CN102800101A

CN102800101A - Satellite-borne infrared remote sensing image airport ROI rapid detection method

Info

Publication number: CN102800101A
Application number: CN2012102807779A
Authority: CN
Inventors: 韩军伟; 姚西文; 郭雷; 钱晓亮; 赵天云; 程塨
Original assignee: Northwestern Polytechnical University
Current assignee: Northwestern Polytechnical University
Priority date: 2012-08-09
Filing date: 2012-08-09
Publication date: 2012-11-28

Abstract

The invention relates to a method for quickly detecting an airport ROI of a space-borne infrared remote sensing image. First, the infrared remote sensing image is converted from the RGB color space to the Lab color space; then the remote sensing image is Gaussian filtered to avoid the influence of noise, and then, for the filtered image, the difference between each pixel value and the average value of the image pixel is calculated , get the remote sensing image difference map, and finally perform k-means clustering and segmentation on the remote sensing image difference map to get the airport ROI. Experimental results show that the invention has high speed and high robustness, effectively reduces the processing difficulty of remote sensing images, and has great value and significance for real-time detection of airport targets in remote sensing images.

Description

A fast detection method for airport ROI in space-borne infrared remote sensing images

技术领域 technical field

本发明属于红外遥感图像处理技术领域，具体涉及一种星载红外遥感图像机场ROI快速检测方法，The invention belongs to the technical field of infrared remote sensing image processing, and in particular relates to a rapid detection method for an airport ROI of a spaceborne infrared remote sensing image.

背景技术 Background technique

随着卫星遥感技术的快速发展和成像数据的日益增多，人们迫切需要能够对遥感图像进行智能处理，从中快速、准确地检测出感兴趣目标。机场作为一类特定目标，它的自动检测在飞机导航、军事侦察和精确打击等领域有着重要的实用价值，受到人们越来越多的关注。目前，多数检测算法试图通过检测机场跑道的线性特征、分析跑道的几何特征来解决机场的检测问题，但是遥感图像中存在大量的公路、河流、人工建筑物边缘等具有和机场类似的线性特征，仅依靠线性特征会造成大量的误检、漏检。基于跑道特征的机场检测方法有一定的局限性，无法有效解决遥感图像中的机场检测问题。还有一类算法直接对原始遥感图像采用图像分割的方法得到疑似机场的候选区域，然后对候选区域进行辨识，进一步确认机场目标区域。算法性能严重依赖图像分割的效果，而且速度慢、效率低。With the rapid development of satellite remote sensing technology and the increasing amount of imaging data, people urgently need to be able to intelligently process remote sensing images to quickly and accurately detect objects of interest. As a specific target, the automatic detection of the airport has important practical value in the fields of aircraft navigation, military reconnaissance and precision strike, and has attracted more and more attention. At present, most detection algorithms try to solve the airport detection problem by detecting the linear characteristics of the airport runway and analyzing the geometric characteristics of the runway. However, there are a large number of roads, rivers, and edges of artificial buildings in remote sensing images that have linear characteristics similar to airports. Only relying on linear features will cause a large number of false detections and missed detections. The airport detection method based on runway features has certain limitations and cannot effectively solve the problem of airport detection in remote sensing images. There is another type of algorithm that directly uses the method of image segmentation on the original remote sensing image to obtain the candidate area of the suspected airport, and then identifies the candidate area to further confirm the airport target area. The performance of the algorithm depends heavily on the effect of image segmentation, and it is slow and inefficient.

发明内容 Contents of the invention

要解决的技术问题technical problem to be solved

为了避免现有技术的不足之处，本发明提出一种星载红外遥感图像机场ROI快速检测方法。In order to avoid the deficiencies of the prior art, the present invention proposes a method for rapidly detecting an airport ROI in a space-borne infrared remote sensing image.

