CN111144335A - Method and device for building deep learning model - Google Patents

Method and device for building deep learning model Download PDF

Info

Publication number
CN111144335A
CN111144335A CN201911390457.7A CN201911390457A CN111144335A CN 111144335 A CN111144335 A CN 111144335A CN 201911390457 A CN201911390457 A CN 201911390457A CN 111144335 A CN111144335 A CN 111144335A
Authority
CN
China
Prior art keywords
remote sensing
building
sample
sensing image
deep learning
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN201911390457.7A
Other languages
Chinese (zh)
Inventor
王忠武
尤淑撑
杜磊
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Ministry Of Natural Resources Land Satellite Remote Sensing Application Center
Original Assignee
Ministry Of Natural Resources Land Satellite Remote Sensing Application Center
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Ministry Of Natural Resources Land Satellite Remote Sensing Application Center filed Critical Ministry Of Natural Resources Land Satellite Remote Sensing Application Center
Priority to CN201911390457.7A priority Critical patent/CN111144335A/en
Publication of CN111144335A publication Critical patent/CN111144335A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/10—Terrestrial scenes
    • G06V20/176—Urban or other man-made structures
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/11—Region-based segmentation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
    • G06V10/462—Salient features, e.g. scale invariant feature transforms [SIFT]
    • G06V10/464—Salient features, e.g. scale invariant feature transforms [SIFT] using a plurality of salient features, e.g. bag-of-words [BoW] representations
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10032—Satellite or aerial image; Remote sensing
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20084—Artificial neural networks [ANN]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20112—Image segmentation details
    • G06T2207/20132—Image cropping

Abstract

The invention provides a method and a device for constructing a deep learning model of a building, wherein the method for constructing the deep learning model of the building comprises the following steps: aiming at each remote sensing image in a remote sensing image set collected in advance, acquiring morphological building index characteristics of the remote sensing image; obtaining a new remote sensing image sample set based on the morphological building index features and the remote sensing images corresponding to the morphological building index features; marking buildings contained in each new remote sensing image sample of the new remote sensing image sample set to obtain label images, and constructing a sample pair set according to each label image and the new remote sensing image sample corresponding to the label image; and establishing a deep learning model of the building according to the deep learning network and the sample pair set. The building detection precision of the constructed building deep learning model can be improved.

