CN106128037A - A kind of monitoring early-warning device for natural disaster - Google Patents

Abstract

The invention provides a kind of monitoring early-warning device for natural disaster, including monitoring early-warning device and coupled speckle registration apparatus, described monitoring early-warning device includes: mounting shell.The present invention has that images match precision is of a relatively high, the comparatively faster advantage of matching speed.

Description

A kind of monitoring early-warning device for natural disaster

Technical field

The present invention relates to monitoring field, be specifically related to a kind of monitoring early-warning device for natural disaster.

Background technology

Image registration be by different time, with different view or with different sensors shooting Same Scene two width or Multiple image spatially carries out the process alignd.Its main purpose be eliminate reference picture with between image subject to registration by becoming slice Geometric deformation caused by part difference, so that the two has Space Consistency.Image registration algorithm of the prior art can By the complementarity between enhancing information, reduce the uncertainty that scene is understood.In process of image registration, due to Same Scene Imaging results show bigger vision difference, bring bigger difficulty to registration, and existing registration Algorithm is in precision, effect The aspect such as rate, Stability and adaptability can not fully meet application demand.

Summary of the invention

For the problems referred to above, the present invention provides a kind of monitoring early-warning device for natural disaster.

The purpose of the present invention realizes by the following technical solutions:

A kind of monitoring early-warning device for natural disaster, registrates dress including monitoring early-warning device and coupled speckle Putting, described monitoring early-warning device includes: mounting shell, is provided with for regular, timing acquiring natural disaster scene in described mounting shell The Monitoring Data of situation and the collection in worksite module of graphic record, for storing data and the expert of photo information that expert provides System module, the data provided for Monitoring Data that collection in worksite module is collected and graphic record and expert system module It is fitted analyzing and providing the Monitoring Data Fitting Analysis module of analytical data, divided by Monitoring Data matching with photo information The analysis analytical data that is given of module carry out the long-term trend prediction of disaster, disaster alarm in mid-term, face calamity disaster alert prediction pre- Alert model module, it is characterised in that described mounting shell is provided with the function keyboard for inputting control command, for picture and text and number According to the display screen of display, for providing the battery area of power supply for equipment, showing pre-for the operating switch and being used for of starting device The display lamp of alert signal, described Monitoring Data Fitting Analysis module, prediction and warning model module are divided by display screen video data Analysis and prediction and warning result, and provide early warning signal by display lamp, also include embedded memory card, USB interface, serial line interface and Communication interface, described embedded memory card for natural disaster basic data and monitoring and warning process and result thereof are stored, Described USB interface, serial line interface and communication interface are for transmitting the data in embedded memory card.

Preferably, described collection in worksite module includes for taking pictures, imaging, and forms comprehensive image, attribute data loading High-definition camera.

Preferably, described mounting shell is additionally provided with it is connected with Monitoring Data Fitting Analysis module, prediction and warning model module The speech input interface connect and voice output interface.

Preferably, described speckle registration apparatus includes: pretreatment module, feature detection module, feature description module, feature Matching module and spatial alternation module；

(1) pretreatment module, for reference picture and image subject to registration are converted into gray level image, defines conversion formula For:

I (x, y)=k (G (x, y)+R (x, y)+B (x, y))+2k (G (and x, y)+R (x, y))+3k (x, y)

Wherein, (x, y) at coordinate, (x, y) grey scale pixel value at place, (x, y), (x, y), (x y) distinguishes B R G representative image I (x, y) the red, green, blue intensity level at place, k is the weight factor set, and the span of k is [0.08,0.12] to represent coordinate；

(2) feature detection module, including local features detection sub-module and Patch properties detection sub-module, described office District of portion

Characteristic of field detection sub-module for detecting the local of pretreated two images by Mexican hat wavelet function Provincial characteristics, described Patch properties detection sub-module is for using the local extremum of approximation Hessian matrix at custom dimensions Space is detected the Patch properties in pretreated two images, the locus of output Patch properties and the feature chi at place Degree；

The description form of described approximation Hessian matrix is:

$H(x,y,\mathrm{\σ})=\left[\begin{array}{cc}{E}_{xx}(x,y,\mathrm{\σ})& E_{xy}(x,y,\mathrm{\σ})\\ E_{xy}(x,y,\mathrm{\σ})& E_{yy}(x,y,\mathrm{\σ})\end{array}\right]$

In formula, σ is the standard deviation of Gaussian function, i.e. scale factor；E_xx(x,y,σ)、E_xy(x,y,σ)、E_yy(x, y, σ) point Wei Gauss second-order differentialApproximation template after discretization and cutting point (x, y) place with The convolution of image；Set E_xx、E_xyAnd E_yyRepresent the convolution results of first, second, and third 9 × 9 approximation template and image, E_xx、 E_xyAnd E_yyComputing formula be defined as follows:

$E_{xx}=\frac{1}{P}(A_1-2A_2+A_3)$

$E_{xy}=\frac{1}{Q}(A_1-A_2-A_3)$

$E_{yy}=\frac{1}{P}(A_1-2A_2+A_3)$

Wherein, take the first approximation template 3 × 3 regions from left to right as the first marked region, take in the second approximation template Region, portion 2 × 2, as the second marked region, takes the 3rd approximation template 3 × 3 regions from top to bottom as the 3rd marked region, A₁、 A₂And A₃Be respectively first, second, and third marked region cover lower image pixel gray level and, P, Q are marked region area, divide Deng Yu 9 and 4；

The characteristic point receptance function of described approximation Hessian matrix is:

DET (H)=E_xxE_yy-(0.9E_xy)²

The description form of described Mexican hat wavelet function is:

$M(x,y,\mathrm{\σ})=(1-q)(2-\frac{x^2+y^2}{{\mathrm{\σ}}^2})\exp (-\frac{x^2+y^2}{2{\mathrm{\σ}}^2})$

Wherein, q is the running parameter constructing described custom dimensions space, and the relation between q and σ is σ=2^-q；

Preferably,

(3) feature description module, the Patch properties detected is described also by it by setting up weighted intensity description Formed describe vector, set up weighted intensity describe the period of the day from 11 p.m. to 1 a.m by centered by Patch properties, be perpendicular to Patch properties principal direction and Size is that the central area of l × l is divided into multiple sub-block, described weighted intensity to describe son to be:

WD={NP⁺,NP^-}

Herein

P⁺=Σ { f (D_i)×d(m_i)|d(m_i)=I (m_i)-I (m) ＞ 0, i=1,2 ... l²}

P^-=Σ { f (D_i)×d(m_i)|d(m_i)=I (m_i)-I (m) ＜ 0, i=1,2 ... l²}

In formula, WD represents that weighted intensity describes son, P⁺Represent positive gray scale difference value histogram, P^-Represent negative gray scale difference value Nogata Figure, I (m_i) and I (m) be respectively the gray average of each sub-block and the ash of whole central area using bilinear interpolation to ask for Degree average, D_iFor the distance of each sub-block Yu center, wherein, i=1,2 ... l², f (D_i) represent weighting function, NP⁺、NP^-Expression is returned The one positive and negative gray scale difference rectangular histogram changed；

(4) characteristic matching module, for mating pretreated reference picture and image subject to registration, including successively The regional area matched sub-block, the region that connect divide submodule, feature packet submodule and Patch properties matched sub-block, institute State regional area matched sub-block for the local features of pretreated reference picture and image subject to registration being carried out Joining, described region divides submodule for carrying out pretreated reference picture and image subject to registration according to local features Image-region after image-region divides and will divide is converted into standard round region, if the image-region quantity divided is N, N's Span is [200,1000], and described feature packet submodule is for the model divided according to image-region by described Patch properties Enclosing and be grouped, described Patch properties matched sub-block is for carrying out the description vector representing Patch properties in each group Join；

(5) spatial alternation module, for being mapped to the coordinate of reference picture by image subject to registration by geometric transformation model In system, completing image registration, the parameter of described geometric transformation model uses RANSAC algorithm to estimate.

Wherein, described custom dimensions space is divided into many groups, and often group comprises the filter template of three different scales；Described In first group of custom dimensions space, template size corresponding to smallest dimension is 9 × 9, and marked region increment is set to 4, template Increment is 12, and second is followed successively by 21 × 21 and 33 × 33 with the 3rd template size that template is corresponding；Except custom dimensions is empty Between first group outside other groups in, first template often organized is identical with the second of previous group template size, and template increase Amount is 4 times of previous group.

Wherein, the center of gravity that the center of circle is local features in described standard round region, the radius in standard round region is local Point in edges of regions is to the average of described centroidal distance.

The invention have the benefit that

1, the image pre-processing module arranged considers visual custom and the human eye same color of perceptibility to different color The non-linear relation of intensity, it is possible to describe image the most accurately；

2, the Patch properties detection sub-module arranged, it is possible to make full use of cell type based on integral image filtering and wave filter The characteristic that size is unrelated, constant speed builds the metric space of image, and owing to there is no the down-sampled operation of image, it is possible to avoid aliasing Phenomenon occurs；

3, the Patch properties detected is described by feature description module by setting up weighted intensity description, it is possible to more Utilize the local message in feature neighborhood to build fully and describe vector；

4, the characteristic matching module arranged first carries out the Patch properties point division that regional area coupling carries out in group again, carries The high speed of images match, and the region being provided with divides the image-region after submodule will divide and is converted into standard round district Territory, reduces the complexity that the packet of Patch properties point judges, because whether Patch properties point is positioned at region and only need to compare its limit The edge distance to the center of circle and the radius of circle, also reduce the impact on packet registration of the regional area extraction precision simultaneously, because of For not using edges of regions to use edge to do, to the average of regional barycenter, the border divided.

Accompanying drawing explanation

The invention will be further described to utilize accompanying drawing, but the embodiment in accompanying drawing does not constitute any limit to the present invention System, for those of ordinary skill in the art, on the premise of not paying creative work, it is also possible to obtain according to the following drawings Other accompanying drawing.

Fig. 1 is that monitoring early-warning device of the present invention forms schematic diagram.

Fig. 2 is the connection diagram of each module of speckle registration apparatus of the present invention.

Detailed description of the invention

The invention will be further described with the following Examples.

Embodiment 1

See Fig. 1, Fig. 2, a kind of monitoring early-warning device for natural disaster of the present embodiment, including monitoring early-warning device and Coupled speckle registration apparatus, described monitoring early-warning device includes: mounting shell, be provided with in described mounting shell for periodically, The Monitoring Data of timing acquiring natural disaster field conditions and the collection in worksite module of graphic record, for storing what expert provided Data and the expert system module of photo information, for the Monitoring Data that collection in worksite module is collected and graphic record and specially Data and photo information that family's system module provides are fitted analyzing and providing the Monitoring Data Fitting Analysis mould of analytical data Block, the analytical data be given by Monitoring Data Fitting Analysis module carry out the long-term trend prediction of disaster, disaster alarm in mid-term, Face the prediction and warning model module of calamity disaster alert, it is characterised in that described mounting shell is provided with for inputting control command Function keyboard, the display screen shown for picture and text and data, for providing the battery area of power supply for equipment, for starting device Operating switch and for showing the display lamp of early warning signal, described Monitoring Data Fitting Analysis module, prediction and warning model module By display screen display data analysis and prediction and warning result, and provide early warning signal by display lamp, also include built-in storage Card, USB interface, serial line interface and communication interface, described embedded memory card is for natural disaster basic data and monitoring and warning Process and result thereof store, and described USB interface, serial line interface and communication interface are for transmitting the number in embedded memory card According to.

Preferably, described collection in worksite module includes for taking pictures, imaging, and forms comprehensive image, attribute data loading High-definition camera.

Preferably, described mounting shell is additionally provided with it is connected with Monitoring Data Fitting Analysis module, prediction and warning model module The speech input interface connect and voice output interface.