技术方案Technical solutions

一种星载红外遥感图像机场ROI快速检测方法，其特征在于包括以下步骤：A kind of airport ROI fast detection method of space-borne infrared remote sensing image is characterized in that comprising the following steps:

步骤1：将红外遥感图像从RGB颜色空间转换到Lab颜色空间，转换后的Lab颜色空间中每一个像素的位置是[L,a,b]^T的向量；Step 1: Convert the infrared remote sensing image from the RGB color space to the Lab color space, and the position of each pixel in the converted Lab color space is a vector of [L, a, b] ^T ;

步骤2：对转换后的Lab颜色空间的遥感图像采用高斯算子进行卷积运算，得到高斯滤波后的图像；所述高斯算子的计算公式为Step 2: Convolute the remote sensing image of the converted Lab color space using a Gaussian operator to obtain a Gaussian filtered image; the calculation formula of the Gaussian operator is

${G G}_{i i,, j j} = = \frac{11}{22 {πσ πσ}^{22}} {e e}^{- - \frac{{((i i - - \frac{n no + + 11}{22}))}^{22} + + {((j j - - \frac{n no + + 11}{22}))}^{22}}{22 {σ σ}^{22}}}$

其中σ为方差，n为高斯算子核矩阵的维数；Where σ is the variance, and n is the dimension of the Gaussian operator kernel matrix;

步骤3：计算高斯滤波后的图像的每一个像素值与图像像素平均值的差值，得到遥感图像差值图Step 3: Calculate the difference between each pixel value of the Gaussian filtered image and the average value of the image pixel, and obtain the remote sensing image difference map

S(x,y)＝||I_μ-I_G(x,y)||S(x,y)＝||I _μ -I _G (x,y)||

其中，I_μ是图像像素特征向量的均值，I_G(x,y)是高斯滤波后的相应的像素特征向量，||||是欧几里得距离；Among them, I _μ is the mean value of the image pixel feature vector, I _G (x, y) is the corresponding pixel feature vector after Gaussian filtering, |||| is the Euclidean distance;

步骤4：在遥感图像差值图上作k均值聚类分割，根据k均值聚类分割的结果在遥感图像差值图上分割出得到机场ROI。Step 4: Carry out k-means clustering segmentation on the remote sensing image difference map, and segment the airport ROI on the remote sensing image difference map according to the result of k-means clustering segmentation.

所述σ＝3，n＝3，k＝2。The σ=3, n=3, k=2.

有益效果Beneficial effect

本发明提出的一种星载红外遥感图像机场ROI快速检测方法，首先将遥感图像从RGB颜色空间转为Lab颜色空间，然后在Lab颜色空间中，对遥感图像进行高斯滤波，去除噪声的影响，接着求取图像中每一个像素值与图像像素均值的差值，得到差值图像，最后对差值图像采用k均值聚类分割，得到机场ROI。The present invention proposes a method for quickly detecting the airport ROI of a space-borne infrared remote sensing image. First, the remote sensing image is converted from the RGB color space to the Lab color space, and then in the Lab color space, the remote sensing image is Gaussian filtered to remove the influence of noise. Then calculate the difference between each pixel value in the image and the average value of the image pixel to obtain the difference image, and finally use k-means clustering to segment the difference image to obtain the airport ROI.

与现有技术相比，本发明并没有对原始遥感图像直接进行分割来得到机场目标候选区域，而是对其进行了一些预处理，在预处理之后的图像上再进行分割，得到机场ROI。本发明减少了计算量，大大提高了处理速度和检测精度，实用性很强。Compared with the prior art, the present invention does not directly segment the original remote sensing image to obtain the airport target candidate area, but performs some preprocessing on it, and then performs segmentation on the preprocessed image to obtain the airport ROI. The invention reduces the calculation amount, greatly improves the processing speed and detection accuracy, and has strong practicability.

附图说明 Description of drawings

图1是本发明的实现流程图；Fig. 1 is the realization flowchart of the present invention;

图2是原始红外遥感图像；Figure 2 is the original infrared remote sensing image;

图3是本发明中遥感图像高斯滤波之后的结果图；Fig. 3 is the result figure after the remote sensing image Gaussian filtering among the present invention;

图4是对图3进行差值计算的结果图；Fig. 4 is the result figure that Fig. 3 is carried out difference calculation;

图5是对图4进行分割的结果图。FIG. 5 is a graph of the result of segmenting FIG. 4 .

