CN101582159A - Infrared image background suppression method based on unsupervised kernel regression analysis - Google Patents

Infrared image background suppression method based on unsupervised kernel regression analysis Download PDF

Info

Publication number
CN101582159A
CN101582159A CN 200910072383 CN200910072383A CN101582159A CN 101582159 A CN101582159 A CN 101582159A CN 200910072383 CN200910072383 CN 200910072383 CN 200910072383 A CN200910072383 A CN 200910072383A CN 101582159 A CN101582159 A CN 101582159A
Authority
CN
China
Prior art keywords
formula
window
background
unsupervised
step
Prior art date
Application number
CN 200910072383
Other languages
Chinese (zh)
Inventor
晔 张
晨 王
谷延锋
Original Assignee
哈尔滨工业大学
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 哈尔滨工业大学 filed Critical 哈尔滨工业大学
Priority to CN 200910072383 priority Critical patent/CN101582159A/en
Publication of CN101582159A publication Critical patent/CN101582159A/en

Links

Abstract

The invention discloses an infrared image background suppression method based on unsupervised kernel regression analysis, belonging to the image processing field. The infrared image background suppression method solves the technical problems that priori knowledge is needed and self-adaptability is poor in the field of infrared image background clutter suppression. Firstly, a sliding window is set to predict background, and Gaussian function is determined to serve as kernel function for unsupervised kernel regression analysis; a background predicting clutter sample is substituted in the function to calculate unsupervised kernel regression equation, and a central test sample (O) is input into the unsupervised kernel regression equation so as to obtain the predicted value of the central test sample (O); then, the central test sample (O) value subtracts the predicted value; the sliding window moves, the above process is repeated until the whole image is processed, and a background suppression result image is output. The invention can effectively improve target detectability and positioning accuracy of an infrared target recognition and tracking system, an infrared image monitoring system, etc.

Description

基于无监督核回归分析的红外图像背景抑制方法 Infrared image background unsupervised kernel regression analysis method of inhibiting

技术领域 FIELD

本发明涉及一种红外图像背景抑制方法,属于图像处理领域。 The present invention relates to a method of inhibiting background infrared image belongs to the field of image processing. 背景技术 Background technique

在红外自动目标探测系统中,为了尽可能早地发现目标,使红外制导系统有足够的反应时间并提高防御武器的预警距离,要求在很远的距离上就能够检测到目标,从而可以尽早发现目标。 Infrared automatic target detection system, in order to find the target as early as possible, the infrared guidance system has enough reaction time and increase warning distance defensive weapons, requiring on long distances will be able to detect the target, which can be found as soon as possible aims. 当探测距离和成像视场增大时,即使目标本身很大,在成像平面内也仅表现为几个像素,甚至不到一个像素,称为小目标。 When the probe and the imaging distance increases the field of view, even if the target itself is large in the imaging plane but also showed only a few pixels, even less than one pixel, called the small target. 此时,可检测信号相对较弱,特别是在非平稳的起伏背景干扰下,目标甚至被大量复杂的噪声(杂波)所淹没,图像信噪比极低,使点目标检测工作变得很困难。 At this time, the detection signal is relatively weak, especially in the non-stationary background noise fluctuation, the target even number of complex noise (clutter) overwhelmed, image noise ratio is very low, so that the work becomes target detection difficult. 因此,为了突出小目标,提高信噪比,从而提高目标检测概率,对红外小目标图像进行检测前的背景抑制和噪声滤除是十分必要的。 Therefore, in order to highlight small target, to improve signal to noise ratio, thereby increasing the probability of target detection, infrared small target image background noise suppression and detection of pre-filtering is necessary. 在自动目标识别系统中,背景抑制和噪声削减统称为检测前滤波预处理。 In the automatic target recognition system, the background noise suppression and reduction referred to as pre-filtering prior to detection. 红外图像背景抑制的最终目的是为了消除杂波干扰、保存目标信息。 The ultimate goal of the infrared image background suppression is to eliminate clutter, save destination information.

目前,在红外图像背景抑制处理中,空域滤波是较为重要、应用广泛的一 Currently, in the infrared image background suppression processing, spatial filtering is more important, a widely used

大类方法。 Categories of methods. 典型的空域滤波方法包括线性背景预测、Top-hat变换(形态学滤波)、 非线性滤波(均值滤波、高斯滤波)等,它们通过空域滤波来实现红外图像背景的预测,利用预测图和原始图做差得到背景抑制的结果,从而达到抑制背景的目的。 Typical background spatial filter comprises a linear prediction method, Top-hat Transform (morphological filtering), nonlinear filtering (mean filter, a Gaussian filter) and the like, are achieved by the predicted image of the background infrared spatial filtering, using the prediction and the original image of FIG. calculating the difference between the results obtained background suppression, so as to achieve the purpose of suppressing background.

近年来,神经网络、支持向量回归等有监督机器学习方法也被逐渐应用于红外图像背景抑制处理中,并取得了较好的抑制效果。 In recent years, neural networks, support vector regression supervised machine learning methods are also increasingly used in infrared background suppression image processing, and achieved good inhibitory effect. 但有监督学习方法需要有大量先验知识(即训练样本),事先进行背景抑制模型的学习训练,无法满足实际处理中无先验知识情况,且不具有自适应性。 However, supervised learning method requires a lot of prior knowledge (ie training samples), background suppression advance learning training model can not meet the actual process no prior knowledge, and not adaptive.

发明内容 SUMMARY

本发明为解决在红外图像背景杂波抑制领域中存在的需要先验知识、自适应性较差的技术难题,提供一种基于无监督核回归分析的红外图像背景抑制方法。 The present invention to solve the problems in the field of infrared image background clutter suppression requires prior knowledge, adaptability to poor technical problems, there is provided a method of inhibiting for Infrared Image unsupervised kernel regression analysis.

本发明包括以下步骤:步骤一、设定滑动窗口用于背景预测;滑动窗口采用双窗口模式,内窗的中心为中心测试样本;外窗中的样本为预测背景杂波样本; The present invention includes the following steps: Step 1, a sliding window is set for the background prediction; sliding window double window mode, the window is centered on the center of the test sample; sample outside the window of background clutter prediction samples;

步骤二、采用高斯函数作为无监督核回归分析的核函数; Step two, using analysis as unsupervised kernel regression Gaussian kernel function;

步骤三、利用当前外窗的预测背景杂波样本信息作为回归数据样值代入到无监督核回归分析的核函数中,计算得到无监督核回归方程; Step three, outside the window using the current prediction information as the background clutter return samples of data samples into substituting unsupervised kernel kernel regression analysis, unsupervised nuclear calculated regression equation;

步骤四、将当前内窗的中心测试样本信息输入到所述的无监督核回归方程,得到中心测试样本的预测杂波灰度值; Step four, the center of the test sample within the current window information is input to the unsupervised kernel regression equation to obtain the prediction value of the center clutter gradation test sample;

步骤五、利用当前内窗的中心测试样本灰度值减去步骤四得到的中心测试样本的预测杂波灰度值,从而抑制红外图像的背景杂波; Step five, the center within the window using the current test sample gradation value by subtracting the prediction value of the center clutter gradation test sample obtained in step four, so as to suppress background clutter in infrared images;

步骤六、移动滑动窗口,其移动步长为l,返回步骤三,直到遍历全图, 输出背景抑制后的红外图像。 Step six, moving sliding window moving step is L, is returned to step three, the infrared images down through the whole map, outputs the background suppression.

本发明与现有技术相比,具有如下优点: Compared with the prior art the present invention has the following advantages:

(1) 采用核回归技术,能够有效地处理强起伏、复杂红外背景杂波数据, 具有良好的非线性数据预测能力。 (1) The regression of nuclear technology, can effectively deal scintillation complex infrared background clutter data, data having good nonlinear predictive power.

(2) 无需事先训练回归模型(即无需训练样本先验知识),根据测试图像自身数据进行无监督学习,具有良好的局部自适应预测能力。 (2) without prior training regression model (that is, without a priori knowledge of training samples), unsupervised learning based on test data of the image itself, with good local adaptive predictive power.

(3) 采用双窗口对红外图像局部区域进行回归预测,抑制背景效果良好, 同时有效保存重要目标,可大大提高如红外目标识别与跟踪、红外图像监测等系统的目标探测能力和定位精度。 (3) dual-image infrared window region local regression prediction, good background suppression effect, while effectively save important objective, can greatly improve the positioning accuracy and target detection target recognition and tracking as infrared, infrared image monitoring system.

