band is repaired to obtain a shadow repair image. The result is carried out through two aspects of visual analysis and effect evaluation, and experiments prove that the method is effective in repairing the missing information of the shadow area.

By identifying the non-penetration surface of the high-resolution remote sensing image, the method can ensure higher identification precision in the non-shadow area and can acquire more non-penetration surface information.

Aiming at non-penetration surface identification of a high-resolution remote sensing image, the shadow classification problem in a data source is mainly considered, shadow metering and repairing are adopted, and a multi-decision tree combined classifier is combined to classify shadow areas. Firstly, a shadow metering method based on multi-feature components is provided in the aspect of shadow metering, secondly, for the problem of high-resolution remote sensing image classification, shadow areas are classified, the recognition capability of a non-permeable surface is improved, and finally, a multi-decision tree joint classifier is adopted to classify and recognize the high-resolution remote sensing image, and the recognition accuracy rate reaches 96.38%.

Claims

1. The method for identifying the non-permeable surface of the remote sensing shadow measurement restored image is characterized in that a high-resolution remote sensing initial image of a non-permeable surface identification area is subjected to preprocessing, shadow measurement and restoration, then the fused image is subjected to object-oriented classification based on multi-decision tree combination, the influence of shadow on information loss in the non-permeable surface identification process is solved, and a high-precision non-permeable surface identification result of the area is obtained and subjected to precision evaluation analysis:

2. The remote sensing shadow metering repairing image non-penetration surface identification method according to claim 1, wherein a set of high-resolution remote sensing image non-penetration surface identification method is designed based on a high-resolution remote sensing image, around metering repairing of a shadow region on the high-resolution remote sensing image, interpretation of land types in the shadow region and refined urban non-penetration surface identification, and mainly comprises the following steps:

3. The method for identifying the non-penetration surface of the remote sensing shadow metering inpainting image as claimed in claim 1, wherein the multi-factor scale wave band segmentation comprises the following steps: by combining small segmentation objects, the method ensures that the average heterogeneity between the objects is minimum, and realizes the multi-factor scale band segmentation of the initial image by adopting region combination under the condition that the homogeneity requirement among pixels inside the objects is maximum;

4. The method for identifying the non-penetration surface of the remote sensing shadow metering repair image as claimed in claim 1, wherein the spectrum is divided differently: optimizing on the basis of multi-factor scale band segmentation, and determining whether the brightness difference of adjacent segmented objects meets a set critical value or not by analyzing and judging to merge the target objects or not; firstly, performing multi-factor scale band segmentation on an image, further classifying and merging a plurality of objects in the multi-factor scale band segmented image by judging that a critical value meets a condition based on a multi-factor scale band segmentation result, merging target objects with small brightness value difference by adopting a brightness value index, and reducing the number of segmented objects;

the spectral dissimilarity partition contains two types of parameters:

5. The method for identifying the non-penetration surface of the remote sensing shadow measurement restoration image according to claim 1, wherein the object-oriented fusion multi-feature shadow measurement comprises the following steps: respectively adopting principal component transformation, HIS transformation, HSV transformation and c_lc₂c₃Carrying out color transformation on the remote sensing image to obtain characteristic components of the shadow in each space, carrying out rough identification on the shadow by adopting a rule set facing an object according to shadow characteristics, and carrying out fine identification on plant confusion, water body confusion and confusion of dark land plants and the shadow by combining NDVI, NDWI and texture characteristics; the principal component of PCA in the method is converted into a first principal component, and shadow measurement is roughly carried out on the first principal component of the PCA to be used as an initial value of precise measurement; the I component in the HSI transformation is obviously different from the characteristics of other ground features based on the illumination intensity of a shadow area, the shadow is distinguished by adopting the average value and the standard deviation of an object of the I component, but other types of characteristic components need to be introduced for eliminating the confusion problem of the dark ground features; v component in HSV conversion is removed, water body and dark ground object are removed, interference to shadow is weakened, but dark plants are easily measured as shadow area; c. C_lc₂c₃C in colour conversion₃Component, distinguishing shadow and non-shadow areas, but c₃The components are easy to be mistakenly divided into shadow areas for partial highlight areas and have instability, and the shadow metering errors are large when the components are singly used as the shadow metering errors, so that the components are cooperatively identified by combining other components; in addition, in the establishment of a rule set based on multi-feature component identification, the problem of confusion of shadow and plants is further weakened by adopting an NDVI (normalized difference of magnitude) index, the problem of confusion of water and plants is further weakened by adopting an NDWI index, the problem of confusion of dark ground objects and shadows is further weakened by adopting texture features, and finally the problem of confusion of dark ground objects and shadows is obtainedAnd (4) reliable shadow identification results.

6. The method for identifying the non-penetration surface of the remote sensing shadow metering inpainting image according to claim 5, wherein the method for metering the shadow of the object-oriented fusion multi-feature comprises the following main steps:

7. The method for identifying the non-penetration surface of the remote sensing shadow metering repair image according to claim 1, wherein the blue band control component: the method for carrying out fusion control on the blue band component in the RGB band of the remote sensing image comprises the following steps:

8. The method for identifying the non-penetration surface of the remote sensing shadow metering repair image according to claim 1, wherein a shadow homogeneous region is set: adjusting the brightness and color of a shadow area by adopting an HIS space, and improving the approximation degree between the shadow and the non-shadow;

rule one is as follows: homogeneous regions contain no shadows;

9. The method for identifying the non-penetration surface of the remote sensing shadow metering restoration image according to claim 1, wherein HIS spatial adjustment and near infrared band restoration are as follows: converting an image RGB wave band into an HIS color space, and performing repair correction on H, S, I, 3 components of a shadow region based on the mean value and standard deviation of a homogeneous non-shadow region corresponding to shadow, wherein the specific repair is based on formula 2:

10. The method for identifying the non-penetration surface of the remote sensing shadow metering inpainting image according to claim 1, wherein the non-penetration surface identification step based on multi-decision tree combination mainly comprises the following steps: