CN108230376A

CN108230376A - Remote sensing image processing method, device and electronic equipment

Info

Publication number: CN108230376A
Application number: CN201611264368.4A
Authority: CN
Inventors: 李聪
Original assignee: Beijing Sensetime Technology Development Co Ltd
Current assignee: Beijing Sensetime Technology Development Co Ltd
Priority date: 2016-12-30
Filing date: 2016-12-30
Publication date: 2018-06-29
Anticipated expiration: 2036-12-30
Also published as: CN108230376B

Abstract

This application discloses remote sensing image processing method, device and electronic equipments.Method includes：The feature and the feature with reference to remote sensing images that even color remote sensing images are treated in matching, obtain multiple features pair；Based on multiple features pair, Feature Mapping relationship between image is determined；According to Feature Mapping relationship between image, determine to treat even color remote sensing images and with reference to the overlapping region between remote sensing images；According to the overlapping region and the overlapping region with reference to remote sensing images for treating even color remote sensing images, treat even color remote sensing images and carry out even color processing, obtain the remote sensing images after even color.The embodiment eliminates a large amount of unnecessary redundant datas in the processing procedure of remote sensing images, improves the treatment effeciency of remote sensing images and improves the processing accuracy of remote sensing images.

Description

Remote sensing image processing method, device and electronic equipment

Technical field

This application involves field of computer technology, and in particular to computer image processing technology field more particularly to remote sensing Image processing method, device and electronic equipment.

Background technology

In the application of remote sensing images, it is often necessary to the image of complete covering survey region is obtained, due to most of image It usually obtains, exists by the factors image such as season, weather, illumination and different zones, between each image apparent under different phases Color and the differences such as brightness.If different influence direct splicings is complete image, visual effect is poor, while many qualitative Or in quantitative remote sensing image information interpretation business, seriously affect information extraction precision and extraction efficiency.

In order to improve the quality of remote sensing image processing, in traditional remote sensing image processing, usually as follows into Row remote sensing image processing：First according to pending remote sensing images, the reference remote sensing images close or similar to its are selected, it is right later It treats even color remote sensing images and is matched with reference to remote sensing images, pending remote sensing images and reference are spliced according to matching result later Remote sensing images.Wherein, when carrying out images match, mainly using geographical coordinate registration and feature registration two schemes.

Invention content

The application proposes a kind of technical solution of remote sensing image processing.

In a first aspect, this application provides a kind of remote sensing image processing method, method includes：Even color remote sensing images are treated in matching Feature with reference to remote sensing images feature, obtain multiple features pair；Based on multiple features pair, Feature Mapping is closed between determining image System；According to Feature Mapping relationship between image, determine to treat even color remote sensing images and with reference to the overlapping region between remote sensing images；According to The overlapping region and the overlapping region with reference to remote sensing images for treating even color remote sensing images, treat even color remote sensing images and carry out at even color Reason, obtains the remote sensing images after even color.

In some embodiments, it according to Feature Mapping relationship between image, determines to treat even color remote sensing images with referring to remote sensing figure Overlapping region as between includes：According to multiple features pair, determine to treat even color remote sensing images and with reference to first between remote sensing images Beginning overlapping region；It treats even color remote sensing images and carries out cloud detection with reference to the initial overlapping region between remote sensing images, obtain cloud Masked areas；It removes and treats even color remote sensing images and with reference to the cloud masked areas in the initial overlapping region between remote sensing images, obtain To the overlapping region treated between even color remote sensing images and reference remote sensing images.

In some embodiments, even color remote sensing images are treated and with reference to the initial overlapping region between remote sensing images into racking Detection, obtains cloud masked areas and includes：Even color remote sensing images are treated and with reference between remote sensing images based on deep neural network Initial overlapping region carries out cloud detection, obtains cloud masked areas.

In some embodiments, according to treating the overlapping regions of even color remote sensing images and the overlapping region with reference to remote sensing images, It treats even color remote sensing images and carries out even color processing, obtain the remote sensing images after even color and include：According to the weight for treating even color remote sensing images Folded region and the overlapping region with reference to remote sensing images determine to treat the overlapping region of even color remote sensing images and the weight with reference to remote sensing images Mapping relations between folded region；According to the overlapping region for treating even color remote sensing images between the overlapping region with reference to remote sensing images Mapping relations, treat the overlapping regions of even color remote sensing images, to obtain treating even color by mapping to reference to the overlapping region of remote sensing images Overlapping region after remote sensing images mapping；It treats the overlapping region after even color remote sensing images mapping and carries out even color processing, obtain even Remote sensing images after color.

In some embodiments, according to treating the overlapping regions of even color remote sensing images and the overlapping region with reference to remote sensing images, It determines to treat that the mapping relations between the overlapping region of even color remote sensing images and the overlapping region of reference remote sensing images include：Treat even color The overlapping region of remote sensing images respectively includes at least one mutual corresponding mapping area with the overlapping region with reference to remote sensing images； Determine the overlapping region for treating even color remote sensing images and mutual corresponding each mapping area in the overlapping region with reference to remote sensing images Between mapping relations.

In some embodiments, according to treat the overlapping regions of even color remote sensing images and the overlapping region with reference to remote sensing images it Between mapping relations, treat the overlapping regions of even color remote sensing images, to obtain treating even by mapping to reference to the overlapping region of remote sensing images Overlapping region after the mapping of color remote sensing images includes：According to the overlapping region and the weight with reference to remote sensing images for treating even color remote sensing images Mapping relations in folded region between mutual corresponding each mapping area are treated being mapped to reference to the overlapping region of remote sensing images The overlapping region of even color remote sensing images obtains the overlapping region after even color remote sensing images map.

In some embodiments, it determines to treat in the overlapping region of even color remote sensing images and the overlapping region with reference to remote sensing images Mapping relations between mutual corresponding each mapping area include following any one：Based on mutual corresponding each map section The color difference of feature pair in domain determines the mapping relations between mutual corresponding each mapping area；Based on to mutually right The color of feature pair in each mapping area answered carries out machine learning, determines between mutual corresponding each mapping area Mapping relations；And based on the Histogram Matching between mutual corresponding each mapping area, mutual corresponding each reflect is determined Penetrate the mapping relations between region.

In some embodiments, it treats the overlapping region after even color remote sensing images mapping and carries out even color processing, obtain even color Remote sensing images afterwards include：At least one preliminary dimension is chosen along the boundary of the overlapping region after the mapping of even color remote sensing images Extended area；Each extended area is merged respectively, obtains the remote sensing images after even color.

In some embodiments, each extended area is merged respectively, obtains the figure after the even color of the extended area As including：For each extended area, the image after the even color of the extended area is obtained according to following any one：To extension Region is merged using graph cut algorithm, obtains the image after the even color of the extended area；Picture in extended area Distance of the element away from boundary carries out linear weighted function fusion to extended area, obtains the image after the even color of the extended area；And base Split is carried out in the minimum split path that extended area determines in dynamic programming algorithm, obtains the figure after the even color of the extended area Picture.

In some embodiments, extended area using graph cut algorithm is merged, obtains the even of the extended area Image after color includes：Calculate the energy function of extended area；Boundary based on extended area, establishes constraint equation；Based on energy Flow function and constraint equation using graph cut algorithm fusion extended area, obtain the image after even color.

In some embodiments, extended area includes the local window that more than one meets preliminary dimension, each local window Mouth includes at least one pixel.

In some embodiments, it based on energy function and constraint equation, using graph cut algorithm fusion extended area, obtains Image after to even color includes：Based on energy function and constraint equation, each window is merged using graph cut algorithm successively, is obtained Video in window to after even color；Splice the video in window after all even colors, obtain the image after even color.

In some embodiments, it based on energy function and constraint equation, using graph cut algorithm fusion extended area, obtains Image after to even color includes following any one：Based on gradient and constraint equation, using graph cut algorithm fusion expansion area Domain；Based on divergence and constraint equation, using graph cut algorithm fusion extended area；Based on angle point measurement and constraint equation, adopt With graph cut algorithm fusion extended area；And based on histogram of gradients and constraint equation, using graph cut algorithm fusion Extended area.