Description

Method and device for building deep learning model
Technical Field
The invention relates to the technical field of remote sensing technology and information science, in particular to a method and a device for building deep learning models.
Background
The deep learning technology, especially the deep learning technology based on the convolutional neural network, has the characteristic of high detection precision, and is widely applied to the field of building detection based on remote sensing images. However, when building deep learning models based on convolutional neural networks are constructed, a large number of sample remote sensing images (data) with small internal feature difference between buildings and large feature difference between buildings and non-buildings are required to be used as supports, but in actual work, massive sample remote sensing images with required quantity and quality are difficult to obtain, and engineering application of building deep learning models based on remote sensing images is severely restricted. For example, in a building deep learning model based on a U-type network such as UNet, which is widely used at present, when the sample remote sensing images are too few to completely cover all types of buildings, the detection accuracy of the buildings in the remote sensing images is not high, and particularly, in the case that the buildings intersect with the ground such as roads and cement pavements, the accuracy of the constructed building deep learning model is low, so that the detection accuracy of the buildings detected by the building deep learning model is low.
Disclosure of Invention
In view of the above, the present invention provides a method and an apparatus for building a deep learning model of a building, so as to improve the detection accuracy of the building deep learning model of the building.
In a first aspect, an embodiment of the present invention provides a method for building a deep learning model of a building, including:
aiming at each remote sensing image in a remote sensing image set collected in advance, acquiring morphological building index characteristics of the remote sensing image;
obtaining a new remote sensing image sample set based on the morphological building index features and the remote sensing images corresponding to the morphological building index features;
marking buildings contained in each new remote sensing image sample of the new remote sensing image sample set to obtain label images, and constructing a sample pair set according to each label image and the new remote sensing image sample corresponding to the label image;
and establishing a deep learning model of the building according to the deep learning network and the sample pair set.
With reference to the first aspect, an embodiment of the present invention provides a first possible implementation manner of the first aspect, where the building deep learning model is built according to a deep learning network and the sample pair set, and the building deep learning model includes:
splitting the sample pair set into a training sample pair set and a test sample pair set;
constructing a U-shaped semantic segmentation deep learning model framework comprising coding, multi-scale features, skip connection, decoding, feature synthesis and classification, and initializing parameters of the U-shaped semantic segmentation deep learning model framework;
taking a training sample pair concentrated remote sensing image new sample as the input of the U-shaped semantic segmentation deep learning model frame, taking a label image corresponding to the remote sensing image new sample as the output of the U-shaped semantic segmentation deep learning model frame, and optimizing the parameters of the U-shaped semantic segmentation deep learning model frame;
inputting the new concentrated remote sensing image sample of the test sample pair into an optimized U-shaped semantic segmentation deep learning model frame to obtain a test result;
and acquiring the precision of the optimized U-shaped semantic segmentation deep learning model frame based on the test result and the label image corresponding to the test result, and if the precision meets a preset precision threshold, taking the optimized U-shaped semantic segmentation deep learning model frame as a building detection deep learning model.
With reference to the first aspect or the first possible implementation manner of the first aspect, an embodiment of the present invention provides a second possible implementation manner of the first aspect, where after the constructing of the sample pairs, before building a building deep learning model according to a deep learning network and the sample pair set, the method further includes:
cutting the sample pair to obtain a cut sample pair;
and calculating the proportion of building pixels in the cutting label images of the cutting sample pairs aiming at each cutting sample pair in the cutting sample pair set, obtaining the cutting sample pairs with the proportion not less than a preset proportion threshold value, obtaining effective sample pairs, and establishing a building deep learning model according to the deep learning network and the effective sample pairs.
With reference to the second possible implementation manner of the first aspect, an embodiment of the present invention provides a third possible implementation manner of the first aspect, where the trimming the sample pair to obtain a trimmed sample pair includes:
and respectively cutting the new remote sensing image sample in the sample pair and the corresponding label image by adopting a preset window size to obtain a cut remote sensing image new sample and a cut label image, and forming a cut sample pair set according to the cut remote sensing image new sample and the cut label image.
With reference to the first aspect or the first possible implementation manner of the first aspect, an embodiment of the present invention provides a fourth possible implementation manner of the first aspect, where the method further includes:
and acquiring a remote sensing image to be detected, inputting the remote sensing image to be detected into the building deep learning model so as to detect the building contained in the remote sensing image to be detected and obtain a building detection result.
With reference to the first aspect or the first possible implementation manner of the first aspect, an embodiment of the present invention provides a fifth possible implementation manner of the first aspect, where the obtaining of the morphological building index feature of the remote sensing image includes:
determining the neighborhood scale and the domain direction of the remote sensing image according to the resolution of the remote sensing image and the average size of the building;
acquiring a multi-scale difference form state sequence corresponding to the remote sensing image according to the neighborhood scale and the domain direction;
and acquiring the morphological building index characteristic of the remote sensing image based on the multi-scale difference morphological sequence, the neighborhood scale and the domain direction.
With reference to the first aspect or the first possible implementation manner of the first aspect, an embodiment of the present invention provides a sixth possible implementation manner of the first aspect, where the obtaining a new sample set of remote sensing images based on the morphological building index feature and the remote sensing images corresponding to the morphological building index feature includes:
adding the morphological building index features serving as single bands to the remote sensing image corresponding to the morphological building index features to obtain a new remote sensing image sample;
and obtaining a new remote sensing image sample set based on the new remote sensing image samples corresponding to the remote sensing images in the remote sensing image set.
In a second aspect, an embodiment of the present invention further provides an apparatus for building a building deep learning model, including:
the MBI characteristic extraction module is used for acquiring morphological building index characteristics of the remote sensing images aiming at each remote sensing image in a remote sensing image set acquired in advance;
the new sample acquisition module is used for obtaining a new sample set of remote sensing images based on the morphological building index features and the remote sensing images corresponding to the morphological building index features;
the sample pair construction module is used for marking buildings contained in each new remote sensing image sample of the new remote sensing image sample set to obtain a label image, and constructing a sample pair set according to each label image and the new remote sensing image sample corresponding to the label image;
and the model building module is used for building a deep learning model of the building according to the deep learning network and the sample pair set.
In a third aspect, an embodiment of the present application provides a computer device, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor, where the processor implements the steps of the method when executing the computer program.
In a fourth aspect, the present application provides a computer-readable storage medium, on which a computer program is stored, which, when executed by a processor, performs the steps of the above-mentioned method.
According to the method and the device for constructing the building deep learning model, provided by the embodiment of the invention, the morphological building index characteristics of the remote sensing image are obtained by aiming at each remote sensing image in the remote sensing image set which is collected in advance; obtaining a new remote sensing image sample set based on the morphological building index features and the remote sensing images corresponding to the morphological building index features; marking buildings contained in each new remote sensing image sample of the new remote sensing image sample set to obtain label images, and constructing a sample pair set according to each label image and the new remote sensing image sample corresponding to the label image; and establishing a deep learning model of the building according to the deep learning network and the sample pair set. Therefore, under the condition that samples are few and can not cover all building types, the morphological building index features of the remote sensing images are calculated one by one, the MBI features are added to the corresponding remote sensing images as independent wave bands, and the MBI features can effectively reduce the difference of the internal features of the building types and enlarge the difference of the features between the building types and the non-building types, so that the parameter optimization in the building deep learning model building process is facilitated, and the building detection precision of the built building deep learning model can be effectively improved.
In order to make the aforementioned and other objects, features and advantages of the present invention comprehensible, preferred embodiments accompanied with figures are described in detail below.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings needed to be used in the embodiments will be briefly described below, it should be understood that the following drawings only illustrate some embodiments of the present invention and therefore should not be considered as limiting the scope, and for those skilled in the art, other related drawings can be obtained from the drawings without inventive effort.
FIG. 1 is a flow chart illustrating a method for building a deep learning model of a building according to an embodiment of the invention;
FIG. 2 is a schematic structural diagram of an apparatus for building a deep learning model of a building according to an embodiment of the present invention;
fig. 3 is a schematic structural diagram of a computer device 300 according to an embodiment of the present application.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. The components of embodiments of the present invention generally described and illustrated in the figures herein may be arranged and designed in a wide variety of different configurations. Thus, the following detailed description of the embodiments of the present invention, presented in the figures, is not intended to limit the scope of the invention, as claimed, but is merely representative of selected embodiments of the invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments of the present invention without any inventive step, are within the scope of protection of the present invention.
The embodiment of the invention provides a method and a device for building deep learning models, which are described in the following through embodiments.
At present, for a building deep learning model constructed based on a remote sensing image with a small sample size, because a high-quality massive remote sensing image is generally difficult to obtain, the building deep learning model constructed based on the remote sensing image has low detection precision on a building, therefore, the embodiment of the invention provides a building deep learning model constructed under the condition of the remote sensing image with the small sample size, and a high-efficiency and high-precision building deep learning model is formed under the conditions of less sample remote sensing images and incomplete types.
Fig. 1 shows a flow chart of a method for building a deep learning model of a building according to an embodiment of the invention. As shown in fig. 1, the method includes:
step 101, aiming at each remote sensing image in a remote sensing image set collected in advance, obtaining the morphological building index characteristics of the remote sensing image;
in the embodiment of the present invention, as an optional embodiment, obtaining a Morphological Building Index (MBI) feature of the remote sensing image includes:
and calculating the morphological building index characteristics of the remote sensing image according to the resolution of the remote sensing image and the average size of the building.
In the embodiment of the invention, the MBI characteristic of each scene of the remote sensing image is calculated according to the resolution of the remote sensing image and the average size of the building. As an alternative embodiment, the calculating the morphological building index feature of the remote sensing image according to the resolution and the average building size of the remote sensing image comprises:
a11, determining the neighborhood scale and the domain direction of the remote sensing image according to the resolution of the remote sensing image and the average size of the building;
in the embodiment of the present invention, as an optional embodiment, the neighborhood scale of the remote sensing image is calculated by using the following formula:
Figure BDA0002344772870000071
in the formula (I), the compound is shown in the specification,
s is the scale of a neighborhood window of the ith scene remote sensing image, namely the domain scale;
BS is the average length of the building;
and PS is the resolution of the remote sensing image.
In the embodiment of the invention, the average building size is the average building length. As an alternative embodiment, the average building length may be obtained by statistical methods.
A12, acquiring a multi-scale difference form sequence corresponding to the remote sensing image according to the neighborhood scale and the domain direction;
in the embodiment of the present invention, as an optional embodiment, the multi-scale difference morphology sequence is calculated by using the following formula:
Figure BDA0002344772870000081
in the formula (I), the compound is shown in the specification,
d is the direction of a neighborhood window of the pixel in the ith scene remote sensing image;
s is the scale of a neighborhood window of the pixel in the ith scene remote sensing image;
TI_DMPi (d,s)a multi-scale difference form sequence corresponding to the ith scene remote sensing image;
TIi (d,s)(b) performing top-hat change based on open reconstruction on the brightness remote sensing image corresponding to the ith scene remote sensing image under the condition that the direction of the neighborhood window is d and the scale is s;
and deltas is the scale difference between adjacent scales.
A13, obtaining the morphological building index features of the remote sensing image based on the multi-scale difference morphological sequence, the neighborhood scale and the domain direction.
In the embodiment of the present invention, as an alternative embodiment, the MBI characteristic is calculated by using the following formula:
Figure BDA0002344772870000082
in the formula (I), the compound is shown in the specification,
TIi_MBIthe MBI characteristic of the ith scene remote sensing image.
102, obtaining a new remote sensing image sample set based on the morphological building index features and the remote sensing images corresponding to the morphological building index features;
in an embodiment of the present invention, as an optional embodiment, obtaining a new sample set of remote sensing images based on the morphological building index feature and the remote sensing images corresponding to the morphological building index feature includes:
a21, adding the morphological building index features serving as separate wave bands into the remote sensing image corresponding to the morphological building index features to obtain a new remote sensing image sample;
in the embodiment of the invention, the MBI characteristics are added to the remote sensing image as the single wave band, so that the internal characteristic difference of various buildings and the characteristic difference between a large building and a non-building are reduced, the parameter optimization of a building deep learning detection model is facilitated, and the building detection precision is improved.
In the embodiment of the invention, for the ith scene remote sensing image TIiK bands, the ith view remote sensing image TIiCan be expressed as:
TIi＝[TIi1,TIi2,...,TIik]
the ith scene remote sensing image TIiThe morphological building index feature of (1) is added to the ith scene remote sensing image TI as a single wave bandiIn the method, the obtained new sample of the remote sensing image is as follows:
TIi_MBI＝[TIi1,TIi2,...,TIik,TIi_MBI]
in the embodiment of the invention, the neighborhood scale is determined according to the resolution of the remote sensing image, the MBI characteristic of the remote sensing image is calculated for each remote sensing image, and the MBI characteristic is added to the corresponding remote sensing image as a single wave band to synthesize a new remote sensing image.
A22, obtaining a new remote sensing image sample set based on the new remote sensing image samples corresponding to the remote sensing images in the remote sensing image set.
In the embodiment of the invention, a new sample set of the remote sensing image consisting of new samples of the remote sensing image is as follows:
newTI＝[TI1_MBI,TI2_MBI,...,TIn_MBI]
wherein the content of the first and second substances,
newTI is a new sample set of the remote sensing image;
n is the number of remote sensing images contained in the remote sensing image set.
103, marking a building contained in each new remote sensing image sample of the new remote sensing image sample set to obtain a label image, and constructing a sample pair set according to each label image and the new remote sensing image sample corresponding to the label image;
in the embodiment of the invention, the building sample marking is carried out on each new remote sensing image sample in the new remote sensing image sample set to generate the corresponding label image. When marking, the complete outline of the building needs to be marked in its entirety.
In the embodiment of the invention, the sample pair set is as follows:
TID＝[TI1_MBI-TI1_label,TI2_MBI-TI2_label,...,TIn_MBI-TIn_label]
wherein the content of the first and second substances,
TID is a sample pair set;
TIi_labeland (3) a label image corresponding to the new sample of the ith scene remote sensing image, wherein i is 1, 2.
In the embodiment of the present invention, as an alternative embodiment, the gray scale value within the marking range (building) is filled with 1, and the gray scale value outside the marking range is filled with 0. Wherein all tag images constitute a set of tag images.
And step 104, establishing a deep learning model of the building according to the deep learning network and the sample pair set.
In the embodiment of the present invention, as an optional embodiment, the building deep learning model is established according to a deep learning network and the pair set of samples, including:
a41, splitting the sample pair set into a training sample pair set and a test sample pair set;
in the embodiment of the invention, the proportion of the training sample pair set and the test sample pair set can be determined according to actual needs. For example, if the sample pair set contains 100 pairs of sample pairs, 80 pairs of sample pairs may be used as training sample pairs, and the remaining 20 pairs of sample pairs may be used as testing sample pairs.
A42, constructing a U-shaped semantic segmentation deep learning model frame comprising coding, multi-scale features, skip connection, decoding, feature synthesis and classification, and initializing parameters of the U-shaped semantic segmentation deep learning model frame;
in the embodiment of the present invention, the parameters of the U-shaped semantic segmentation deep learning model framework include, but are not limited to: coding depth (ED), number of convolution Kernels (KN), multi-scale depth (MD), multi-scale step size (MS), feature synthesis domain size (FCD), loss function type (LT), pooling mode type (PT), activation mode type (AT), etc.
In the embodiment of the present invention, as an optional embodiment, for the coded depth, the maximum coded depth may be calculated according to the sample pair size. Wherein the content of the first and second substances,
ED∈[2,ED_max]
ED_max＝MIN(FLOOR(LOG2(Sample_H/8)),FLOOR(LOG2(Sample_W/8)))
wherein the content of the first and second substances,
ED is the coded depth;
ED_maxis the maximum coding depth;
LOG2 is base 2 logarithmic;
FLOOR is a FLOOR rounding function;
MIN () is a minimum function;
sample _ H is the Sample height in the Sample pair;
sample _ W is the Sample width in the Sample pair.
And for the number of the convolution kernels, determining the number of the convolution kernels in three stages of encoding, multi-scale feature and decoding according to the initial number of the convolution kernels. Wherein the content of the first and second substances,
KN_Encoderi＝IntialKN*2(i-1)
KN_MFCi＝IntialKN*2(ED-1)
KN_Decoderi＝IntialKN*2(ED-i)
wherein the content of the first and second substances,
IntialKN is the number of initial convolution kernels;
KN _ Encoder is the number of convolution kernels of the coding module layer;
KN _ MFC constructs the number of module layer convolution kernels for the multi-scale features;
KN _ Decoder is the number of convolution kernels of the decoding module layer;
i is the number of layers of the coding module.
For multi-scale depth, it can be determined from the minimum detection area, the (spatial) resolution of the remote sensing image.
[FACTORIAL(MD),FACTORIAL(MD+1)]＝SQRT(MinArea)/R/3
Wherein the content of the first and second substances,
MD is multi-scale depth;
FACTORIAL () is a FACTORIAL function;
SQRT is an evolution function;
MinArea is the minimum detection area;
and R is the spatial resolution.
From the multi-scale depth, a multi-scale step size may be determined.
MSi＝i*3
Wherein the content of the first and second substances,
MSiis the multi-scale step size of the ith scale layer.
From the coded depth and the multi-scale depth, a feature synthesis domain size may be determined.
FCD∈{1,ED+MD-1}
Wherein the content of the first and second substances,
FCD is the characteristic composite domain size.
In the embodiment of the present invention, the types of the loss function include, but are not limited to: cross entropy loss functions, Dice loss functions, local loss functions, and the like.
LT∈{CELoss,DiceLoss,FocalLoss}
Wherein the content of the first and second substances,
LT is a loss function type;
CELOSs is a cross entropy loss function;
DiceLoss is a Dice loss function;
focallloss is a local loss function.
Types of pooling include, but are not limited to: maximum pooling, average pooling, etc.
PT∈{MaxPool,AvePool}
PT is a pooling mode type;
pooling with MaxPool as a maximum value;
AvePool pooled as an average.