Preferably, described speckle registration apparatus includes: pretreatment module, feature detection module, feature description module, feature Matching module and spatial alternation module；

(1) pretreatment module, for reference picture and image subject to registration are converted into gray level image, defines conversion formula For:

I (x, y)=k (G (x, y)+R (x, y)+B (x, y))+2k (G (and x, y)+R (x, y))+3k (x, y)

Wherein, (x, y) at coordinate, (x, y) grey scale pixel value at place, (x, y), (x, y), (x y) distinguishes B R G representative image I (x, y) the red, green, blue intensity level at place, k is the weight factor set, and the span of k is [0.08,0.12] to represent coordinate；

(2) feature detection module, including local features detection sub-module and Patch properties detection sub-module, described office District of portion

Characteristic of field detection sub-module for detecting the local of pretreated two images by Mexican hat wavelet function Provincial characteristics, described Patch properties detection sub-module is for using the local extremum of approximation Hessian matrix at custom dimensions Space is detected the Patch properties in pretreated two images, the locus of output Patch properties and the feature chi at place Degree；

The description form of described approximation Hessian matrix is:

$H(x,y,\mathrm{\σ})=\left[\begin{array}{cc}{E}_{xx}(x,y,\mathrm{\σ})& E_{xy}(x,y,\mathrm{\σ})\\ E_{xy}(x,y,\mathrm{\σ})& E_{yy}(x,y,\mathrm{\σ})\end{array}\right]$

In formula, σ is the standard deviation of Gaussian function, i.e. scale factor；E_xx(x,y,σ)、E_xy(x,y,σ)、E_yy(x, y, σ) point Wei Gauss second-order differentialApproximation template after discretization and cutting point (x, y) place with The convolution of image；Set E_xx、E_xyAnd E_yyRepresent the convolution results of first, second, and third 9 × 9 approximation template and image, E_xx、 E_xyAnd E_yyComputing formula be defined as follows:

$E_{xx}=\frac{1}{P}(A_1-2A_2+A_3)$

$E_{xy}=\frac{1}{Q}(A_1-A_2-A_3)$

$E_{yy}=\frac{1}{P}(A_1-2A_2+A_3)$

Wherein, take the first approximation template 3 × 3 regions from left to right as the first marked region, take in the second approximation template Region, portion 2 × 2, as the second marked region, takes the 3rd approximation template 3 × 3 regions from top to bottom as the 3rd marked region, A₁、 A₂And A₃Be respectively first, second, and third marked region cover lower image pixel gray level and, P, Q are marked region area, divide Deng Yu 9 and 4；

The characteristic point receptance function of described approximation Hessian matrix is:

DET (H)=E_xxE_yy-(0.9E_xy)²

The description form of described Mexican hat wavelet function is:

$M(x,y,\mathrm{\σ})=(1-q)(2-\frac{x^2+y^2}{{\mathrm{\σ}}^2})\exp (-\frac{x^2+y^2}{2{\mathrm{\σ}}^2})$

Wherein, q is the running parameter constructing described custom dimensions space, and the relation between q and σ is σ=2^-q；

Preferably,

(3) feature description module, the Patch properties detected is described also by it by setting up weighted intensity description Formed describe vector, set up weighted intensity describe the period of the day from 11 p.m. to 1 a.m by centered by Patch properties, be perpendicular to Patch properties principal direction and Size is that the central area of l × l is divided into multiple sub-block, described weighted intensity to describe son to be:

WD={NP⁺,NP^-}

Herein

P⁺=Σ { f (D_i)×d(m_i)|d(m_i)=I (m_i)-I (m) ＞ 0, i=1,2 ... l²}

P^-=Σ { f (D_i)×d(m_i)|d(m_i)=I (m_i)-I (m) ＜ 0, i=1,2 ... l²}

In formula, WD represents that weighted intensity describes son, P⁺Represent positive gray scale difference value histogram, P^-Represent negative gray scale difference value Nogata Figure, I (m_i) and I (m) be respectively the gray average of each sub-block and the ash of whole central area using bilinear interpolation to ask for Degree average, D_iFor the distance of each sub-block Yu center, wherein, i=1,2 ... l², f (D_i) represent weighting function, NP⁺、NP^-Expression is returned The one positive and negative gray scale difference rectangular histogram changed；

(4) characteristic matching module, for mating pretreated reference picture and image subject to registration, including successively The regional area matched sub-block, the region that connect divide submodule, feature packet submodule and Patch properties matched sub-block, institute State regional area matched sub-block for the local features of pretreated reference picture and image subject to registration being carried out Joining, described region divides submodule for carrying out pretreated reference picture and image subject to registration according to local features Image-region after image-region divides and will divide is converted into standard round region, if the image-region quantity divided is N, N's Span is [200,1000], and described feature packet submodule is for the model divided according to image-region by described Patch properties Enclosing and be grouped, described Patch properties matched sub-block is for carrying out the description vector representing Patch properties in each group Join；

(5) spatial alternation module, for being mapped to the coordinate of reference picture by image subject to registration by geometric transformation model In system, completing image registration, the parameter of described geometric transformation model uses RANSAC algorithm to estimate.