In some embodiments, according to treating the overlapping regions of even color remote sensing images and the overlapping region with reference to remote sensing images, It determines to treat that the mapping relations between the overlapping region of even color remote sensing images and the overlapping region of reference remote sensing images include：According to treating An at least channel for the overlapping region of even color remote sensing images and the overlapping region with reference to remote sensing images, determines to treat even color remote sensing images Overlapping region and with reference to remote sensing images overlapping region between mapping relations.

Second aspect, this application provides a kind of remote sensing image processing device, device includes：Feature uses matching unit The feature of even color remote sensing images and the feature of reference remote sensing images are treated in matching, obtain multiple features pair；Mapping relations determine list Member for being based on multiple features pair, determines Feature Mapping relationship between image；Overlapping region determination unit, for according between image Feature Mapping relationship determines to treat even color remote sensing images and with reference to the overlapping region between remote sensing images；Even color processing unit, is used for According to the overlapping region and the overlapping region with reference to remote sensing images for treating even color remote sensing images, treat even color remote sensing images and carry out even color Processing, obtains the remote sensing images after even color.

In some embodiments, overlapping region determination unit includes：Prime area determination subelement, for according to multiple spies Sign pair determines to treat even color remote sensing images and with reference to the initial overlapping region between remote sensing images；Masked areas detection sub-unit is used In treating even color remote sensing images and carrying out cloud detection with reference to the initial overlapping region between remote sensing images, cloud masked areas is obtained； Masked areas removes subelement, and even color remote sensing images are treated and with reference in the initial overlapping region between remote sensing images for removing Cloud masked areas obtains treating even color remote sensing images and with reference to the overlapping region between remote sensing images.

In some embodiments, masked areas detection sub-unit is further used for：Even color is treated based on deep neural network Initial overlapping region between remote sensing images and reference remote sensing images carries out cloud detection, obtains cloud masked areas.

In some embodiments, even color processing unit includes：Mapping relations determination subelement treats even color remote sensing for basis The overlapping region of image and the overlapping region with reference to remote sensing images determine to treat the overlapping region of even color remote sensing images with referring to remote sensing Mapping relations between the overlapping region of image；Overlapping region maps subelement, for according to the overlapping for treating even color remote sensing images Region and with reference to remote sensing images overlapping region between mapping relations, by with reference to remote sensing images overlapping region map to treat it is even The overlapping region of color remote sensing images obtains the overlapping region after even color remote sensing images map；Even color handles subelement, for pair Overlapping region after the mapping of even color remote sensing images carries out even color processing, obtains the remote sensing images after even color.

In some embodiments, mapping relations determination subelement is further used for：Treat the overlapping region of even color remote sensing images At least one mutual corresponding mapping area is respectively included with the overlapping region with reference to remote sensing images；It determines to treat even color remote sensing images Overlapping region and with reference to remote sensing images overlapping region in mapping relations between mutual corresponding each mapping area.

In some embodiments, mapping subelement in overlapping region is further used for：According to the overlapping for treating even color remote sensing images Mapping relations in region and the overlapping region of reference remote sensing images between mutual corresponding each mapping area, will refer to remote sensing The overlapping region of image maps to the overlapping region for treating even color remote sensing images, obtains the overlay region after even color remote sensing images map Domain.

In some embodiments, mapping relations determination subelement is further used for following any one：Based on corresponding Each mapping area in feature pair color difference, determine the mapping relations between mutual corresponding each mapping area； Based on machine learning is carried out to the color of the feature pair in mutual corresponding each mapping area, mutual corresponding each reflect is determined Penetrate the mapping relations between region；And it based on the Histogram Matching between mutual corresponding each mapping area, determines mutual Mapping relations between corresponding each mapping area.

In some embodiments, even color processing subelement is further used for：Along the overlapping after the mapping of even color remote sensing images Choose the extended area of at least one preliminary dimension in the boundary in region；Each extended area is merged respectively, obtains even color Remote sensing images afterwards.

In some embodiments, even color processing subelement is further used for：For each extended area, according to following Meaning one obtains the image after the even color of the extended area：Extended area using graph cut algorithm is merged, is somebody's turn to do Image after the even color of extended area；Distance of the pixel away from boundary in extended area carries out extended area linearly to add Power fusion, obtains the image after the even color of the extended area；And based on the minimum that dynamic programming algorithm is determined in extended area Split path carries out split, obtains the image after the even color of the extended area.

In some embodiments, even color processing subelement is further used for：Calculate the energy function of extended area；Based on expansion The boundary in exhibition section domain, establishes constraint equation；Based on energy function and constraint equation, using graph cut algorithm fusion expansion area Domain obtains the image after even color.

In some embodiments, the extended area in even color processing subelement includes more than one office for meeting preliminary dimension Portion's window, each local window include at least one pixel.

In some embodiments, even color processing subelement is further used for：Based on energy function and constraint equation, using pool Loose blending algorithm merges each window successively, obtains the video in window after even color；Splice the video in window after all even colors, obtain Image after even color.

In some embodiments, even color processing subelement is further used for following any one：Based on gradient and constraint side Journey, using graph cut algorithm fusion extended area；Based on divergence and constraint equation, using graph cut algorithm fusion expansion area Domain；Based on angle point measurement and constraint equation, using graph cut algorithm fusion extended area；And based on histogram of gradients peace treaty Shu Fangcheng, using graph cut algorithm fusion extended area.

In some embodiments, mapping relations determination subelement is further used for：According to the overlapping for treating even color remote sensing images An at least channel in region and the overlapping region with reference to remote sensing images determines to treat that the overlapping region of even color remote sensing images and reference are distant Feel the mapping relations between the overlapping region of image.

The third aspect, this application provides a kind of electronic equipment, including：Memory stores executable instruction；One or more A processor communicates with memory and completes following operate to perform executable instruction：The spy of even color remote sensing images is treated in matching Sign and the feature with reference to remote sensing images, obtain multiple features pair；Based on multiple features pair, Feature Mapping relationship between image is determined； According to Feature Mapping relationship between image, determine to treat even color remote sensing images and with reference to the overlapping region between remote sensing images；According to treating The overlapping region of even color remote sensing images and the overlapping region with reference to remote sensing images, treat even color remote sensing images and carry out even color processing, Obtain the remote sensing images after even color.

Remote sensing image processing method, device and the electronic equipment that the application provides, even color remote sensing figure is treated by matching first The feature of picture and the feature with reference to remote sensing images, obtain multiple features pair, later based on multiple features pair, determine feature between image Mapping relations later according to Feature Mapping relationship between image, determine to treat even color remote sensing images and with reference to the weight between remote sensing images Folded region, it is last according to the overlapping region for treating even color remote sensing images and the overlapping region with reference to remote sensing images, treat even color remote sensing Image carries out even color processing, obtains the remote sensing images after even color, so as to be eliminated in the processing procedure of remote sensing images largely not Necessary redundant data improves the treatment effeciency of remote sensing images and improves the processing accuracy of remote sensing images.

Further, in some embodiments, remote sensing image processing method, device and the electronic equipment that the application provides are also It provides and when even color image includes cloud masked areas, how to determine to treat between even color remote sensing images and reference remote sensing images Overlapping region, so as to avoid cloud or snow of the remote sensing images caused by being influenced by the highlighted cartographic feature such as cloud or snow Severe distortion occurs for regional color, so as to improve the accuracy rate of subsequent image processing.

