US 10,949,703 B2 · Method and system of extraction of impervious surface of remote sensing image

Abstract

A method of extraction of an impervious surface of a remote sensing image. The method includes: 1) obtaining a remote sensing image of a target region, performing normalization for image data, and dividing the normalized target region image into a sample image and a test image; 2) extracting an image feature of each sample image by constructing a deep convolutional network for feature extraction of the remote sensing image; 3) performing pixel-by-pixel category prediction for each sample image respectively; 4) constructing a loss function by using an error between a prediction value and a true value of the sample image and performing update training for network parameters of the deep convolutional network and network parameters relating to the category prediction; and 5) extracting an image feature from the test image through the deep convolutional network based on the training result obtained in 4).

A method, comprising:

1) obtaining a remote sensing image of a target region, performing normalization for image data, and dividing the normalized target region image into a sample image and a test image;
2) extracting an image feature of each sample image by constructing a deep convolutional network for feature extraction of the remote sensing image, wherein the deep convolutional network is formed by a plurality of convolution layers, pooling layers and corresponding unpooling layers and deconvolution layers;
3) performing pixel-by-pixel category prediction for each sample image respectively by using the image feature obtained by extraction;
4) constructing a loss function by using an error between a prediction value and a true value of the sample image and performing update training for network parameters of the deep convolutional network and network parameters relating to the category prediction;
and
5) extracting an image feature from the test image through the deep convolutional network based on a training result obtained in 4), performing the pixel-by-pixel category prediction by using the image feature obtained by extraction, and then, constructing a conditional random field model of the image by using semantic association information between pixel points, performing global optimization for a prediction result of the test image, obtaining a classification result, and realizing impervious surface extraction;
wherein:
an energy function of the conditional random field model comprises both a data item and a smooth item, the data item penalizes inconsistency between a prediction category label and a real category label, the smooth item penalizes inconsistency of adjacent pixel category labels, and the accurate impervious surface extraction is finally realized by balancing a relationship of the two items;
and
xi refers to a label variable of the i-th pixel and is a 0-1 vector, when a pixel corresponding to xi belongs to the k-th terrain category, the k-th element xi,k of xi is 1, others are 0, x={xi}i=1N is a label variable set, and the conditional random field model is expressed as an energy function E(x) as follows:
E
⁡
(
x
)
=
∑
i
⁢
Ψ
u
⁡
(
x
i
)
+
∑
(
i
,
j
)
∈
E
⁢
Ψ
p
⁡
(
x
i
,
x
j
)
wherein, (i,j)∈E means that the i-th pixel and the j-th pixel are adjacent vertexes, and the data item Ψu(xi) constrains a distance between a label and a prediction probability {circumflex over (p)}i,k;
Ψu(xi)=Σi,jΣkxi,k ln {circumflex over (p)}i,k,
wherein the prediction probability {circumflex over (p)}i,k is a probability that the i-th pixel belongs to the k-th category, which is obtained by prediction;
and
the smooth item Ψp(xi,xj) penalizes the inconsistency of labels between adjacent pixels by assigning this case a larger value;
Ψp(xi,xj)=μ(xi,xj)(k(1)(Ii,Ij)+λk(2)(Ii,Ij))
wherein
μ
⁡
(
x
i
,
x
j
)
=
{
1
,
x
i
=
x
j
0
,
else
refers to an indicative function of the label variables xi and xj, k(1)(Ii,Ij) and k(2)(Ii,Ij) require the labels of spatially adjacent pixels to be consistent and adjacent pixel categories with consistent spectrums to be consistent respectively, Ii refers to a spectrum feature of the i-th pixel point, Ij refers to a spectrum feature of the j-th pixel point, and λ is a controllable parameter.

72 more elements

Method and system of extraction of impervious surface of remote sensing image

Life of this patent

Method and system of extraction of impervious surface of remote sensing image

Life of this patent

Abstract

Claims3