具体实施方式 Detailed ways

现结合实施例、附图对本发明作进一步描述：Now in conjunction with embodiment, accompanying drawing, the present invention will be further described:

一种星载红外遥感图像机场ROI快速检测方法的步骤如下：The steps of a method for quickly detecting an airport ROI in a spaceborne infrared remote sensing image are as follows:

步骤1：将红外遥感图像从RGB颜色空间转换到Lab颜色空间，转换的图像每一个像素的位置是[L,a,b]^T的向量。具体步骤如下：Step 1: Convert the infrared remote sensing image from the RGB color space to the Lab color space, and the position of each pixel of the converted image is a vector of [L, a, b] ^T. Specific steps are as follows:

步骤a:从RGB颜色空间转换到XYZ颜色空间，转换公式为Step a: convert from RGB color space to XYZ color space, conversion formula is

$[\begin{matrix} X x \\ Y Y \\ Z Z \end{matrix}] = = [\begin{matrix} 0.412 0.412 & 0.358 0.358 & 0.180 0.180 \\ 0.213 0.213 & 0.715 0.715 & 0.072 0.072 \\ 0.019 0.019 & 0.119 0.119 & 0.950 0.950 \end{matrix}] [\begin{matrix} R R \\ G G \\ B B \end{matrix}]$

步骤b：从XYZ颜色空间转换到Lab颜色空间，转换公式为Step b: Convert from XYZ color space to Lab color space, the conversion formula is

$L L = = \{\begin{matrix} 166166 * * {((Y Y / / {Y Y}_{n no}))}^{11 / / 33} - - 1616 & Y Y / / {Y Y}_{n no} > > 0.008856 0.008856 \\ 903.3 903.3 * * ((Y Y / / {Y Y}_{n no})) & Y Y / / {Y Y}_{n no} \leq \leq 0.008856 0.008856 \end{matrix}$

a＝500*(f(X/X_n)-f(Y/Y_n))a=500*(f(X/X _n )-f(Y/Y _n ))

b＝200*(f(Y/Y_n)-f(Z/Z_n))b=200*(f(Y/Y _n )-f(Z/Z _n ))

其中：X_n、Y_n、Z_n是白光的三色刺激值；Among them: X _n , Y _n , Z _n are tristimulus values of white light;

$f f ((t t)) = = \{\begin{matrix} {t t}^{11 / / 33} & t t > > 0.008856 0.008856 \\ 7.787 7.787 t t + + 1616 / / 116116 & t t \leq \leq 0.008856 0.008856 \end{matrix}$

步骤2：对转换为Lab颜色空间的遥感图像采用高斯算子进行卷积运算，得到高斯滤波后的图像，离散高斯算子核矩阵计算公式为Step 2: Convolute the remote sensing image converted into the Lab color space using a Gaussian operator to obtain a Gaussian filtered image. The calculation formula of the discrete Gaussian operator kernel matrix is

其中σ为方差，n为高斯核矩阵的维数；Where σ is the variance and n is the dimension of the Gaussian kernel matrix;

步骤3：对于高斯模糊后的图像，求每一个像素值与图像像素平均值的差值，得到遥感图像差值图；Step 3: For the Gaussian blurred image, calculate the difference between each pixel value and the average value of the image pixel to obtain the remote sensing image difference map;

S(x,y)＝||I_μ-I_G(x,y)||S(x,y)＝||I _μ -I _G (x,y)||

其中，I_μ是图像像素特征向量的均值，I_G(x,y)是高斯滤波后的相应的像素特征向量。||||是欧几里得距离。Among them, I _μ is the mean value of image pixel feature vector, and I _G (x, y) is the corresponding pixel feature vector after Gaussian filtering. |||| is the Euclidean distance.