附图说明 BRIEF DESCRIPTION

图1为滑动窗口的示意图;图2是滑动窗口的内窗B的中心测试样本O 位于红外图像C的边缘的位置示意图。 FIG 1 is a schematic diagram of sliding window; FIG. 2 is a schematic view of the position of the window B in the test sample of the sliding window center O situated on the edge of the infrared image C. 具体实施方式 Detailed ways

具体实施方式一:结合图1说明本实施方式,本实施方式步骤如下: 步骤一、设定滑动窗口用于背景预测;滑动窗口采用双窗口模式对背景杂波进行预测,滑动窗口由内窗B和外窗A组成;内窗B位于滑动窗口的中心, 内窗B的中心为中心测试样本O,内窗B用于保护位于滑动窗口中心的中心测试样本O信息,中心测试样本O以外的内窗B区域相当于保护区域,防止在测试样本O为目标像素的情况下在选择背景杂波样本的过程中选择到与目标相关的样本,因此产生了保护区域;滑动窗口内窗B的外部为外窗A,外窗A中的样本为预测背景杂波样本,外窗A用于选择预测背景杂波样本信息; A specific embodiment: FIG. 1 explained in conjunction with the present embodiment, the present embodiment the following steps: step a, setting a sliding window used for background prediction; sliding window uses the double window background clutter prediction mode, a sliding window within the window B a composition and an outer window; sliding window within the window B is in the center, the center of the inner window centered test sample B O, B within the window in the center of the test sample to protect the sliding window information center O, the center of the test sample than O protected area corresponding to the window region B, to prevent the case where the test sample as the target pixel O selected background clutter in the sample selection process related to the target sample, thus creating protection areas; external sliding windows for window B a outside the window, the sample a is outside the window background clutter prediction samples outside the window for selecting a prediction a sample background clutter information;

步骤二、采用高斯函数作为无监督核回归分析的核函数;用于度量背景杂波样本之间的相似性; Step two, using unsupervised analysis as a kernel regression Gaussian kernel function; a measure of similarity between the background noise samples;

步骤三、利用当前外窗A的预测背景杂波样本信息作为回归数据样值代入到无监督核回归分析的核函数中,计算得到无监督核回归方程; Step three, background clutter using the current prediction information of the sample outside the window A as the return value into the data samples unsupervised kernel kernel regression analysis, unsupervised nuclear calculated regression equation;

步骤四、将当前内窗B的中心测试样本O信息输入到所述的无监督核回归方程,得到中心测试样本O的预测杂波灰度值; Step four, the information center O test sample within the current window to the B input of the unsupervised kernel regression equation to obtain the prediction value clutter gradation center O of the test sample;

步骤五、利用当前内窗B的中心测试样本O的灰度值减去步骤四得到的中心测试样本O的预测杂波灰度值,从而抑制红外图像的背景杂波;若当前内窗B的中心测试样本O是背景杂波,则所述的中心测试样本O的灰度值将被去除;若当前内窗B的中心测试样本O是目标像素,则抑制叠加在内窗B 的中心测试样本O上的背景杂波灰度值; Step 5 using the current value of the gradation center O of the test sample within the window B gradation values ​​obtained by subtracting the predicted center clutter O test samples obtained in four steps, thereby suppressing the infrared image background clutter; if the current is within the window B O is the center of the test sample background clutter, then the value of the gradation center O of the test sample is to be removed; if the current test sample center O within the window B is the target pixel, the center of the test sample, including window B is superimposed inhibition O on gray background clutter value;

步骤六、移动滑动窗口,其移动步长为l,返回步骤三,直到遍历全图, 输出背景抑制后的红外图像。 Step six, moving sliding window moving step is L, is returned to step three, the infrared images down through the whole map, outputs the background suppression.