In the embodiment of the present invention, the activation mode types include, but are not limited to: modified linear activation, vulnerability modified linear activation, truncated modified linear activation, and the like.
AT∈{Relu,LeakyRelu,ClippedRelu}
Wherein the content of the first and second substances,
AT is an activation mode type;
relu is modified linear activation;
LeakyRelu is loophole correction linear activation;
ClippedRelu modifies the linear activation for truncation.
A43, taking a concentrated remote sensing image new sample of a training sample pair as the input of the U-shaped semantic segmentation deep learning model frame, taking a label image corresponding to the remote sensing image new sample as the output of the U-shaped semantic segmentation deep learning model frame, and optimizing the parameters of the U-shaped semantic segmentation deep learning model frame;
in the embodiment of the invention, aiming at each new remote sensing image sample in the training sample pair set, the new remote sensing image sample is used as the input of the U-shaped semantic segmentation deep learning model frame, the label image corresponding to the new remote sensing image sample is used as the output of the U-shaped semantic segmentation deep learning model frame, and the miniband gradient descent method is utilized to carry out parameter optimization training so as to obtain the local optimal value of the parameter.
A44, inputting the concentrated remote sensing image new sample of the test sample pair into an optimized U-shaped semantic segmentation deep learning model frame to obtain a test result;
and A45, acquiring the precision of the optimized U-shaped semantic segmentation deep learning model frame based on the test result and the label image corresponding to the test result, and if the precision meets a preset precision threshold, taking the optimized U-shaped semantic segmentation deep learning model frame as a building detection deep learning model.
In the embodiment of the invention, aiming at each new remote sensing image sample in the test sample pair set, based on the optimized U-shaped semantic segmentation deep learning model frame, the test result corresponding to the new remote sensing image sample can be obtained, the test result is matched with the label image corresponding to the new remote sensing image sample, the matching degree of the test result and the label image is determined, the higher the matching degree is, the higher the precision of the optimized U-shaped semantic segmentation deep learning model frame is, and the precision of the optimized U-shaped semantic segmentation deep learning model frame can be determined through the statistics of the matching degree of each test result and the corresponding label image.
In the embodiment of the invention, if the precision can not meet the preset precision threshold, the number of the sample pairs can be properly increased by adjusting the proportional threshold, so that the number of the samples in the training sample pair set and the test sample pair set is properly increased, and the parameter optimization training can be performed on the optimized U-shaped semantic segmentation deep learning model frame until the output precision of the optimized U-shaped semantic segmentation deep learning model frame meets the precision threshold.
In the embodiment of the invention, the morphological building index characteristic of the remote sensing image is acquired aiming at each remote sensing image in a remote sensing image set acquired in advance; obtaining a new remote sensing image sample set based on the morphological building index characteristics and the remote sensing images corresponding to the morphological building index characteristics; marking buildings contained in each new remote sensing image sample of the new remote sensing image sample set to obtain label images, and constructing a sample pair set according to each label image and the new remote sensing image sample corresponding to the label image; and establishing a deep learning model of the building according to the deep learning network and the sample pair set. Thus, under the condition that the sample is less and can not cover all building types, the morphological building index characteristics are calculated one by one according to the resolution of the remote sensing image and the average size of the building, the MBI characteristics are added into the corresponding remote sensing image as a single wave band, constructing a new sample set of remote sensing images, establishing a deep learning model of the building based on the new sample set of remote sensing images, by adding the MBI characteristics, the method is beneficial to reducing the internal characteristic difference of the building type, enlarging the characteristic difference between the building and the non-building type, and optimizing the parameters of the U-shaped semantic segmentation deep learning model frame, thereby leading the building deep learning model constructed by utilizing the U-shaped semantic segmentation deep learning model frame to have high precision, the building is further identified by using the high-precision building deep learning model, and the identification or detection precision is high; and under the condition of a small amount of samples, the detection precision of the frame of the U-shaped semantic segmentation deep learning model can be guaranteed, so that the high-efficiency and high-precision rapid verification of the building detection sample data is realized.
In this embodiment, as an optional embodiment, after the building of the sample pairs, before the building deep learning model is built according to a deep learning network and the sample pair set, the method further includes:
b11, cutting the sample pair to obtain a cut sample pair;
in this embodiment of the present invention, as an optional embodiment, the cutting the sample pair to obtain a cut sample pair includes:
and respectively cutting the new remote sensing image sample in the sample pair and the corresponding label image by adopting a preset window size to obtain a cut remote sensing image new sample and a cut label image, and forming a cut sample pair set according to the cut remote sensing image new sample and the cut label image.
In an embodiment of the invention, a window of fixed size (TS × TS) is used for each pair of samples (TI) in the set of sample pairsi_MBI-TIi_label) And cutting to obtain a new sample of the cut remote sensing image and a cut label image.
In the embodiment of the invention, after a sample pair is cut, one or more cut sample pairs can be formed, and all cut sample pairs formed by cutting form a cut sample pair set.
In the embodiment of the present invention, as an optional embodiment, the window size is set according to the following formula:
Figure BDA0002344772870000161
wherein the content of the first and second substances,
TS is the window size, and the pixel is taken as a unit;
INT is an integer function;
MGB is Graphics Processing Unit (GPU) card memory capacity, Unit Gb, i.e. GPU card memory capacity of the terminal running the building deep learning model;
k is the number of wave bands contained in the remote sensing image;
GR is the number of gray scale bits of the remote-sensing image.
B12, calculating the proportion of the building pixels in the cutting label images of the cutting sample pairs aiming at each cutting sample pair in the cutting sample pair set, obtaining the cutting sample pairs with the proportion not less than a preset proportion threshold value, obtaining effective sample pairs, and establishing a building deep learning model according to the deep learning network and the effective sample pairs.
In the embodiment of the invention, for each cutting sample pair in the cutting sample pair set, the cutting label image of the cutting sample pair is extracted, and the proportion of the pixels corresponding to the building in the extracted cutting label image to all the pixels in the cutting label image is calculated.
In the embodiment of the present invention, as an optional embodiment, the ratio is calculated according to the following formula:
Figure BDA0002344772870000162
wherein the content of the first and second substances,
Pjin the image of the cutting label of the jth pair of cutting samples, the proportion of the non-building pixels to all the pixels is determined;
SP _ Building is the number of Building pixels in the cropped label image of the jth pair of cropped samples.
In the embodiment of the invention, the size of the cut label image is the window size, so the number of pixels contained in the cut label image is the window size.
In the embodiment of the present invention, as an alternative embodiment, if P of the label image is cutjIf the image size is more than 0.95, deleting the trimming sample pair containing the trimming label image, otherwise, keeping the trimming sample containing the trimming label imageThe pair. Of course, in practical application, the proportion threshold value may also be set according to actual needs, as long as the effective sample pair includes the building pixels and the non-building pixels in the appropriate proportion, so as to form positive and negative samples, which is convenient for building a deep learning model of the building.
In this embodiment, as another optional embodiment, the method further includes:
and acquiring a remote sensing image to be detected, inputting the remote sensing image to be detected into the building deep learning model so as to detect the building contained in the remote sensing image to be detected and obtain a building detection result.
The following describes embodiments of the present invention in detail with reference to specific examples.
From the aerial image data website:
(https:// project. inria. fr/aesialimagelabeling /) selecting a remote sensing image with a square kilometer of 405 and ground truth value data thereof, wherein imaging areas comprise five areas such as austin, chicago, vienna and the like, including cities with dense buildings and suburbs with sparse buildings, each area is provided with a three-waveband true-color remote sensing image with 36 scenes and 5000 x 5000 pixels, the remote sensing image accounts for 180 scenes in total, and the spatial resolution is 0.3 m. Wherein the content of the first and second substances,
and selecting 80 remote sensing images as test remote sensing images, and using the corresponding label images as ground truth values for checking the precision of the data verification of the building sample. And respectively selecting 15, 50 and 100 scene remote sensing images and corresponding label images thereof from the rest 100 scene remote sensing images as training remote sensing images and label images, and obtaining a cut sample pair through sample cutting, removing and other processing. Table 1 shows training samples (training sample pair set) and test samples (test sample pair set) selected in the embodiment of the present invention.
TABLE 1
Figure BDA0002344772870000181
In the embodiment of the invention, a 256 × 256 square window is adopted to cut a new sample of the remote sensing image and the label image.
In the embodiment of the invention, the value of the coding depth is 4, the number of convolution kernels is 64, the multi-scale depth is 1, the multi-scale step length is 1, the size of a characteristic synthesis domain is 5, the loss type is CELOSs, the pooling type is MaxPool, and the activation type is Relu.
In the embodiment of the invention, the maximum iteration number of the parameters of the U-shaped semantic segmentation deep learning model frame is 5, the batch sample size is 12, and the learning rate is 0.01.
In the embodiment of the invention, when the number of effective samples is 3349, 9803 and 20275, the accuracy of the constructed building deep learning model is 0.5441, 0.5814 and 0.6549 respectively, and is 0.5225, 0.4395 and 0.6471 respectively compared with the accuracy of the existing building deep learning model, so that the processing method of the remote sensing image has a better effect; with the increase of the number of samples, the precision of the building deep learning model is stably improved, and compared with the precision of the existing method, the precision is high in time, low in time, extremely unstable and incapable of meeting the requirement of rapid sample verification, the embodiment of the invention has good stability and can meet the requirement of rapid sample verification. As the MBI features can better reduce the feature difference between different buildings and enlarge the feature difference between the buildings and non-buildings, the accuracy of more than 0.54-0.65 can be achieved under the condition that samples are few and can not cover all building types, and the absolute accuracy of the building deep learning model of 2-15 percentage points can be effectively improved.