Description of the drawings

By reading the detailed description made to non-limiting example made with reference to the following drawings, the application's is other Feature, objects and advantages will become more apparent upon：

Fig. 1 is the schematic flow chart according to one embodiment of the remote sensing image processing method of the application；

Fig. 2 is to treat even color remote sensing images and with reference to remote sensing images feature based point to obtaining overlapping region according to the application An exemplary application scene schematic diagram；

Fig. 3 is to determine to treat even color remote sensing images and with reference to overlapping between remote sensing images according to Feature Mapping relationship between image The schematic flow chart of one embodiment of the method in region；

Fig. 4 is to treat even color remote sensing images and refer to cloud detection of the remote sensing images based on deep learning to obtain according to the application The schematic diagram of one exemplary application scene of cloud mask image；

Fig. 5 is to treat even color remote sensing with the overlapping region with reference to remote sensing images according to the overlapping region for treating even color remote sensing images Image carries out even color processing to obtain the schematic flow chart of one embodiment of the method for the remote sensing images after even color；

Fig. 6 a are according to the accumulative histogram of the overlapping region of the reference remote sensing images of the application and treat even color remote sensing images Overlapping region accumulative histogram an exemplary application scene schematic diagram；

Fig. 6 b are that even color remote sensing figure is approximately treated in the overlapping region with referring to remote sensing images after the mapping according to the application The schematic diagram of the accumulative histogram of the overlapping region of picture；

Fig. 7 be according to the overlapping region of the application establish one of extended area for even color BORDER PROCESSING it is exemplary should With the schematic diagram of scene；

Fig. 8 a be using the application remote sensing image processing method before the one embodiment for treating even color remote sensing images Schematic diagram；

Fig. 8 b are the images after the even color of remote sensing image processing method to Fig. 8 a application the embodiment of the present application；

Fig. 9 a be using the application remote sensing image processing method before another embodiment for treating even color remote sensing images Schematic diagram；

Fig. 9 b are the images after the even color of remote sensing image processing method to Fig. 9 a application the embodiment of the present application；

Figure 10 is the structure diagram according to one embodiment of the remote sensing image processing device of the application；

Figure 11 is adapted for for realizing that the terminal device of the embodiment of the present application or the structure of the computer system of server show It is intended to.

Specific embodiment

The application is described in further detail with reference to the accompanying drawings and examples.It is understood that this place is retouched The specific embodiment stated is used only for explaining related invention rather than the restriction to the invention.It also should be noted that in order to Convenient for description, illustrated only in attached drawing and invent relevant part with related.

It should be noted that in the absence of conflict, the feature in embodiment and embodiment in the application can phase Mutually combination.The application is described in detail below with reference to the accompanying drawings and in conjunction with the embodiments.

Fig. 1 shows the schematic flow chart of one embodiment of the remote sensing image processing method according to the application.

As shown in Figure 1, remote sensing image processing method 100, includes the following steps：

Step 101, the feature and the feature with reference to remote sensing images that even color remote sensing images are treated in matching, obtain multiple features pair.

In the present embodiment, electronic equipment (such as the service shown in FIG. 1 of remote sensing image processing method operation thereon Device) it can extract respectively and treat the features of even color remote sensing images and the feature with reference to remote sensing images, later to two image zooming-outs Feature is matched, and obtains feature pair.Here reference remote sensing images, to be manually selected close with treating even color remote sensing images or It is similar and with the history remote sensing images of higher image quality (such as clarity high, without covering etc.).

Here the feature and the feature with reference to remote sensing images that even color remote sensing images are treated in matching can be that matching treats that even color is distant The feature and the characteristic point with reference to remote sensing images for feeling image, or matching treats the feature of even color remote sensing images and with reference to remote sensing The characteristic curve of image can also be feature and the other feature with reference to remote sensing images that even color remote sensing images are treated in matching.

Illustratively, below by taking characteristic point as an example, illustrate to treat the characteristic point of even color remote sensing images and reference remote sensing images The extraction and matching of characteristic point.

Characteristic point is also known as point of interest, key point, it is prominent in the picture and with some the apparent points for representing meaning. Extraction characteristic point is those pixels for being easiest to identification to be found in two images to be matched, such as angle point, by image The artis of some geometries is formed, and is all much the intersection point generated between lines, in another example the object edge of texture-rich Point.The algorithm of characteristic point is extracted, can be can have scale anti-to a certain degree, color in the prior art or the technology of future development Color, affine variation, has the algorithm of robust features point extractability, and the application does not limit this.For example, ruler can be passed through Degree invariant features convert (SIFT) algorithm to extract characteristic point.

Extraction characteristic point can usually include following two steps：Extraction detection：It is sought in two images to be matched The pixel (angle point) that those is looked for be easiest to identification, such as object edge point of texture-rich etc.；Extraction description, can use one Detection is described in feature mathematically a bit, and such as histogram of gradients, local random binary feature etc. obtains retouching for characteristic point State son.

After characteristic point is extracted, the characteristic point for treating even color remote sensing images and the feature with reference to remote sensing images can be matched Point.It is possible, firstly, to judge their correspondences in two images, judge that algorithm used can be the prior art or future The algorithm of correspondence of the judging characteristic o'clock in two images, the application do not limit this in the technology of development.It for example, can To judge correspondence of description in two images using the quick nearest neighbor algorithm (FLANN) of high dimensional data.In order to go Except the match point of mistake, retain correct match point, the de-noising in the prior art or the technology of future development can also be used to calculate Method carries out de-noising to obtained description, and the application does not limit this.For example, random sampling consistency may be used (RANSAC) algorithm carries out de-noising to obtained description.

Step 102, based on multiple features pair, Feature Mapping relationship between image is determined.

In the present embodiment, on the basis of the characteristic point pair obtained in step 101, Feature Mapping is closed between can obtaining image System.Here Feature Mapping relationship, it is mutually " right between the feature for treating even color remote sensing images and the feature of reference remote sensing images to refer to Should " relationship, can be there are many form of expression, the application not limit this.It for example, can be with this performance shape of mapping matrix Formula is come the Feature Mapping relationship that shows.

Below by taking characteristic point pair as an example, illustrate to determine mapping matrix between image when Feature Mapping relationship is mapping matrix Process：

Assuming that (x₁,y₁) and (x₂,y₁) it is that corresponding image points (represents practically between the two images extracted by SIFT algorithms A pair of of pixel of same position in object, referred to as corresponding image points) in a pair, following mapping is met between that corresponding image points and is closed System：

Wherein, A is affine transformation matrix, wherein parameter a₁₁...a₃₃It has co-expressed flat in the relativeness between image Therefore the design parameters such as shifting, scaling, rotation, shearing, can utilize corresponding image points to solve the parameter matrix A in above formula, Obtain mapping matrix between image.Here the calculation of the determining parameter matrix in the prior art or the technology of future development may be used Method determines parameter matrix, and the application do not limit this.For example, RANSAC algorithms may be used to determine parameter matrix.

Step 103, it according to Feature Mapping relationship between image, determines to treat even color remote sensing images and with reference between remote sensing images Overlapping region.

In the present embodiment, it based on Feature Mapping relationship between the image obtained in step 102, calculates with reference to remote sensing images The transformed conversion point set in vertex, the vertex set by treating even color remote sensing images and the conversion point set with reference to remote sensing images, calculate Intersection of polygon.According between image Feature Mapping relationship it is inverse, it is original in reference to remote sensing images to calculate the intersection of polygon Point set obtains overlapping region.

Illustratively, overlapping region between image for mapping matrix, is calculated between determining image by feature based point：

It please refers to Fig.2, in fig. 2, based on mapping matrix between image, the overlapping region of A images can be calculated：A images Overlapping region is polygon, and the vertex for setting polygon is respectively point P₁、P₂、P₃、P₄, point P₂For the bottom right point coordinates of A images, point P₁For corresponding image points coordinate of the B images upper left point coordinates on A images, the mapping relations obtained by step 103 in above-mentioned Fig. 1, It can obtain P₁Actual coordinate point in image A is calculating point P₃Coordinate when, can by the lower-left point coordinates of B images, count Calculate its corresponding image points P on A images₃' coordinate, line P₁P₃', P₁P₃' the intersection point with A image lower boundaries is point P₃Coordinate；Together Reason is calculating point P₄Coordinate when, corresponding image points P of the upper right point coordinates of B images on A images can be calculated₄' coordinate, even Line P₁P₄', P₁P₄' the intersection point with A image right margins is point P₄Coordinate.Point P is calculated₁、P₂、P₃、P₄Coordinate in image A, Also the overlapping region of A images is just obtained, according to the mapping relations that step 103 in Fig. 1 obtains, the overlay region of image B can be obtained Domain then obtains the boundary of overlapping region：P₁P₃、P₃P₂、P₂P₄、P₄P₁。

Fig. 1 is returned, step 104, according to treating the overlapping regions of even color remote sensing images and the overlapping region with reference to remote sensing images, It treats even color remote sensing images and carries out even color processing, obtain the remote sensing images after even color.

In the present embodiment, can the overlay region of even color remote sensing images be treated according to the overlapping region with reference to remote sensing images Domain is handled, and is obtained that treated and is treated the overlapping region of even color remote sensing images, can treat even color remote sensing to treated later The boundary of the overlapping region of image carries out even color processing, so as to obtain the remote sensing images after even color.