具体实施方式二:结合图2说明本实施方式,本实施方式与具体实施方式一不同点在于若当前处理的内窗B的中心测试样本O位于红外图像C的边缘,则滑动窗口中的缺失部分的样本采用镜像对称方式来获得,即采用镜像对称方式来获得缺失部分像素点的灰度值。 DETAILED DESCRIPTION II: in conjunction with FIG. 2 illustrates the present embodiment, the present embodiment and the exemplary first embodiment except that if the current within the window B processing center of the test sample O at the edge of the infrared image C, the sliding missing part of the window the sample obtained using the mirror-symmetrical manner, i.e. using a mirror-symmetrical manner to obtain a gray value pixels of the missing portion. 其它步骤与具体实施方式一相同。 Other steps are the same as a specific embodiment.

具体实施方式三:本实施方式与具体实施方式一不同点在于步骤三中求得无监督核回归方程的步骤如下: DETAILED Embodiment 3: Embodiment of the present embodiment and the exemplary embodiment except that a determined step 3 step unsupervised kernel regression equation is as follows:

回归估计公式如下- Regression estimation formula is as follows -

_y,=z(x,) + s,,/ = l,,户, (1) 其中,;c,.为2xl维向量,表示二维空间的坐标,乂代表相应的图像灰度值; 称作回归函数,e,为随机误差或随机干扰,它是一个分布与^无关的随机变量, 它是均值为0的正态分布随机变量;将芈,)在邻域展开,可叫得到如下公式: _y, = z (x,) + s ,, / = l ,, households, (1) wherein,; c ,. 2xl dimensional vector is, a graph, qe represents the corresponding two-dimensional space of the image intensity; said as a regression function, e, random interference or is a random error, which is a random variable ^ and independent, which is normally distributed random variable with mean 0; and Mi,) to expand in the neighborhood, can be called to obtain the following formula :

<formula>formula see original document page 6</formula><formula>formula see original document page 7</formula> <Formula> formula see original document page 6 </ formula> <formula> formula see original document page 7 </ formula>

而参数&是通过求解下面最优化问题得到的: The parameter & is by solving the following optimization problem is:

<formula>formula see original document page 7</formula>其中<formula>formula see original document page 7</formula>i^W为核加权函数,H称为平滑矩阵; <Formula> formula see original document page 7 </ formula> where <formula> formula see original document page 7 </ formula> i ^ W is a weighting function core, H is called smoothing matrix;

利用数学运算方法进行化简,求得z(;c)的零阶估计值为-<formula>formula see original document page 7</formula> Using a mathematical operation method for simplification, to obtain z (; c) is the estimated value of the zero-order - <formula> formula see original document page 7 </ formula>

其中《是一个第一行元素为i,其它为o的列向量, Wherein "a first row elements i, o the other column vector,

<formula>formula see original document page 7</formula> <Formula> formula see original document page 7 </ formula>

由上面的公式可以看出,评价的结果^c)部分取决于平滑矩阵的选择;这里, 使用一个简单且计算效率较高的模型来表示: As can be seen from the above formula, the results of the evaluation ^ c) depends in part on the selected smoothing matrix; Here, the high computational efficiency and a simple model to represent:

<formula>formula see original document page 7</formula>在上式中,",表征数据采样密集程度(一般令",=1), a称作平滑参数,它的取 <Formula> formula see original document page 7 </ formula> In the above formula, "Characterization data sampling intensive (typically Order", = 1), a smoothing parameter called, it takes

值是通过一系列迭代公式计算出来的;为了计算简便, 一般情况下,/z的取值在"2"附近; Value is calculated through a series of iterative formula; easy to calculate, in general, / z values ​​in the "2" nearby;

这样,只要确定图像局部邻域内每个像素点的灰度值,并按照公式9至公式12计算,将计算结果代入公式8,即可得到局部邻域中心像素的回归估计值; 一般地,核函数均采用高斯径向基函数形式;此时回归核函数形式《及其参数/z和",.如下:K为高斯核函数形式<formula>formula see original document page 7</formula> /z取值为2,",.取值为1。 Thus, as long as the gray scale value is determined for each pixel local neighborhood of an image, and calculated according to Formula 9 to Formula 12, the calculation result is substituted into Equation 8 to obtain a local neighborhood regression estimation value of the center pixel; Generally, nuclear functions are Gaussian radial basis function forms; case regression kernel formation "and its parameters / z and" as follows: K is the Gaussian kernel function of the form <formula> formula see original document page 7 </ formula> / z taken. value is 2, "the value of 1 ,.. 无监督核回归方程是当前外窗A的预测背景杂波样本作为回归数据样值代入到公式3至公式7和公式9至公式12中,计算出Xx和Wx的值后, 代入到公式8中所得到的。 Unsupervised kernel regression equation is the current prediction background clutter samples outside the window A as the return data sample values ​​are substituted into Equation 3 to Equation 7 and Equation 9 to Equation 12, the calculated value Xx and Wx and substituted into Equation 8 obtained. 其它步骤与具体实施方式一相同。 Other steps are the same as a specific embodiment.