Table 2 shows a comparison of the Intersection-to-Union ratio (IOU) between the model constructed by the method of the embodiment of the present invention and the existing model.
TABLE 2
Figure BDA0002344772870000191
Figure BDA0002344772870000201
Fig. 2 shows a structural diagram of an apparatus for building a deep learning model of a building according to an embodiment of the present invention. As shown in fig. 2, the apparatus includes:
the MBI characteristic extraction module is used for acquiring morphological building index characteristics of the remote sensing images aiming at each scene of the remote sensing image set collected in advance;
in the embodiment of the invention, the morphological building index characteristic of the remote sensing image is calculated according to the resolution of the remote sensing image and the average size of the building.
In this embodiment of the present invention, as an optional embodiment, the MBI feature extraction module includes:
the scale determining unit (not shown in the figure) is used for determining the neighborhood scale and the domain direction of each remote sensing image in the remote sensing image set which is collected in advance according to the resolution of the remote sensing image and the average size of the building;
the sequence acquisition unit is used for acquiring a multi-scale difference form sequence corresponding to the remote sensing image according to the neighborhood scale and the domain direction;
and the MBI feature extraction unit is used for acquiring the morphological building index features of the remote sensing image based on the multi-scale difference morphological sequence, the neighborhood scale and the domain direction.
The new sample acquisition module is used for obtaining a new sample set of remote sensing images based on the morphological building index features and the remote sensing images corresponding to the morphological building index features;
in the embodiment of the invention, the neighborhood scale is determined according to the resolution of the remote sensing image, the MBI characteristic of the remote sensing image is calculated for each remote sensing image, and the MBI characteristic is added to the corresponding remote sensing image as a single wave band to synthesize a new remote sensing image.
In this embodiment of the present invention, as an optional embodiment, the new sample obtaining module includes:
a feature expansion unit (not shown in the figure) for adding the morphological building index features as separate wave bands to the remote sensing image corresponding to the morphological building index features to obtain a new remote sensing image sample;
and the new sample acquisition unit is used for obtaining a new sample set of the remote sensing images based on the new remote sensing image samples corresponding to the remote sensing images in the remote sensing image set.
The sample pair construction module is used for marking buildings contained in each new remote sensing image sample of the new remote sensing image sample set to obtain a label image, and constructing a sample pair set according to each label image and the new remote sensing image sample corresponding to the label image;
in the embodiment of the invention, the building sample marking is carried out on each new remote sensing image sample in the new remote sensing image sample set to generate the corresponding label image. When marking, the complete outline of the building needs to be marked in its entirety.
In the embodiment of the present invention, as an alternative embodiment, the gray scale value within the marking range (building) is filled with 1, and the gray scale value outside the marking range is filled with 0. Wherein all tag images constitute a set of tag images.
And the model building module is used for building a deep learning model of the building according to the deep learning network and the sample pair set.
In this embodiment, as an optional embodiment, the model building module includes:
a splitting unit (not shown in the figure) for splitting the sample pair set into a training sample pair set and a test sample pair set;
the model initial construction unit is used for constructing a U-shaped semantic segmentation deep learning model framework comprising coding, multi-scale features, skip connection, decoding, feature synthesis and classification, and initializing parameters of the U-shaped semantic segmentation deep learning model framework;
the training unit is used for taking a concentrated remote sensing image new sample of a training sample pair as the input of the U-shaped semantic segmentation deep learning model frame, taking a label image corresponding to the remote sensing image new sample as the output of the U-shaped semantic segmentation deep learning model frame and optimizing the parameters of the U-shaped semantic segmentation deep learning model frame;
in the embodiment of the invention, aiming at each new remote sensing image sample in the training sample pair set, the new remote sensing image sample is used as the input of the U-shaped semantic segmentation deep learning model frame, the label image corresponding to the new remote sensing image sample is used as the output of the U-shaped semantic segmentation deep learning model frame, and the miniband gradient descent method is utilized to carry out parameter optimization training so as to obtain the local optimal value of the parameter.
The testing unit is used for inputting the concentrated remote sensing image new samples of the testing sample pair into the optimized U-shaped semantic segmentation deep learning model frame to obtain a testing result;
and the model acquisition unit is used for acquiring the precision of the optimized U-shaped semantic segmentation deep learning model frame based on the test result and the label image corresponding to the test result, and if the precision meets a preset precision threshold, the optimized U-shaped semantic segmentation deep learning model frame is used as a building detection deep learning model.
In this embodiment of the present invention, as an optional embodiment, the apparatus further includes:
a sample optimizing module (not shown in the figure) for cutting the sample pair to obtain a cut sample pair;
and calculating the proportion of building pixels in the cutting label images of the cutting sample pairs aiming at each cutting sample pair in the cutting sample pair set, obtaining the cutting sample pairs with the proportion not less than a preset proportion threshold value, obtaining effective sample pairs, and establishing a building deep learning model according to the deep learning network and the effective sample pairs.
In this embodiment of the present invention, as an optional embodiment, the cutting the sample pair to obtain a cut sample pair includes:
and respectively cutting the new remote sensing image sample in the sample pair and the corresponding label image by adopting a preset window size to obtain a cut remote sensing image new sample and a cut label image, and forming a cut sample pair set according to the cut remote sensing image new sample and the cut label image.
In this embodiment, as another optional embodiment, the apparatus further includes:
and the detection module (not shown in the figure) is used for acquiring the remote sensing image to be detected, inputting the remote sensing image to be detected into the building deep learning model, and detecting the building contained in the remote sensing image to be detected to obtain a building detection result.
As shown in fig. 3, an embodiment of the present application provides a computer device 300 for executing the method for building a deep learning model of a building in fig. 1, the device includes a memory 301, a processor 302 and a computer program stored in the memory 301 and executable on the processor 302, wherein the processor 302 implements the steps of the method for building a deep learning model of a building when executing the computer program.
Specifically, the memory 301 and the processor 302 can be general-purpose memory and processor, and are not limited to specific examples, and the processor 302 can execute the method for building the deep learning model of the building when executing the computer program stored in the memory 301.
Corresponding to the method for building a deep learning model of a building in fig. 1, an embodiment of the present application further provides a computer-readable storage medium, on which a computer program is stored, and the computer program is executed by a processor to perform the steps of the method for building a deep learning model of a building.
In particular, the storage medium can be a general-purpose storage medium, such as a removable disk, a hard disk, or the like, and when executed, the computer program on the storage medium can execute the above method for constructing the deep learning model of the building.
In the embodiments provided in the present application, it should be understood that the disclosed system and method may be implemented in other ways. The above-described system embodiments are merely illustrative, and for example, the division of the units is merely a logical division, and there may be other divisions when actually implemented, and for example, a plurality of units or components may be combined or integrated into another system, or some features may be omitted, or not executed. In addition, the shown or discussed mutual coupling or direct coupling or communication connection may be an indirect coupling or communication connection of systems or units through some communication interfaces, and may be in an electrical, mechanical or other form.
The units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may also be distributed on a plurality of network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of the embodiment.
In addition, functional units in the embodiments provided in the present application may be integrated into one processing unit, or each unit may exist alone physically, or two or more units are integrated into one unit.
The functions, if implemented in the form of software functional units and sold or used as a stand-alone product, may be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application may be essentially implemented or contributed to by the prior art or parts thereof in the form of a software product stored in a storage medium, and including instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method according to the embodiments of the present application. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk or an optical disk, and other various media capable of storing program codes.
It should be noted that: like reference numbers and letters refer to like items in the following figures, and thus, once an item is defined in a figure, it need not be further defined or explained in subsequent figures, and moreover, the terms "first," "second," "third," etc. are used merely to distinguish one description from another, and are not to be construed as indicating or implying relative importance.
Finally, it should be noted that: the above-mentioned embodiments are merely specific embodiments of the present application, which are used for illustrating the technical solutions of the present application and not for limiting the same, and the protection scope of the present application is not limited thereto, although the present application is described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: any person skilled in the art can modify or easily conceive the technical solutions described in the foregoing embodiments or equivalent substitutes for some technical features within the technical scope disclosed in the present application; such modifications, changes or substitutions do not depart from the spirit and scope of the present disclosure, which should be construed in light of the above teachings. Are intended to be covered by the scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.