As mentioned in the background art, traditional remote sensing image processing generally includes following steps：Basis treats even color first Remote sensing images select the reference remote sensing images close or similar to its, treat even color remote sensing images later and with reference to remote sensing images It is matched, pending remote sensing images is spliced and with reference to remote sensing using emergence processing in splicing regions according to matching result later Image.Wherein, when carrying out images match, mainly using geographical coordinate registration and feature registration two schemes.

However, when carrying out images match in the even color processing of traditional remotely-sensed data image, geographical coordinate registration usually by Registration accuracy is caused to be restricted, and for image DecryptDecryption in geographical coordinate error and image itself distortion, many times Image does not have geographical coordinate.Characteristic matching scheme becomes more general scheme thus, although main stream approach is real now Existing Automatic signature extraction, but still can there are a large amount of erroneous matchings in matching.And pending remote sensing is spliced according to matching result later When image is with reference to remote sensing images, is usually handled in the range of full figure, include mass of redundancy data in the data of processing, Lead to that the treatment effeciency of remote sensing images is relatively low and processing accuracy is poor, and handled in splicing regions using emergence, can be caused thin It saves smudgy.

In contrast, the remote sensing image processing method that the above embodiments of the present application provide, even color remote sensing figure is treated by matching The feature of picture and the feature with reference to remote sensing images, obtain multiple features pair；Based on multiple features pair, Feature Mapping between image is determined Relationship；According to Feature Mapping relationship between image, determine to treat even color remote sensing images and with reference to the overlapping region between remote sensing images；Root According to the overlapping region and the overlapping region with reference to remote sensing images for treating even color remote sensing images, treat even color remote sensing images and carry out at even color Reason, obtains the remote sensing images after even color, so as to eliminate a large amount of unnecessary redundant datas in the processing procedure of remote sensing images, It improves the treatment effeciency of remote sensing images and improves the processing accuracy of remote sensing images.

With further reference to Fig. 3, Fig. 3, which is shown, to be determined to treat even color remote sensing images and reference according to Feature Mapping relationship between image The schematic flow chart of one embodiment of the method for the overlapping region between remote sensing images.

This determines to treat even color remote sensing images and with reference to the overlay region between remote sensing images according to Feature Mapping relationship between image The method in domain shows when even color image includes cloud masked areas, how to determine to treat that even color remote sensing images and reference are distant Feel the overlapping region between image.

As shown in figure 3, it is determined to treat even color remote sensing images and with reference between remote sensing images according to Feature Mapping relationship between image The method 300 of overlapping region include：

In step 301, it according to multiple features pair, determines to treat even color remote sensing images and with reference to initial between remote sensing images Overlapping region.

In the present embodiment, according to multiple features pair, it may be determined that treat even color remote sensing images and with reference between remote sensing images Image between Feature Mapping relationship, later according to mapping relations between image, calculate transformed turn of the vertex with reference to remote sensing images Point set is changed, the vertex set by treating even color remote sensing images and the conversion point set with reference to remote sensing images calculate intersection of polygon.According to Feature Mapping relationship is inverse between image, original point set of the intersection of polygon in reference to remote sensing images is calculated, so that it is determined that treating Overlapping region between even color remote sensing images, and using the determining overlapping region as initial overlapping region.

In step 302, even color remote sensing images are treated and are examined with reference to the initial overlapping region between remote sensing images into racking It surveys, obtains cloud masked areas.

In the present embodiment, Color Statistical is carried out to the overlapping region of even color image during even color, due to cloud on the image Usually highlighted white area, it is impossible to which the realistic colour of true reflection atural object simultaneously brings very big error, more manages in order to obtain The even color effect thought needs to know the range of cloud and ignores the region.It should be appreciated that cloud detection here can detect cloud, The shape of shade caused by snow and other shelters, for the convenience of description, unifying to carry out exemplary description with cloud detection here Deng.Here cloud mask, refers to the cloud for covering image or object.

The method for carrying out cloud detection in the application to overlapping region, may be used in the prior art or the technology of future development For the cloud detection method of optic of remote sensing images, the application does not limit this.It is applied to for example, multispectral physical features may be used The physical method that is detected in single pixel, the detection method of texture based on cloud and spatial character, pattern-recognition detection Optimizing detection method that method and more algorithm synthesis use etc. obtains the cloud masked areas of overlapping region.

Fig. 4 shows the cloud detection effect based on deep learning, in Fig. 4, using deep learning algorithm to image A and figure As the overlapping region P of B₁、P₂、P₃、P₄Cloud detection is carried out, the cloud mask image of image A and image B, image A and figure can be obtained As B cloud mask image in respectively illustrate cloud masked areas 410,420 and 430.Here cloud mask image refers to：It identifies The image of the position of cloud, usually a width binary map have cloud sector domain in the conversion of subsequent histogram in cloud mask image It is ignored.

Fig. 3 is returned, in some optional realization methods of the present embodiment, treats even color remote sensing images with referring to remote sensing images Between initial overlapping region carry out cloud detection, obtain cloud masked areas and include：Even color remote sensing is treated based on deep neural network Initial overlapping region between image and reference remote sensing images carries out cloud detection, obtains cloud masked areas.

In this realization method, treated based on deep neural network between even color remote sensing images and reference remote sensing images Initial overlapping region carries out cloud detection, when obtaining cloud masked areas, since neural network has very strong nonlinear fitting ability, and And learning rules are simple, realized convenient for computer and with very strong robustness, memory capability, non-linear mapping capability and strong Big self-learning capability, thus when can improve cloud detection caused by cloud, snow and other shelters shade accuracy of identification, And the accuracy of recognition result can be improved.

In step 303, it removes and treats even color remote sensing images and with reference to the cloud in the initial overlapping region between remote sensing images Masked areas obtains treating even color remote sensing images and with reference to the overlapping region between remote sensing images.

In the present embodiment, after the cloud masked areas for obtaining overlapping region in step 302, can obtain respectively treat it is even The overlapping regions of color remote sensing images and with reference to remote sensing images this two parts overlapping region of overlapping region in except cloud masked areas it Outer region, and using the region of acquisition as overlapping region.Overlapping region at this time has eliminated the highlighted atural object such as cloud or snow The influence of image so as to improve the accuracy rate of subsequent processing, and reduces data redundancy, improves the effect of subsequent processing Rate.

What the above embodiments of the present application provided determines to treat even color remote sensing images and reference according to Feature Mapping relationship between image The method of overlapping region between remote sensing images, can be to avoid remote sensing images due to being influenced by the highlighted cartographic feature such as cloud or snow Severe distortion occurs for caused cloud or snow regional color, so as to improve the accuracy rate of subsequent image processing.

With further reference to Fig. 5, Fig. 5 is shown according to the overlapping region and the weight with reference to remote sensing images for treating even color remote sensing images It treats even color remote sensing images and carries out even color processing to obtain one embodiment of the method for the remote sensing images after even color in folded region Schematic flow chart.

As shown in figure 5, according to treat even color remote sensing images overlapping region and with reference to remote sensing images overlapping region treat it is even Color remote sensing images carry out even color processing and are included with obtaining the method schematic flow chart 500 of the remote sensing images after even color：

In step 501, it according to the overlapping region and the overlapping region with reference to remote sensing images for treating even color remote sensing images, determines Treat the mapping relations between the overlapping region of even color remote sensing images and the overlapping region of reference remote sensing images.

It in the present embodiment, can be according to the overlapping region and the overlapping region with reference to remote sensing images for treating even color remote sensing images In the mutual corresponding subregion of at least part or pixel, to determine to treat the overlapping region of even color remote sensing images with referring to remote sensing figure Mapping relations between the overlapping region of picture.

In some optional realization methods of the present embodiment, according to the overlapping region for treating even color remote sensing images and with reference to remote sensing The overlapping region of image determines to treat the mapping between the overlapping region of even color remote sensing images and the overlapping region of reference remote sensing images Relationship includes：The overlapping region for treating even color remote sensing images and the overlapping region for referring to remote sensing images respectively include at least one mutual Corresponding mapping area；Determine to treat that the overlapping region of even color remote sensing images is corresponded with the overlapping region with reference to remote sensing images Each mapping area between mapping relations.