本发明内容不仅限于上述各实施方式的内容,其中一个或几个具体实施方式的组合同样也可以实现发明的目的。 SUMMARY The present invention is not limited to the above embodiments in which one or a combination of several specific embodiments also object of the invention can be achieved.

Claims (5)

1、基于无监督核回归分析的红外图像背景抑制方法,其特征在于它的步骤如下: 步骤一、设定滑动窗口用于背景预测;滑动窗口采用双窗口模式,内窗(B)的中心为中心测试样本(O);外窗(A)中的样本为预测背景杂波样本; 步骤二、采用高斯函数作为无监督核回归分析的核函数; 步骤三、利用当前外窗(A)的预测背景杂波样本信息作为回归数据样值代入到无监督核回归分析的核函数中,计算得到无监督核回归方程; 步骤四、将当前内窗(B)的中心测试样本(O)信息输入到所述的无监督核回归方程,得到中心测试样本(O)的预测杂波灰度值; 步骤五、利用当前内窗(B)的中心测试样本(O)灰度值减去步骤四得到的中心测试样本(O)的预测杂波灰度值,从而抑制红外图像的背景杂波; 步骤六、移动滑动窗口,其移动步长为1,返回步骤三,直到遍历全图,输出背景抑 1, based on the infrared image background unsupervised kernel regression analysis method for inhibiting, characterized in that it is the following steps: step a, setting a sliding window used for background prediction; sliding window double window mode, the center of the window (B) for Center the test specimen (O); samples outside the window (a) of the background clutter prediction samples; unsupervised kernel regression analysis as step two, the Gaussian kernel function; step three, outside the window using the current (a) a prediction background clutter sample information into a return data samples substituting the kernel unsupervised kernel regression analysis, calculated unsupervised kernel regression equation; step four, the input current within the window (B) of the center of the test specimen (O) information to the unsupervised kernel regression equation to obtain the prediction value of the center clutter gradation test sample (O); and step five, the window with the current (B) (O) of the center of the test sample gradation value obtained by subtracting the step four Center the test specimen (O) clutter gradation value predicted to suppress background clutter infrared image; step six, moving the sliding window that moves in steps of 1, three returns to step down through the whole map, it outputs the background suppression 后的红外图像。 After the infrared image.
2、 根据权利要求1所述的基于无监督核回归分析的红外图像背景抑制方法,其特征李于若当前处理的内窗(B)的中心测试样本(O)位于红外图像C的边缘,则滑动窗口中的缺失部分的样本采用镜像对称方式来获得。 2, inhibition for Infrared Image unsupervised kernel regression analysis method according to claim 1, wherein Li in when the window (B) is currently processing center of the test specimen (O) at the edge of the infrared image C, the sliding window sample missing portion in a mirror-symmetrical way to get employed.
3、 根据权利要求1所述的基于无监督核回归分析的红外图像背景抑制方法,其特征在于步骤二中计算无监督核回归方程的步骤如下:回归估计公式如下:=啦)+ £,,/ = 1,……,尸, (1)其中,x,.为2xl维向量,表示二维空间的坐标,y,代表相应的图像灰度值;z0c,)称作回归函数,s,.为随机误差或随机干扰;将z(x,)在邻域展开,可以得到如下公式:= A) + — x) + y9/wc/;((x, — — x)r}+… (2)定义m^(.)是对对称矩阵下三角部分的向量化处理:|> 6 (3)、a 6J乂A=Vz(x)=if(4)而参数a是通过求解下面最优化问题得到的:<formula>formula see original document page 3</formula>其中<formula>formula see original document page 3</formula>i^W为核加权函数,/f称为平滑矩阵; 」利用数学运算方法进行化简,求得z(;c)的零阶估计值为:<formula>formula see original document page 3</formula>其中ef是一个第一行元素为1,其它为0 3, suppressing the background image based on the infrared unsupervised kernel regression analysis method according to claim 1, wherein the calculating step two step unsupervised kernel regression equation as follows: regression estimation formula is as follows: it =) + £ ,, / = 1, ......, dead, (1) wherein, x ,. 2xl dimensional vector of coordinates indicating two-dimensional space, y, representing a respective image gray value; z0c,) called regression function, s ,. the z (x,) in the neighborhood deployment, can obtain the following equation; random errors or random interference: = A) + - x) + y9 / wc /; ((x, - - x) r} + ... (2 ) defines m ^ () is the quantization processing on the symmetric matrix lower triangular portion: |.> 6 (3), a 6J qe a = Vz (x) = if (4) and the parameter a is an optimization problem by solving the following obtained: <formula> formula see original document page 3 </ formula> where <formula> formula see original document page 3 </ formula> i ^ W is a weighting function core, / f is called smoothing matrices; "using a mathematical operation method for simplification, to obtain z (; c) is the zero-order estimation: <formula> formula see original document page 3 </ formula> ef wherein a first element of a row, the other is 0 列向量,<formula>formula see original document page 3</formula>由上面的公式可以看出,评价的结果fOc)部分取决于平滑矩阵Z/的选择;平滑矩阵//由模型表示为:<formula>formula see original document page 3</formula>上式中,",表征数据采样密集程度,/;称作平滑参数。 Column vector, <formula> formula see original document page 3 </ formula> can be seen from the above formula, the results of the evaluation fOc) depends in part on smoothing matrix Z / selection; // smoothing matrix is ​​represented by the model: <formula > formula see original document page 3 </ formula> above formula, "Characterization data intensive sampling, /; called smoothing parameters.
4、 根据权利要求1所述的基于无监督核回归分析的红外图像背景抑制方法,其特征在于步骤五中的当前内窗(B)的中心测试样本(O)若是背景杂波,则中心测试样本(O)的灰度值将被去除。 4, according to claim suppressed for Infrared Image unsupervised nuclear regression analysis according to claim 1, wherein the step of this inner window (B) of the fifth center of test sample (O) if the background clutter, the center of the test sample gradation value (O) to be removed.
5、 根据权利要求1所述的基于无监督核回归分析的红外图像背景抑制方法,其特征在于步骤五中的当前内窗(B)的中心测试样本(O)若是目标像素,则抑制叠加在内窗(B)的中心测试样本(O)上的背景杂波的灰度值。 5, according to claim suppressed for Infrared Image unsupervised nuclear regression analysis according to claim 1, wherein the step of this inner window (B) of the fifth center of test sample (O) if the target pixel is suppressed superimposed gray value of the background clutter on the window (B) of the center of the test specimen (O).
CN 200910072383 2009-06-26 2009-06-26 Infrared image background suppression method based on unsupervised kernel regression analysis CN101582159A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN 200910072383 CN101582159A (en) 2009-06-26 2009-06-26 Infrared image background suppression method based on unsupervised kernel regression analysis