Claims (10)

1. A method of constructing a deep learning model of a building, comprising:
aiming at each remote sensing image in a remote sensing image set collected in advance, acquiring morphological building index characteristics of the remote sensing image;
obtaining a new remote sensing image sample set based on the morphological building index features and the remote sensing images corresponding to the morphological building index features;
marking buildings contained in each new remote sensing image sample of the new remote sensing image sample set to obtain label images, and constructing a sample pair set according to each label image and the new remote sensing image sample corresponding to the label image;
and establishing a deep learning model of the building according to the deep learning network and the sample pair set.
2. The method of claim 1, wherein building deep learning models are built from the deep learning network and the sample pair set, comprising:
splitting the sample pair set into a training sample pair set and a test sample pair set;
constructing a U-shaped semantic segmentation deep learning model frame comprising coding, multi-scale features, skip connection, decoding, feature synthesis and classification, and initializing parameters of the U-shaped semantic segmentation deep learning model frame;
taking a training sample pair concentrated remote sensing image new sample as the input of the U-shaped semantic segmentation deep learning model frame, taking a label image corresponding to the remote sensing image new sample as the output of the U-shaped semantic segmentation deep learning model frame, and optimizing the parameters of the U-shaped semantic segmentation deep learning model frame;
inputting the new concentrated remote sensing image sample of the test sample pair into an optimized U-shaped semantic segmentation deep learning model frame to obtain a test result;
and acquiring the precision of the optimized U-shaped semantic segmentation deep learning model frame based on the test result and the label image corresponding to the test result, and if the precision meets a preset precision threshold, taking the optimized U-shaped semantic segmentation deep learning model frame as a building detection deep learning model.
3. The method of claim 1 or 2, wherein after the constructing the sample pairs, before building a building deep learning model from a deep learning network and the set of sample pairs, the method further comprises:
cutting the sample pair to obtain a cut sample pair;
and calculating the proportion of building pixels in the cut label images of the cut sample pairs aiming at each cut sample pair in the cut sample pair set, obtaining the cut sample pairs with the proportion not less than a preset proportion threshold value, obtaining effective sample pairs, and establishing a building deep learning model according to the deep learning network and the effective sample pairs.
4. The method of claim 3, wherein said cropping said sample pair to obtain a cropped sample pair comprises:
and respectively cutting the new remote sensing image sample in the sample pair and the corresponding label image by adopting a preset window size to obtain a cut remote sensing image new sample and a cut label image, and forming a cut sample pair set according to the cut remote sensing image new sample and the cut label image.
5. The method according to claim 1 or 2, characterized in that the method further comprises:
and acquiring a remote sensing image to be detected, inputting the remote sensing image to be detected into the building deep learning model, and detecting the building contained in the remote sensing image to be detected to obtain a building detection result.
6. The method of claim 1 or 2, wherein said obtaining the morphological building index features of the remote sensing image comprises:
determining the neighborhood scale and the domain direction of the remote sensing image according to the resolution of the remote sensing image and the average size of the building;
acquiring a multi-scale difference form sequence corresponding to the remote sensing image according to the neighborhood scale and the domain direction;
and acquiring the morphological building index characteristic of the remote sensing image based on the multi-scale difference morphological sequence, the neighborhood scale and the domain direction.
7. The method according to claim 1 or 2, wherein the obtaining of a new sample set of remote sensing images based on the morphological building index feature and the remote sensing images corresponding to the morphological building index feature comprises:
adding the morphological building index features serving as single bands to the remote sensing image corresponding to the morphological building index features to obtain a new remote sensing image sample;
and obtaining a new remote sensing image sample set based on the new remote sensing image samples corresponding to the remote sensing images in the remote sensing image set.
8. An apparatus for constructing a deep learning model of a building, comprising:
the MBI characteristic extraction module is used for acquiring morphological building index characteristics of the remote sensing images aiming at each remote sensing image in a remote sensing image set acquired in advance;
the new sample acquisition module is used for obtaining a new sample set of remote sensing images based on the morphological building index features and the remote sensing images corresponding to the morphological building index features;
the sample pair construction module is used for marking buildings contained in each new remote sensing image sample of the new remote sensing image sample set to obtain a label image, and constructing a sample pair set according to each label image and the new remote sensing image sample corresponding to the label image;
and the model building module is used for building a deep learning model of the building according to the deep learning network and the sample pair set.
9. An electronic device, comprising: a processor, a memory and a bus, the memory storing machine-readable instructions executable by the processor, the processor and the memory communicating over the bus when the electronic device is running, the machine-readable instructions when executed by the processor performing the steps of the method of building deep learning models according to any one of claims 1 to 7.
10. A computer-readable storage medium, having stored thereon a computer program which, when being executed by a processor, carries out the steps of the method of constructing a deep learning model of a building as claimed in any one of claims 1 to 7.
CN201911390457.7A 2019-12-30 2019-12-30 Method and device for building deep learning model Pending CN111144335A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201911390457.7A CN111144335A (en) 2019-12-30 2019-12-30 Method and device for building deep learning model