In this realization method, it is contemplated that remote sensing images size is larger, therefore according to the overlapping for treating even color remote sensing images Region and the overlapping region with reference to remote sensing images determine to treat that the overlapping region of even color remote sensing images is overlapping with reference remote sensing images During mapping relations between region, overlapping region can be divided into more than one mapping area, later according to more than one Mapping area, determine to treat even color image and the mapping relations with reference to remote sensing images, it is mutual corresponding each so as to be based on The more detail content of mapping area determines mapping relations, improves the precision of mapping relations, and due to single treatment Data are less, can improve the treatment effeciency of remote sensing images.Here mapping area refers in overlapping region for the son of mapping Region.

In some optional realization methods of the present embodiment, according to the overlapping region for treating even color remote sensing images and with reference to remote sensing The overlapping region of image determines to treat the mapping between the overlapping region of even color remote sensing images and the overlapping region of reference remote sensing images Relationship includes：According at least channel for treating the overlapping regions of even color remote sensing images and the overlapping region with reference to remote sensing images, really Surely the mapping relations between the overlapping region of even color remote sensing images and the overlapping region of reference remote sensing images are treated.

In this realization method, due to treating even color remote sensing images with having multiple channels with reference to remote sensing images, it is possible to It is matched by channel, obtains treating even color remote sensing images and the mapping relations with reference to the overlapping region between remote sensing images.For example, Can based on pixel value each channel accumulative histogram, determine to treat the overlapping regions of even color remote sensing images with reference to remote sensing images Overlapping region between mapping relations.Here the channel of remote sensing images refers to that the sensor selection for acquiring remote sensing images connects The ability of electromagnetic wave is received, actually refers to the service band of sensor.Each service band of sensor, can be referred to as For a channel.As soon as sensor can receive several electromagnetic wave bands, referred to as several channel sensors.An if for example, job Wave band is 10 to 20nm, then just more than 50 to 100 a spectrum channels of needs that spectral regions are 0.4 to 2.5 microns.

In step 502, between the overlapping region according to the overlapping region for treating even color remote sensing images and with reference to remote sensing images Mapping relations, treat the overlapping regions of even color remote sensing images, to obtain treating even color by mapping to reference to the overlapping region of remote sensing images Overlapping region after remote sensing images mapping.

In the present embodiment, between the overlapping region according to the overlapping region for treating even color remote sensing images and with reference to remote sensing images Mapping relations, by with reference to remote sensing images overlapping region at least part subregion or pixel, map to and treat even color remote sensing Corresponding at least part subregion or pixel in the overlapping region of image, so as to obtain the weight after even color remote sensing images map Folded region.

In some optional realization methods of the present embodiment, according to the overlapping region for treating even color remote sensing images and with reference to remote sensing The overlapping region of reference remote sensing images is mapped to the weight for treating even color remote sensing images by the mapping relations between the overlapping region of image Folded region obtains the overlapping region after even color remote sensing images map and includes：According to treat the overlapping regions of even color remote sensing images with Mapping relations in overlapping region with reference to remote sensing images between mutual corresponding each mapping area, will be with reference to remote sensing images Overlapping region maps to the overlapping region for treating even color remote sensing images, obtains the overlapping region after even color remote sensing images map.

In this realization method, by using the mapping relations between each mapping area, by the overlapping with reference to remote sensing images Area maps obtain the overlapping region after even color remote sensing images map, single needs to the overlapping region for treating even color remote sensing images Data volume to be processed is less, therefore improves data-handling efficiency, and according to the mapping relations between each mapping area It will be mapped to when the overlapping region of even color remote sensing images with reference to the overlapping region of remote sensing images, more map sections can be included Local detail in domain, therefore the precision of images of the overlapping region after mapping can be improved.

In some optional realization methods of the present embodiment, determine to treat the overlapping region of even color remote sensing images with referring to remote sensing Mapping relations in the overlapping region of image between mutual corresponding each mapping area can include：It can be based on corresponding Each mapping area in feature pair color difference, determine the mapping relations between mutual corresponding each mapping area； It is alternatively possible to carry out machine learning based on the color to the feature pair in mutual corresponding each mapping area, determine mutual Mapping relations between corresponding each mapping area；It is alternatively possible to based between mutual corresponding each mapping area Histogram Matching determines the mapping relations between mutual corresponding each mapping area.

In this realization method, by the color difference of the feature pair in mutual corresponding each mapping area, phase is determined Mapping relations between mutual corresponding each mapping area, can conveniently determine mapping relations, improve mapping relations Determine speed；By carrying out machine learning based on the color to the feature pair in mutual corresponding each mapping area, phase is determined Mapping relations between mutual corresponding each mapping area, accuracy is high, learning ability is strong and the robustness to noise data and Fault-tolerance is stronger；By based on the Histogram Matching between mutual corresponding each mapping area, determining mutual corresponding each Mapping relations between mapping area, can improve the accuracy of mapping result, and retain image detail.

Below with reference to Fig. 6 a and Fig. 6 b, illustrate how in the overlapping region for treating even color remote sensing images and reference remote sensing images The mapping of accumulative histogram is carried out between overlapping region.

Histogram Mapping relationship is being used by the histogram aggregate-value of each pixel in the overlapping region with reference to remote sensing images It maps to when each pixel in the overlapping region of even color remote sensing images, because remote sensing images have multiple channels, therefore can be into Row is by the Histogram Matching of channel.

First, as shown in Figure 6 a, the overlapping region of statistical-reference remote sensing images and even color remote sensing images can be treated respectively The pixel value of each corresponding channel between overlapping region, obtain reference remote sensing images as shown in Figure 6 a overlapping region it is accumulative straight Square Figure 60 1 and the accumulative histogram 602 of overlapping region for treating even color remote sensing images, later, by the accumulative Nogata obtained in Fig. 6 a Figure 60 1,602 establishes the mapping of pixel value between the overlapping region for treating even color remote sensing images and the overlapping region of reference remote sensing images Relationship makes to treat the overlapping region of even color remote sensing images with the same pixel value in the overlapping region with reference to remote sensing images with identical Histogram aggregate-value, so as to treat the pixel value of the overlapping region of even color remote sensing images carry out mapping transformation, make to treat that even color is distant The overlapping region of sense image obtains the approximate accumulative histogram 603 in overlapping region with reference remote sensing images as shown in Figure 6 b.

Here histogram aggregate-value refers to accumulated probability distribution situation of the constituent in gray level of representative image, Each probability value represents the probability less than or equal to this gray value.

Fig. 5 is returned, in step 503, the overlapping region after even color remote sensing images mapping is treated and carries out even color processing, obtain Remote sensing images after even color.

In the present embodiment, since the overlapping region after the mapping of even color remote sensing images is to be mapped from reference to remote sensing images It is coming as a result, compared with other regions in addition to overlapping region of the even color remote sensing images after mapping, in the side such as color, brightness Face has differences, it is therefore desirable to treat the overlapping region after even color remote sensing images mapping and carry out even color processing, to obtain even color Remote sensing images afterwards.

In some optional realization methods of the present embodiment, the overlapping region progress treated after even color remote sensing images mapping is even Color processing, obtains the remote sensing images after even color and includes：Along after even color remote sensing images mapping after overlapping region boundary choose to The extended area of few preliminary dimension；Each extended area is merged respectively, obtains the remote sensing images after even color.

In this realization method, preliminary dimension needs to determine the size of extended area and phase according to the size of remote sensing images The range of fusion is hoped to determine.Here extended area can inwardly and/or outwardly extend, this Shen along the boundary of overlapping region Please this is not limited.

In some optional realization methods of the present embodiment, each extended area is merged respectively, obtains the extension Image after the even color in region can include：For each extended area, which is obtained according to following any one Even color after image：Extended area using graph cut algorithm (possion blending) is merged, obtains the expansion Image after the even color in exhibition section domain；It is alternatively possible to distance of the pixel away from boundary in extended area, to extended area into Line Weighted Fusion obtains the image after the even color of the extended area；It is alternatively possible to it is being extended based on dynamic programming algorithm The minimum split path that region determines carries out split, obtains the image after the even color of the extended area.