步骤4：在遥感图像差值图上作k均值聚类分割，具体步骤如下：Step 4: Perform k-means clustering and segmentation on the remote sensing image difference map, the specific steps are as follows:

步骤a：任意设定2个初始质心即分割灰度值f₁，f₂；Step a: Arbitrarily set two initial centroids, that is, segment gray values f ₁ and f ₂ ;

步骤b：把遥感图像差值图的每一个像素p分配到灰度值与之相距最近的那个质心所代表的聚类中，背景赋值为0，目标赋值为255；Step b: Assign each pixel p of the remote sensing image difference map to the cluster represented by the centroid closest to the gray value, assign the background value to 0, and assign the target value to 255;

步骤c：根据公式计算分配后的新的质心f₁、f₂，其中N_j是第j类中像素数目，g(x)表示类内像素p的灰度值，j＝1,2；Step c: According to the formula Calculate the new centroids f ₁ and f ₂ after assignment, where N _j is the number of pixels in the jth class, g(x) represents the gray value of the pixel p in the class, j=1,2;

步骤d：重复步骤b和步骤c直到f₁和f₂不再改变，便可在遥感图像差值图上分割出机场ROI。Step d: Repeat step b and step c until f ₁ and f ₂ do not change, then the airport ROI can be segmented on the remote sensing image difference map.

具体实施例的硬件环境是：因特尔酷睿2双核2.93G计算机、2.0GB内存、512M显卡，运行的软件环境是：Matlab R2011a，Windows XP。我们用Matlab软件实现了本发明提出的方法。原始遥感图像选用了一张分辨率为400*400的红外遥感图像。The hardware environment of specific embodiment is: Intel Core 2 duo 2.93G computer, 2.0GB memory, 512M graphics card, the software environment of operation is: Matlab R2011a, Windows XP. We have realized the method that the present invention proposes with Matlab software. The original remote sensing image uses an infrared remote sensing image with a resolution of 400*400.

本发明具体实施如下：The present invention is specifically implemented as follows:

1、颜色空间转换：将将原始红外遥感图像2从RGB颜色空间转换到Lab颜色空间，转换的图像每一个像素的位置是[L,a,b]^T的向量。具体转换步骤为：1. Color space conversion: the original infrared remote sensing image 2 will be converted from the RGB color space to the Lab color space, and the position of each pixel of the converted image is a vector of [L, a, b] ^T. The specific conversion steps are:

1）从RGB颜色空间转换到XYZ颜色空间，转换公式为1) Convert from RGB color space to XYZ color space, the conversion formula is

2）从XYZ颜色空间转换到Lab颜色空间，转换公式为2) Convert from XYZ color space to Lab color space, the conversion formula is

a＝500*(f(X/X_n)-f(Y/Y_n))a=500*(f(X/X _n )-f(Y/Y _n ))

b＝200*(f(Y/Y_n)-f(Z/Z_n))b=200*(f(Y/Y _n )-f(Z/Z _n ))

其中：X_n、Y_n、Z_n是白光的三色刺激值。Among them: X _n , Y _n , Z _n are tristimulus values of white light.

2、高斯滤波：采用高斯算子对转换为Lab颜色空间的遥感图像进行卷积滤波运算，离散高斯核矩阵计算公式为2. Gaussian filtering: Gaussian operators are used to perform convolution filtering operations on remote sensing images converted into Lab color space, and the calculation formula of discrete Gaussian kernel matrix is

其中σ为方差，n为高斯核矩阵的维数。本例中选用σ＝3，n＝3，计算高斯核矩阵并对其进行归一化，得到归一化后的离散高斯核矩阵为where σ is the variance and n is the dimension of the Gaussian kernel matrix. In this example, select σ=3, n=3, calculate the Gaussian kernel matrix and normalize it, and obtain the normalized discrete Gaussian kernel matrix as

$[\begin{matrix} 0.1070 0.1070 & 0.1131 0.1131 & 0.1070 0.1070 \\ 0.1131 0.1131 & 0.1196 0.1196 & 0.1131 0.1131 \\ 0.1070 0.1070 & 0.1131 0.1131 & 0.1070 0.1070 \end{matrix}]$

图3即为对图2采用此高斯核进行卷积滤波后的结果图。Figure 3 is the result of convolution filtering using the Gaussian kernel in Figure 2.