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201911390457.7A CN111144335A (en) 2019-12-30 2019-12-30 Method and device for building deep learning model

Publications (1)

Publication Number Publication Date
CN111144335A true CN111144335A (en) 2020-05-12

Family

ID=70521600

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201911390457.7A Pending CN111144335A (en) 2019-12-30 2019-12-30 Method and device for building deep learning model

Country Status (1)

Country Link
CN (1) CN111144335A (en)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112070779A (en) * 2020-08-04 2020-12-11 武汉大学 Remote sensing image road segmentation method based on convolutional neural network weak supervised learning
CN113822220A (en) * 2021-10-09 2021-12-21 海南长光卫星信息技术有限公司 Building detection method and system
CN113989675A (en) * 2021-11-02 2022-01-28 四川睿迈威科技有限责任公司 Geographic information extraction deep learning training sample interactive manufacturing method based on remote sensing image
CN116434009A (en) * 2023-04-19 2023-07-14 应急管理部国家减灾中心(应急管理部卫星减灾应用中心) Construction method and system for deep learning sample set of damaged building

Citations (19)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20100106674A1 (en) * 2009-04-30 2010-04-29 Mclean Donald John Method and system for integrated analysis
CN103839267A (en) * 2014-02-27 2014-06-04 西安科技大学 Building extracting method based on morphological building indexes
CN103984947A (en) * 2014-05-30 2014-08-13 武汉大学 High-resolution remote sensing image house extraction method based on morphological house indexes
CN105528596A (en) * 2016-02-03 2016-04-27 长江大学 High-resolution remote sensing image building automatic extraction method and system by using shadow
CN105608458A (en) * 2015-10-20 2016-05-25 武汉大学 High-resolution remote sensing image building extraction method
CN106683112A (en) * 2016-10-10 2017-05-17 中国交通通信信息中心 High-resolution image-based road region building change extraction method
CN107341795A (en) * 2017-06-30 2017-11-10 武汉大学 A kind of high spatial resolution remote sense image method for detecting automatic variation of Knowledge driving
CN107516100A (en) * 2017-08-31 2017-12-26 北京航天绘景科技有限公司 A kind of image building extracting method based on elevation morphology building index
CN107657246A (en) * 2017-10-19 2018-02-02 武汉大学 A kind of remote sensing image building detection method based on multi-scale filtering building index
CN108596103A (en) * 2018-04-26 2018-09-28 吉林大学 High resolution ratio satellite remote-sensing image building extracting method based on optimal spectrum Index selection
CN108681692A (en) * 2018-04-10 2018-10-19 华南理工大学 Increase Building recognition method in a kind of remote sensing images based on deep learning newly
CN109255334A (en) * 2018-09-27 2019-01-22 中国电子科技集团公司第五十四研究所 Remote sensing image terrain classification method based on deep learning semantic segmentation network
CN109389051A (en) * 2018-09-20 2019-02-26 华南农业大学 A kind of building remote sensing images recognition methods based on convolutional neural networks
CN109446992A (en) * 2018-10-30 2019-03-08 苏州中科天启遥感科技有限公司 Remote sensing image building extracting method and system, storage medium, electronic equipment based on deep learning
CN109472785A (en) * 2018-11-01 2019-03-15 中国石油大学(华东) Based on Multi-structure elements earthquake damaged structure object method for quickly identifying
CN109544579A (en) * 2018-11-01 2019-03-29 上海理工大学 A method of damage building is assessed after carrying out calamity using unmanned plane
CN110175548A (en) * 2019-05-20 2019-08-27 中国科学院光电技术研究所 Remote sensing images building extracting method based on attention mechanism and channel information
CN110321861A (en) * 2019-07-09 2019-10-11 中国水利水电科学研究院 A kind of main crops production moon scale Dynamic Extraction method
CN110503020A (en) * 2019-08-16 2019-11-26 华东师范大学 A method of based on deep learning air remote sensing target detection