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN 200910072383 CN101582159A (en) 2009-06-26 2009-06-26 Infrared image background suppression method based on unsupervised kernel regression analysis

Publications (1)

Publication Number Publication Date
CN101582159A true CN101582159A (en) 2009-11-18

Family

ID=41364295

Family Applications (1)

Application Number Title Priority Date Filing Date
CN 200910072383 CN101582159A (en) 2009-06-26 2009-06-26 Infrared image background suppression method based on unsupervised kernel regression analysis

Country Status (1)

Country Link
CN (1) CN101582159A (en)

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102542540A (en) * 2011-12-31 2012-07-04 华中科技大学 Method for inhibiting infrared image background based on PDE (Partial Differential Equation)
CN103842923A (en) * 2011-07-19 2014-06-04 智能信号公司 System of sequential kernel regression modeling for forecasting and prognostics
CN104104922A (en) * 2014-07-24 2014-10-15 成都市晶林科技有限公司 Archaeological detection system and method
US9256224B2 (en) 2011-07-19 2016-02-09 GE Intelligent Platforms, Inc Method of sequential kernel regression modeling for forecasting and prognostics
US9299008B2 (en) 2012-12-26 2016-03-29 Industrial Technology Research Institute Unsupervised adaptation method and automatic image classification method applying the same
CN107358616A (en) * 2017-06-30 2017-11-17 西安电子科技大学 SAR image edge detection method based on anisotropic morphology direction ratio
CN109033652A (en) * 2018-08-02 2018-12-18 江苏艾佳家居用品有限公司 A kind of indoor autoplacement method based on sliding window feature and regression forecasting