In this realization method, by using graph cut algorithm according to the energy function and extended area of extended area Boundary information, recreate the image pixel in extended area using the method for interpolation to obtain the image after even color, select It selects the process of integration region simply and conveniently, can realize seamless even color processing so that the color of image mosaic borderline region Even transition.

By distance of the pixel in extended area away from boundary, linear weighted function fusion is carried out to extended area, is obtained Image pixel in extended area, and then the image after even color is obtained, it can make the transition effect of image co-registration more naturally, simultaneously And fusion efficiencies are improved, the real-time demand of system when meeting image procossing, and obtain and more managed than traditional algorithm The image syncretizing effect thought.

By carrying out split in the minimum split path that extended area determines according to dynamic programming algorithm, be expanded region Interior image pixel, and then the image after even color is obtained, it needs to find an optimal gap, this gap in extended area Pixel above be it is most unessential, in other words this gap undergo pixel energy accumulation and be it is minimum (can basis Energy function determines that energy function here can be arbitrary in the measurement of gradient, divergence, angle point and histogram of gradients of image One), this gap can be it is vertical extended through down from above, and width be a pixel, the right side can also be extended through from a left side, It is highly a pixel, optimal gap here is also minimum split path.By dynamic programming algorithm in extended area Determining minimum split path carries out split, thus it is possible to vary (reservation is important, and removal is secondary for the image resolution ratio in extended area , importance and secondary property depend on energy function), it can also emphasize picture material (content amplification), also Specific object can be deleted, so as to simply and effectively realize seamless even color processing so that image mosaic borderline region Uniform in color transition.

In some optional realization methods of the present embodiment, extended area using graph cut algorithm is merged, is obtained Image after to the even color of the extended area includes：Calculate the energy function of extended area；Boundary based on extended area is established Constraint equation；Based on energy function and constraint equation, using graph cut algorithm fusion extended area, the figure after even color is obtained Picture.

In this realization method, when calculating the energy function of extended area, energy function can be gradient, divergence, angle point Measurement and histogram of gradients in any one, after computation energy function, can the boundary based on extended area, establish about Shu Fangcheng is finally based on energy function and constraint equation, using graph cut algorithm fusion extended area, obtains the figure after even color Picture.So as to add image boundary constraints on the basis of the energy function, the even color meter of splicing boundary range is realized Calculate, while image detail information is kept as far as possible, realize color at splicing boundary it is uniform excessively.

In some optional realization methods of the present embodiment, extended area includes more than one part for meeting preliminary dimension Window, each local window include at least one pixel.

In this realization method, since remote sensing images size is larger, lead to that needs during even color solve with by processing image The relevant data volume of size is larger, can be based on meeting the window of preliminary dimension to extension in order to solve the problems, such as low memory Region carries out piecemeal processing, so as to improve data-handling efficiency.

In some optional realization methods of the present embodiment, based on energy function and constraint equation, calculated using graph cut Method merges extended area, obtains the image after even color and includes：Based on energy function and constraint equation, using graph cut algorithm according to The secondary each window of fusion, obtains the video in window after even color；Splice the video in window after all even colors, obtain the figure after even color Picture.

In this realization method, when often merging extended area using graph cut algorithm, by handling expansion area successively Each window in domain, the last entire buffer area of traversal processing, it is possible to reduce single data volume to be treated solves even color When need solve with by the relevant data volume of processing picture size it is larger caused by low memory the problem of, realize seamless Even color processing.

In some optional realization methods of the present embodiment, based on energy function and constraint equation, calculated using graph cut Method merges extended area, obtains the image after even color and includes following any one：Based on gradient and constraint equation, melted using Poisson Hop algorithm merges extended area；Based on divergence and constraint equation, using graph cut algorithm fusion extended area；Based on angle point amount Degree and constraint equation, using graph cut algorithm fusion extended area；And based on histogram of gradients and constraint equation, using pool Loose blending algorithm merges extended area.

In this realization method, gradient refers to the gradient of scalar field, and in vector calculus, the gradient of scalar field is one A vector field.Gradient in scalar field on certain point is directed toward the fastest-rising direction of scalar field, and the length of gradient is the corresponding increasing The maximum change rate in long most fast direction.By being based on gradient and constraint equation, using graph cut algorithm fusion expansion area Domain selects the process of integration region simply and conveniently, can realize seamless even color processing so that image mosaic borderline region Uniform in color transition.

Divergence (divergence), available for the degree of strength of characterization spatial points vector field diverging, physically, divergence Meaning is the active property of field.As div F>0, represent that point F has the positive source (emitting source) for distributing flux；As div F<0 represents to be somebody's turn to do Point F has the negative source (hole or remittance) of absorbed flux；As div F=0, represent that point F is passive.Based on divergence and constraint equation, use Graph cut algorithm fusion extended area selects the process of integration region simply and conveniently, can realize seamless even color processing, So that the uniform in color transition of image mosaic borderline region.

Angle point is measured, and is the data for weighing angle point, angle point is the violent point of two dimensional image brightness change or image side The point of curvature maximum on edge curve.It, can using graph cut algorithm fusion extended area based on angle point measurement and constraint equation While image graphics important feature is retained, to efficiently reduce the data volume of information, the content for making its information is very high, effectively Ground improves the speed of calculating, is conducive to the reliable matching of image.

Histogram of gradients, expression be edge (gradient) structure feature, abundant feature set can be formed, can be described The shape information of part.Based on histogram of gradients and constraint equation, using graph cut algorithm fusion extended area, can be based on The quantization in position and direction space, inhibits translation and the influence that brings of rotation, and in regional area normalization histogram, can be with The influence that partial offset illumination variation is brought.

The overlapping region and the weight with reference to remote sensing images that the basis that the above embodiments of the present application provide treats even color remote sensing images Folded region treats even color remote sensing images and carries out even color processing to obtain the method for the remote sensing images after even color, and right in traditional technology The emergence processing of splicing regions is compared, and realizes the smooth excessiveness of color, and remain the contrast or details of data in itself.

Illustratively, as shown in fig. 7, the boundary based on above-mentioned overlapping region can be established for the expansion of even color BORDER PROCESSING Exhibition section domain, extended area here is the inwardly extended obtained buffer area 710 in boundary along overlapping region.Arrow in such as 710 Shown in 711, regional area in buffering area per treatment, seamless even color processing is realized in the last entire buffer area of traversal processing.

Illustratively, Fig. 8 a are shown treats even color remote sensing images using before the embodiment of the present application, treats even color remote sensing figure As 8a includes highlighted cloud 810, Fig. 8 b show even to the remote sensing image processing method of Fig. 8 a application the embodiment of the present application Image after color, it can be seen that the image in Fig. 8 b after even color is not distorted, and the figure after even color in cloud covered areas domain color As remaining detailed information.

Illustratively, Fig. 9 a are shown treats even color remote sensing images using before the embodiment of the present application, treats even color remote sensing figure Shade 910 as caused by 9a includes shelter, Fig. 9 b show the remote sensing image processing to Fig. 9 a application the embodiment of the present application Image after the even color of method, it can be seen that image shadow region color caused by shelter in Fig. 9 b after even color does not occur Distortion, and the image after even color remains detailed information.

With further reference to Figure 10, as the realization to method shown in above-mentioned each figure, this application provides a kind of remote sensing images Processing unit one embodiment, the device embodiment is corresponding with embodiment of the method shown in FIG. 1, which can specifically apply In various electronic equipments.

As shown in Figure 10, the remote sensing image processing device 1000 of the present embodiment can include：

Feature is to matching unit 1010, the feature and the feature with reference to remote sensing images for treating even color remote sensing images for matching, Obtain multiple features pair.

Mapping relations determination unit 1020 for being based on multiple features pair, determines Feature Mapping relationship between image.

Overlapping region determination unit 1030, for according to Feature Mapping relationship between image, determine to treat even color remote sensing images with With reference to the overlapping region between remote sensing images.

Even color processing unit 1040, for overlapping with reference remote sensing images according to the overlapping region for treating even color remote sensing images Region treats even color remote sensing images and carries out even color processing, obtains the remote sensing images after even color.