Wherein, described custom dimensions space is divided into many groups, and often group comprises the filter template of three different scales；Described In first group of custom dimensions space, template size corresponding to smallest dimension is 9 × 9, and marked region increment is set to 4, template Increment is 12, and second is followed successively by 21 × 21 and 33 × 33 with the 3rd template size that template is corresponding；Except custom dimensions is empty Between first group outside other groups in, first template often organized is identical with the second of previous group template size, and template increase Amount is 4 times of previous group.

Wherein, the center of gravity that the center of circle is local features in described standard round region, the radius in standard round region is local Point in edges of regions is to the average of described centroidal distance.

The image pre-processing module that the present embodiment is arranged considers visual custom and the human eye perceptibility to different color Non-linear relation with colouring intensity, it is possible to describe image the most accurately；The Patch properties detection sub-module arranged, it is possible to etc. Speed builds the metric space of image, and it can be avoided that aliasing occurs；The feature description module arranged is by setting up Weighted Grey Degree describes son and is described the Patch properties detected, it is possible to utilizes the local message in feature neighborhood to build more fully and retouches State vector；The characteristic matching module arranged first carries out the Patch properties point division that regional area coupling carries out in group again, improves The speed of images match, and the region being provided with divides the image-region after submodule will divide and is converted into standard round district Territory, reduces complexity and the impact on packet registration of the regional area extraction precision that the packet of Patch properties point judges.This reality Executing example weight factor k value is 0.08, and the image-region quantity N value of division is 200, and images match precision improves relatively 1%, matching speed improves 3%.

Embodiment 2

See Fig. 1, Fig. 2, a kind of monitoring early-warning device for natural disaster of the present embodiment, including monitoring early-warning device and Coupled speckle registration apparatus, described monitoring early-warning device includes: mounting shell, be provided with in described mounting shell for periodically, The Monitoring Data of timing acquiring natural disaster field conditions and the collection in worksite module of graphic record, for storing what expert provided Data and the expert system module of photo information, for the Monitoring Data that collection in worksite module is collected and graphic record and specially Data and photo information that family's system module provides are fitted analyzing and providing the Monitoring Data Fitting Analysis mould of analytical data Block, the analytical data be given by Monitoring Data Fitting Analysis module carry out the long-term trend prediction of disaster, disaster alarm in mid-term, Face the prediction and warning model module of calamity disaster alert, it is characterised in that described mounting shell is provided with for inputting control command Function keyboard, the display screen shown for picture and text and data, for providing the battery area of power supply for equipment, for starting device Operating switch and for showing the display lamp of early warning signal, described Monitoring Data Fitting Analysis module, prediction and warning model module By display screen display data analysis and prediction and warning result, and provide early warning signal by display lamp, also include built-in storage Card, USB interface, serial line interface and communication interface, described embedded memory card is for natural disaster basic data and monitoring and warning Process and result thereof store, and described USB interface, serial line interface and communication interface are for transmitting the number in embedded memory card According to.

Preferably, described collection in worksite module includes for taking pictures, imaging, and forms comprehensive image, attribute data loading High-definition camera.

Preferably, described mounting shell is additionally provided with it is connected with Monitoring Data Fitting Analysis module, prediction and warning model module The speech input interface connect and voice output interface.