Cited By (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103842923B (en) * 2011-07-19 2016-09-07 智能信号公司 Sequence kernel regression modeling system for forecasting and prediction
CN103842923A (en) * 2011-07-19 2014-06-04 智能信号公司 System of sequential kernel regression modeling for forecasting and prognostics
US9250625B2 (en) 2011-07-19 2016-02-02 Ge Intelligent Platforms, Inc. System of sequential kernel regression modeling for forecasting and prognostics
US9256224B2 (en) 2011-07-19 2016-02-09 GE Intelligent Platforms, Inc Method of sequential kernel regression modeling for forecasting and prognostics
CN102542540A (en) * 2011-12-31 2012-07-04 华中科技大学 Method for inhibiting infrared image background based on PDE (Partial Differential Equation)
CN102542540B (en) 2011-12-31 2014-05-07 华中科技大学 Method for inhibiting infrared image background based on PDE (Partial Differential Equation)
US9299008B2 (en) 2012-12-26 2016-03-29 Industrial Technology Research Institute Unsupervised adaptation method and automatic image classification method applying the same
CN104104922A (en) * 2014-07-24 2014-10-15 成都市晶林科技有限公司 Archaeological detection system and method
CN107358616A (en) * 2017-06-30 2017-11-17 西安电子科技大学 SAR image edge detection method based on anisotropic morphology direction ratio
CN109033652A (en) * 2018-08-02 2018-12-18 江苏艾佳家居用品有限公司 A kind of indoor autoplacement method based on sliding window feature and regression forecasting

Similar Documents

Publication Publication Date Title
Gu et al. A kernel-based nonparametric regression method for clutter removal in infrared small-target detection applications
CN102598057B (en) Method and system for automatic object detection and subsequent object tracking in accordance with the object shape
CN101159855B (en) Characteristic point analysis based multi-target separation predicting method
van der Heijden Edge and line feature extraction based on covariance models
CN101727662B (en) SAR image nonlocal mean value speckle filtering method
Zheng et al. Using combined difference image and $ k $-means clustering for SAR image change detection
Luo et al. Adaptive CUSUM control chart with variable sampling intervals
Hao et al. Persymmetric Rao and Wald tests for partially homogeneous environment
Zhai et al. Recent methods and applications on image edge detection
CN101807301B (en) High spectral image target detection method based on high order statistic
CN1514408A (en) Infra red detecting and tracing method for weak target under complex background condition
An et al. An improved iterative censoring scheme for CFAR ship detection with SAR imagery
CN102509300B (en) Defect detection method and system
Niu et al. A moving objects detection algorithm based on improved background subtraction
CN101403676A (en) Insulator hydrophobicity rank amalgamation judging method based on D-S evidence theory
CN100353152C (en) Method for monitoring rotary kiln cylindrical temperature through infrared scanning
US20120328161A1 (en) Method and multi-scale attention system for spatiotemporal change determination and object detection
Yang et al. Variance WIE based infrared images processing
US20190012548A1 (en) Unified deep convolutional neural net for free-space estimation, object detection and object pose estimation
CN1606033A (en) Weak target detecting and tracking method in infrared image sequence
CN103488988B (en) Extraction power line inspection apparatus UAV visible image based on Insulator
CN101493520B (en) SAR image variation detecting method based on two-dimension gamma distribution
CN101539629B (en) Remote sensing image change detection method based on multi-feature evidence integration and structure similarity
Dai et al. A modified CFAR algorithm based on object proposals for ship target detection in SAR images
Liu et al. Multitarget state extraction for the PHD filter using MCMC approach

Legal Events

Date Code Title Description
C06 Publication
C10 Request of examination as to substance
C02 Deemed withdrawal of patent application after publication (patent law 2001)