3、计算差值图：对于高斯滤波后的图像3，求每一个像素值与图像像素平均值的差值，得到遥感图像差值图。3. Calculate the difference map: For the Gaussian filtered image 3, calculate the difference between each pixel value and the average value of the image pixels to obtain the remote sensing image difference map.

S(x,y)＝||I_μ-I_G(x,y)||S(x,y)＝||I _μ -I _G (x,y)||

其中，I_μ是图像像素特征向量的均值，I_G(x,y)是高斯滤波后的相应的像素特征向量。||||是欧几里得距离。本例中，图4即是对图3进行差值计算的结果图。Among them, I _μ is the mean value of image pixel feature vector, and I _G (x, y) is the corresponding pixel feature vector after Gaussian filtering. |||| is the Euclidean distance. In this example, Figure 4 is the result of calculating the difference of Figure 3 .

4、分割：在遥感图像差值图4上作k均值聚类分割，具体步骤如下：4. Segmentation: Perform k-means clustering segmentation on the remote sensing image difference map 4, the specific steps are as follows:

步骤b：把遥感图像差值图的每一个像素p分配到灰度值与之相距最近的那个质心所代表的聚类中，背景赋值为0，目标赋值为255。Step b: Assign each pixel p of the difference image of the remote sensing image to the cluster represented by the centroid closest to the gray value, the background is assigned a value of 0, and the target is assigned a value of 255.

步骤c：根据公式计算分配后的新的质心f₁、f₂，其中N_j是第j类中像素数目，g(x)表示类内像素p的灰度值，j＝1,2。Step c: According to the formula Calculate the new centroids f ₁ and f ₂ after assignment, where N _j is the number of pixels in the jth class, g(x) represents the gray value of the pixel p in the class, and j=1,2.

步骤d：重复步骤b和步骤c直到f₁和f₂不再改变，便可在遥感图像差值图上分割出机场ROI，得到结果图5。Step d: Repeat step b and step c until f ₁ and f ₂ do not change, then the airport ROI can be segmented on the remote sensing image difference map, and the result shown in Figure 5 is obtained.

Claims

1. spaceborne infrared remote sensing image airport ROI method for quick is characterized in that may further comprise the steps:

Step 1: with the infrared remote sensing image from the RGB color space conversion to the Lab color space, each locations of pixels is [L, a, b] in the Lab color space after the conversion ^TVector;

Step 2: the remote sensing images to the Lab color space after the conversion adopt Gauss operator to carry out convolution algorithm, obtain the image behind the gaussian filtering; The computing formula of said Gauss operator does

G_{i, j} = \frac{1}{2 {πσ}^{2}} e^{- \frac{{(i - \frac{n + 1}{2})}^{2} + {(j - \frac{n + 1}{2})}^{2}}{2 σ^{2}}}

Wherein σ is a variance, and n is the dimension of Gauss operator nuclear matrix;

Step 3: each pixel value of the image behind the calculating gaussian filtering and the difference of image pixel mean value obtain the remote sensing images differential chart

S(x,y)＝||I _μ-I _G(x,y)||

Wherein, I _μBe the average of image pixel proper vector, I _G(x y) is corresponding pixel characteristic vector behind the gaussian filtering, || || be Euclidean distance;

Step 4: on the remote sensing images differential chart, make the k mean cluster and cut apart, the result of cutting apart according to the k mean cluster is partitioned on the remote sensing images differential chart and obtains airport ROI.

2. a kind of spaceborne infrared remote sensing image according to claim 1 airport ROI method for quick is characterized in that: said σ=3, n=3, k=2.