Patent Citations (19)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20100106674A1 (en) * 2009-04-30 2010-04-29 Mclean Donald John Method and system for integrated analysis
CN103839267A (en) * 2014-02-27 2014-06-04 西安科技大学 Building extracting method based on morphological building indexes
CN103984947A (en) * 2014-05-30 2014-08-13 武汉大学 High-resolution remote sensing image house extraction method based on morphological house indexes
CN105608458A (en) * 2015-10-20 2016-05-25 武汉大学 High-resolution remote sensing image building extraction method
CN105528596A (en) * 2016-02-03 2016-04-27 长江大学 High-resolution remote sensing image building automatic extraction method and system by using shadow
CN106683112A (en) * 2016-10-10 2017-05-17 中国交通通信信息中心 High-resolution image-based road region building change extraction method
CN107341795A (en) * 2017-06-30 2017-11-10 武汉大学 A kind of high spatial resolution remote sense image method for detecting automatic variation of Knowledge driving
CN107516100A (en) * 2017-08-31 2017-12-26 北京航天绘景科技有限公司 A kind of image building extracting method based on elevation morphology building index
CN107657246A (en) * 2017-10-19 2018-02-02 武汉大学 A kind of remote sensing image building detection method based on multi-scale filtering building index
CN108681692A (en) * 2018-04-10 2018-10-19 华南理工大学 Increase Building recognition method in a kind of remote sensing images based on deep learning newly
CN108596103A (en) * 2018-04-26 2018-09-28 吉林大学 High resolution ratio satellite remote-sensing image building extracting method based on optimal spectrum Index selection
CN109389051A (en) * 2018-09-20 2019-02-26 华南农业大学 A kind of building remote sensing images recognition methods based on convolutional neural networks
CN109255334A (en) * 2018-09-27 2019-01-22 中国电子科技集团公司第五十四研究所 Remote sensing image terrain classification method based on deep learning semantic segmentation network
CN109446992A (en) * 2018-10-30 2019-03-08 苏州中科天启遥感科技有限公司 Remote sensing image building extracting method and system, storage medium, electronic equipment based on deep learning
CN109472785A (en) * 2018-11-01 2019-03-15 中国石油大学(华东) Based on Multi-structure elements earthquake damaged structure object method for quickly identifying
CN109544579A (en) * 2018-11-01 2019-03-29 上海理工大学 A method of damage building is assessed after carrying out calamity using unmanned plane
CN110175548A (en) * 2019-05-20 2019-08-27 中国科学院光电技术研究所 Remote sensing images building extracting method based on attention mechanism and channel information
CN110321861A (en) * 2019-07-09 2019-10-11 中国水利水电科学研究院 A kind of main crops production moon scale Dynamic Extraction method
CN110503020A (en) * 2019-08-16 2019-11-26 华东师范大学 A method of based on deep learning air remote sensing target detection

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
XIN HUANG 等: "Morphological Building/Shadow Index for Building Extraction From High-Resolution Imagery Over Urban Areas", 《IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING》 *
任欣磊 等: "基于改进Ｕ-net的遥感影像建筑物提取", 《激光与光电子学进展》 *
陶 峰 等: "增强型形态学建筑指数的建筑物变化检测", 《测绘科学》 *

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112070779A (en) * 2020-08-04 2020-12-11 武汉大学 Remote sensing image road segmentation method based on convolutional neural network weak supervised learning
CN113822220A (en) * 2021-10-09 2021-12-21 海南长光卫星信息技术有限公司 Building detection method and system
CN113989675A (en) * 2021-11-02 2022-01-28 四川睿迈威科技有限责任公司 Geographic information extraction deep learning training sample interactive manufacturing method based on remote sensing image
CN116434009A (en) * 2023-04-19 2023-07-14 应急管理部国家减灾中心(应急管理部卫星减灾应用中心) Construction method and system for deep learning sample set of damaged building
CN116434009B (en) * 2023-04-19 2023-10-24 应急管理部国家减灾中心(应急管理部卫星减灾应用中心) Construction method and system for deep learning sample set of damaged building

Similar Documents

Publication Publication Date Title
CN111144335A (en) Method and device for building deep learning model
CN110163197B (en) Target detection method, target detection device, computer-readable storage medium and computer equipment
CN112101165B (en) Interest point identification method and device, computer equipment and storage medium
CN109840556B (en) Image classification and identification method based on twin network
CN108520254B (en) Text detection method and device based on formatted image and related equipment
CN108230278B (en) Image raindrop removing method based on generation countermeasure network
CN110929607A (en) Remote sensing identification method and system for urban building construction progress
JP2003523587A (en) Visual attention system
CN111382298B (en) Image retrieval method and device based on picture content and electronic equipment
CN111681273A (en) Image segmentation method and device, electronic equipment and readable storage medium
CN110222598B (en) Video behavior identification method and device, storage medium and server
CN107169425A (en) A kind of recognition methods of item property and device
CN108460400A (en) A kind of hyperspectral image classification method of combination various features information
CN111080641A (en) Crack detection method and device, computer equipment and storage medium
CN114429637B (en) Document classification method, device, equipment and storage medium
CN112070159A (en) Multi-scale sliding window similar picture detection method
CN104036493A (en) No-reference image quality evaluation method based on multifractal spectrum
CN113011420A (en) Character recognition method, model training method, related device and electronic equipment
CN114022770A (en) Mountain crack detection method based on improved self-attention mechanism and transfer learning
Vázquez et al. Using normalized compression distance for image similarity measurement: an experimental study
CN112183542A (en) Text image-based recognition method, device, equipment and medium
CN111899191A (en) Text image restoration method and device and storage medium
CN114494775A (en) Video segmentation method, device, equipment and storage medium
CN115272691A (en) Training method, recognition method and equipment for steel bar binding state detection model
CN113094533B (en) Image-text cross-modal retrieval method based on mixed granularity matching

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
RJ01 Rejection of invention patent application after publication
RJ01 Rejection of invention patent application after publication

Application publication date: 20200512