The (not shown) in some optional realization methods of the present embodiment, overlapping region determination unit include：Initially Region determination subelement, for according to multiple features pair, determining to treat even color remote sensing images and with reference to initial between remote sensing images Overlapping region；Masked areas detection sub-unit, it is initial overlapping between even color remote sensing images and reference remote sensing images for treating Region carries out cloud detection, obtains cloud masked areas；Masked areas removes subelement, and even color remote sensing images and reference are treated for removing The cloud masked areas in initial overlapping region between remote sensing images obtains treating even color remote sensing images and with reference between remote sensing images Overlapping region.

The (not shown) in some optional realization methods of the present embodiment, masked areas detection sub-unit are further used In：Even color remote sensing images are treated based on deep neural network and are examined with reference to the initial overlapping region between remote sensing images into racking It surveys, obtains cloud masked areas.

The (not shown) in some optional realization methods of the present embodiment, even color processing unit include：Mapping relations Determination subelement, for according to the overlapping region and the overlapping region with reference to remote sensing images for treating even color remote sensing images, determining to treat even Mapping relations between the overlapping region of color remote sensing images and the overlapping region of reference remote sensing images；Overlapping region mapping is single Member treats the mapping relations between the overlapping region of even color remote sensing images and the overlapping region of reference remote sensing images for basis, will It maps to reference to the overlapping region of remote sensing images and treats the overlapping regions of even color remote sensing images, obtain after even color remote sensing images map Overlapping region；Even color handles subelement, carries out even color processing for treating the overlapping region after even color remote sensing images mapping, obtains Remote sensing images to after even color.

The (not shown) in some optional realization methods of the present embodiment, mapping relations determination subelement are further used In：The overlapping region for treating even color remote sensing images and the overlapping region for referring to remote sensing images respectively include at least one mutual corresponding Mapping area；Determine to treat the overlapping regions of even color remote sensing images with it is mutual corresponding each in the overlapping region with reference to remote sensing images Mapping relations between mapping area.

The (not shown) in some optional realization methods of the present embodiment, overlapping region mapping subelement are further used In：According to the overlapping region and mutual corresponding each map section in the overlapping region with reference to remote sensing images for treating even color remote sensing images The overlapping region of reference remote sensing images is mapped to the overlapping region for treating even color remote sensing images, obtained by the mapping relations between domain Overlapping region after the mapping of even color remote sensing images.

The (not shown) in some optional realization methods of the present embodiment, mapping relations determination subelement are further used In following any one：Based on the color difference of the feature pair in mutual corresponding each mapping area, determine mutual corresponding Mapping relations between each mapping area；Machine is carried out based on the color to the feature pair in mutual corresponding each mapping area Device learns, and determines the mapping relations between mutual corresponding each mapping area；And based on mutual corresponding each map section Histogram Matching between domain determines the mapping relations between mutual corresponding each mapping area.

The (not shown) in some optional realization methods of the present embodiment, even color processing subelement are further used for： The extended area of at least one preliminary dimension is chosen along the boundary of the overlapping region after the mapping of even color remote sensing images；Respectively to each A extended area is merged, and obtains the remote sensing images after even color.

The (not shown) in some optional realization methods of the present embodiment, even color processing subelement are further used for： For each extended area, the image after the even color of the extended area is obtained according to following any one：Extended area is adopted It is merged with graph cut algorithm, obtains the image after the even color of the extended area；Pixel in extended area is away from side The distance on boundary carries out linear weighted function fusion to extended area, obtains the image after the even color of the extended area；And based on dynamic Planning algorithm carries out split in the minimum split path that extended area determines, obtains the image after the even color of the extended area.

The (not shown) in some optional realization methods of the present embodiment, even color processing subelement are further used for： Calculate the energy function of extended area；Boundary based on extended area, establishes constraint equation；Based on energy function and constraint side Journey using graph cut algorithm fusion extended area, obtains the image after even color.

The (not shown) in some optional realization methods of the present embodiment, even color handle the extended area in subelement Meet the local window of preliminary dimension including more than one, each local window includes at least one pixel.

The (not shown) in some optional realization methods of the present embodiment, even color processing subelement are further used for： Based on energy function and constraint equation, each window is merged using graph cut algorithm successively, obtains the video in window after even color； Splice the video in window after all even colors, obtain the image after even color.

The (not shown) in some optional realization methods of the present embodiment, even color processing subelement be further used for Lower any one：Based on gradient and constraint equation, using graph cut algorithm fusion extended area；Based on divergence and constraint side Journey, using graph cut algorithm fusion extended area；Based on angle point measurement and constraint equation, expanded using graph cut algorithm fusion Exhibition section domain；And based on histogram of gradients and constraint equation, using graph cut algorithm fusion extended area.

The (not shown) in some optional realization methods of the present embodiment, mapping relations determination subelement are further used In：According to an at least channel for the overlapping region and the overlapping region with reference to remote sensing images for treating even color remote sensing images, determine to treat even Mapping relations between the overlapping region of color remote sensing images and the overlapping region of reference remote sensing images.

It should be appreciated that all units described in device 1000 are corresponding with each step in the method described with reference to figure 2. As a result, device 1000 and list wherein included are equally applicable to above with respect to the operation and feature of remote sensing image processing method description Member, details are not described herein.Corresponding units in device 1000 can cooperate with the unit in terminal device and/or server To realize the scheme of the embodiment of the present application.

It will be understood by those skilled in the art that above-mentioned remote sensing image processing device 1000 further includes some other known knots Structure, such as processor, memory etc., in order to unnecessarily obscure embodiment of the disclosure, these well known structures are in Fig. 10 It is not shown.

Present invention also provides a kind of electronic equipment, including：Memory stores executable instruction；One or more processing Device communicates with memory and completes following operate to perform executable instruction：The feature and ginseng of even color remote sensing images are treated in matching The feature of remote sensing images is examined, obtains multiple features pair；Based on multiple features pair, Feature Mapping relationship between image is determined；According to figure The Feature Mapping relationship as between determines to treat even color remote sensing images and with reference to the overlapping region between remote sensing images；According to treating that even color is distant The overlapping region and the overlapping region with reference to remote sensing images for feeling image, treat even color remote sensing images and carry out even color processing, obtain even Remote sensing images after color.

The embodiment of the present invention additionally provides a kind of electronic equipment, such as can be mobile terminal, personal computer (PC), put down Plate computer, server etc..Below with reference to Figure 11, it illustrates suitable for being used for realizing the terminal device of the embodiment of the present application or service The structure diagram of the electronic equipment 1100 of device：As shown in figure 11, computer system 1100 includes one or more processors, leads to Letter portion etc., one or more of processors are for example：One or more central processing unit (CPU) 1101 and/or one or more A image processor (GPU) 1113 etc., processor can be according to the executable instructions being stored in read-only memory (ROM) 1102 Or performed from the executable instruction that storage section 1108 is loaded into random access storage device (RAM) 1103 it is various appropriate Action and processing.Communication unit 1112 may include but be not limited to network interface card, and the network interface card may include but be not limited to IB (Infiniband) Network interface card.

Processor can communicate with read-only memory 1102 and/or random access storage device 1103 to perform executable instruction, It is connected by bus 1104 with communication unit 1112 and is communicated through communication unit 1112 with other target devices, so as to completes the application The corresponding operation of any one method that embodiment provides, for example, the hyperspectral image data marked in advance is obtained, wherein, in advance The hyperspectral image data of mark includes the markup information of at least part characteristics of image in high spectrum image；It randomly selects in advance A part in the hyperspectral image data of mark is as the first image data；And using the first image data as training data The preset high spectrum image interpretation model of training.

In addition, in RAM 1103, it can also be stored with various programs and data needed for device operation.CPU1101、 ROM1102 and RAM1103 is connected with each other by bus 1104.In the case where there is RAM1103, ROM1102 is optional module. RAM1103 stores executable instruction or executable instruction is written into ROM1102 at runtime, and executable instruction makes processor 1101 perform the corresponding operation of above-mentioned communication means.Input/output (I/O) interface 1105 is also connected to bus 1104.Communication unit 1112 can be integrally disposed, may be set to be with multiple submodule (such as multiple IB network interface cards), and in bus link.

I/O interfaces 1105 are connected to lower component：Importation 1106 including keyboard, mouse etc.；Including such as cathode The output par, c 1107 of ray tube (CRT), liquid crystal display (LCD) etc. and loud speaker etc.；Storage section including hard disk etc. 1108；And the communications portion 1109 of the network interface card including LAN card, modem etc..Communications portion 1109 passes through Communication process is performed by the network of such as internet.Driver 1110 is also according to needing to be connected to I/O interfaces 1105.It is detachable to be situated between Matter 1111, such as disk, CD, magneto-optic disk, semiconductor memory etc. are mounted on driver 1110 as needed, so as to In being mounted into storage section 1108 as needed from the computer program read thereon.

Need what is illustrated, framework as shown in figure 11 is only a kind of optional realization method, can root during concrete practice The component count amount and type of above-mentioned Figure 11 are selected, are deleted, increased or replaced according to actual needs；It is set in different function component Put, can also be used it is separately positioned or integrally disposed and other implementations, such as GPU and CPU separate setting or can be by GPU collection Into on CPU, communication unit separates setting, can also be integrally disposed on CPU or GPU, etc..These interchangeable embodiments Each fall within protection domain disclosed by the invention.

Particularly, in accordance with an embodiment of the present disclosure, it may be implemented as computer above with reference to the process of flow chart description Software program.For example, embodiment of the disclosure includes a kind of computer program product, it is machine readable including being tangibly embodied in Computer program on medium, computer program are included for the program code of the method shown in execution flow chart, program code May include it is corresponding perform the corresponding instruction of method and step provided by the embodiments of the present application, for example, the feature of matching full-colour image with The feature of multispectral image obtains multiple features pair；Feature based pair determines mapping matrix between image；It is mapped according between image Matrix determines the overlapping region of full-colour image and multispectral image；Merge overlapping region and the multispectral image of full-colour image Overlapping region, the remote sensing images after being merged.In such embodiments, which can pass through communications portion 11011 are downloaded and installed from network and/or are mounted from detachable media 1111.In the computer program by centre When managing 1101 execution of unit (CPU), the above-mentioned function of being limited in the present processes is performed.

The method, apparatus and electronic equipment of the present invention may be achieved in many ways.For example, can by software, hardware, Firmware or software, hardware, firmware any combinations realize the method, apparatus of the present invention and electronic equipment.For method Merely to illustrate, the step of method of the invention, is not limited to sequence described in detail above, removes for the said sequence of step It is non-to illustrate in other ways.In addition, in some embodiments, the present invention can be also embodied as recording in the recording medium Program, these programs include being used to implement machine readable instructions according to the method for the present invention.Thus, the present invention also covering storage For performing the recording medium of program according to the method for the present invention.

Description of the invention provides for the sake of example and description, and is not exhaustively or will be of the invention It is limited to disclosed form.Many modifications and variations are obvious for the ordinary skill in the art.It selects and retouches It states embodiment and is to more preferably illustrate the principle of the present invention and practical application, and those of ordinary skill in the art is enable to manage The solution present invention is so as to design the various embodiments with various modifications suitable for special-purpose.

Claims

1. a kind of remote sensing image processing method, which is characterized in that the method includes：

The feature and the feature with reference to remote sensing images that even color remote sensing images are treated in matching, obtain multiple features pair；

Based on the multiple feature pair, Feature Mapping relationship between image is determined；

According to Feature Mapping relationship between described image, the even color remote sensing images and described with reference between remote sensing images treated are determined Overlapping region；

According to the overlapping region for treating even color remote sensing images and the overlapping region with reference to remote sensing images, even color is treated to described Remote sensing images carry out even color processing, obtain the remote sensing images after even color.

2. remote sensing image processing method according to claim 1, which is characterized in that described to be reflected according to feature between described image Relationship is penetrated, determines described to treat that even color remote sensing images and the overlapping region with reference between remote sensing images include：

According to the multiple feature pair, determine described to treat even color remote sensing images with described with reference to initial overlapping between remote sensing images Region；

It treats that even color remote sensing images and the initial overlapping region with reference between remote sensing images carry out cloud detection to described, obtains cloud Masked areas；

The cloud mask in even color remote sensing images and the initial overlapping region with reference between remote sensing images is treated described in removing Region obtains described treating even color remote sensing images and the overlapping region with reference between remote sensing images.

3. according to the remote sensing image processing method described in claim 1 to 2 any one, which is characterized in that described in the basis It treats the overlapping region of even color remote sensing images and the overlapping region with reference to remote sensing images, treats that even color remote sensing images carry out to described Even color processing, obtains the remote sensing images after even color and includes：

According to the overlapping region for treating even color remote sensing images and the overlapping region with reference to remote sensing images, determine it is described treat it is even Mapping relations between the overlapping region and the overlapping region with reference to remote sensing images of color remote sensing images；

It is closed according to the mapping between the overlapping region for treating even color remote sensing images and the overlapping region with reference to remote sensing images System, by the overlapping region with reference to remote sensing images map to described in treat the overlapping regions of even color remote sensing images, to obtain treating even color Overlapping region after remote sensing images mapping；

Even color processing is carried out to the overlapping region after the mapping of even color remote sensing images, obtains the remote sensing images after even color.

4. remote sensing image processing method according to claim 3, which is characterized in that even color remote sensing figure is treated described in the basis Mapping relations between the overlapping region and the overlapping region with reference to remote sensing images of picture, by the weight with reference to remote sensing images Folded area maps obtain the overlapping region packet after even color remote sensing images map to the overlapping region for treating even color remote sensing images It includes：

According to the overlapping region for treating even color remote sensing images with it is mutual corresponding in the overlapping region with reference to remote sensing images Mapping relations between each mapping area, by the overlapping region with reference to remote sensing images map to described in treat even color remote sensing figure The overlapping region of picture obtains the overlapping region after even color remote sensing images map.

5. according to the remote sensing image processing method described in claim 3 to 4 any one, which is characterized in that described to be treated to described Overlapping region after even color remote sensing images mapping carries out even color processing, obtains the remote sensing images after even color and includes：

Choose the extended area of at least one preliminary dimension in the boundary of overlapping region described in after the mapping of even color remote sensing images；

Each extended area is merged respectively, obtains the remote sensing images after even color.

6. remote sensing image processing method according to claim 5, which is characterized in that described respectively to each expansion area Domain is merged, and is obtained the image after even color and is included：

For extended area each described, the image after the even color of the extended area is obtained according to following any one：

The extended area using graph cut algorithm is merged, obtains the image after the even color of the extended area；

Distance of the pixel away from boundary in the extended area carries out linear weighted function fusion to the extended area, obtains Image after the even color of the extended area；And

Split is carried out in the minimum split path that the extended area determines based on dynamic programming algorithm, obtains the expansion area Image after the even color in domain.

7. remote sensing image processing method according to claim 6, which is characterized in that described used to the extended area is moored Loose blending algorithm is merged, and is obtained the image after the even color of the extended area and is included：

Calculate the energy function of the extended area；

Based on the boundary of the extended area, constraint equation is established；

Based on the energy function and the constraint equation, using extended area described in graph cut algorithm fusion, even color is obtained Image afterwards.

8. remote sensing image processing method according to claim 7, which is characterized in that described to be based on the energy function and institute Constraint equation is stated, using extended area described in graph cut algorithm fusion, the image after even color is obtained and includes：

The extended area includes more than one local window for meeting the preliminary dimension, and each local window is included extremely A few pixel；

Based on the energy function and the constraint equation, each window is merged using graph cut algorithm successively, obtains even color Video in window afterwards；

Splice the video in window after all even colors, obtain the image after even color.

9. a kind of remote sensing image processing device, which is characterized in that described device includes：

Feature for matching the feature for treating even color remote sensing images and the feature with reference to remote sensing images, obtains multiple matching unit Feature pair；

Mapping relations determination unit for being based on the multiple feature pair, determines Feature Mapping relationship between image；

Overlapping region determination unit, for according to Feature Mapping relationship between described image, determine it is described treat even color remote sensing images with The overlapping region with reference between remote sensing images；

Even color processing unit, for treating the overlapping region of even color remote sensing images with described with reference to the overlapping of remote sensing images according to Region treats that even color remote sensing images carry out even color processing to described, obtains the remote sensing images after even color.

10. a kind of electronic equipment, which is characterized in that including：

Memory stores executable instruction；

One or more processors communicate with memory and complete following operate to perform executable instruction：