Preferably, described speckle registration apparatus includes: pretreatment module, feature detection module, feature description module, feature Matching module and spatial alternation module；

(1) pretreatment module, for reference picture and image subject to registration are converted into gray level image, defines conversion formula For:

I (x, y)=k (G (x, y)+R (x, y)+B (x, y))+2k (G (and x, y)+R (x, y))+3k (x, y)

Wherein, (x, y) at coordinate, (x, y) grey scale pixel value at place, (x, y), (x, y), (x y) distinguishes B R G representative image I (x, y) the red, green, blue intensity level at place, k is the weight factor set, and the span of k is [0.08,0.12] to represent coordinate；

(2) feature detection module, including local features detection sub-module and Patch properties detection sub-module, described office District of portion

Characteristic of field detection sub-module for detecting the local of pretreated two images by Mexican hat wavelet function Provincial characteristics, described Patch properties detection sub-module is for using the local extremum of approximation Hessian matrix at custom dimensions Space is detected the Patch properties in pretreated two images, the locus of output Patch properties and the feature chi at place Degree；

The description form of described approximation Hessian matrix is:

$H(x,y,\mathrm{\σ})=\left[\begin{array}{cc}{E}_{xx}(x,y,\mathrm{\σ})& E_{xy}(x,y,\mathrm{\σ})\\ E_{xy}(x,y,\mathrm{\σ})& E_{yy}(x,y,\mathrm{\σ})\end{array}\right]$

In formula, σ is the standard deviation of Gaussian function, i.e. scale factor；E_xx(x,y,σ)、E_xy(x,y,σ)、E_yy(x, y, σ) point Wei Gauss second-order differentialApproximation template after discretization and cutting point (x, y) place with The convolution of image；Set E_xx、E_xyAnd E_yyRepresent the convolution results of first, second, and third 9 × 9 approximation template and image, E_xx、 E_xyAnd E_yyComputing formula be defined as follows:

$E_{xx}=\frac{1}{P}(A_1-2A_2+A_3)$

$E_{xy}=\frac{1}{Q}(A_1-A_2-A_3)$

$E_{yy}=\frac{1}{P}(A_1-2A_2+A_3)$

Wherein, take the first approximation template 3 × 3 regions from left to right as the first marked region, take in the second approximation template Region, portion 2 × 2, as the second marked region, takes the 3rd approximation template 3 × 3 regions from top to bottom as the 3rd marked region, A₁、 A₂And A₃Be respectively first, second, and third marked region cover lower image pixel gray level and, P, Q are marked region area, divide Deng Yu 9 and 4；

The characteristic point receptance function of described approximation Hessian matrix is:

DET (H)=E_xxE_yy-(0.9E_xy)²

The description form of described Mexican hat wavelet function is:

$M(x,y,\mathrm{\σ})=(1-q)(2-\frac{x^2+y^2}{{\mathrm{\σ}}^2})\exp (-\frac{x^2+y^2}{2{\mathrm{\σ}}^2})$

Wherein, q is the running parameter constructing described custom dimensions space, and the relation between q and σ is σ=2^-q；

Preferably,

(3) feature description module, the Patch properties detected is described also by it by setting up weighted intensity description Formed describe vector, set up weighted intensity describe the period of the day from 11 p.m. to 1 a.m by centered by Patch properties, be perpendicular to Patch properties principal direction and Size is that the central area of l × l is divided into multiple sub-block, described weighted intensity to describe son to be:

WD={NP⁺,NP^-}

Herein

P⁺=Σ { f (D_i)×d(m_i)|d(m_i)=I (m_i)-I (m) ＞ 0, i=1,2 ... l²}

P^-=Σ { f (D_i)×d(m_i)|d(m_i)=I (m_i)-I (m) ＜ 0, i=1,2 ... l²}

In formula, WD represents that weighted intensity describes son, P⁺Represent positive gray scale difference value histogram, P^-Represent negative gray scale difference value Nogata Figure, I (m_i) and I (m) be respectively the gray average of each sub-block and the ash of whole central area using bilinear interpolation to ask for Degree average, D_iFor the distance of each sub-block Yu center, wherein, i=1,2 ... l², f (D_i) represent weighting function, NP⁺、NP^-Expression is returned The one positive and negative gray scale difference rectangular histogram changed；

(4) characteristic matching module, for mating pretreated reference picture and image subject to registration, including successively The regional area matched sub-block, the region that connect divide submodule, feature packet submodule and Patch properties matched sub-block, institute State regional area matched sub-block for the local features of pretreated reference picture and image subject to registration being carried out Joining, described region divides submodule for carrying out pretreated reference picture and image subject to registration according to local features Image-region after image-region divides and will divide is converted into standard round region, if the image-region quantity divided is N, N's Span is [200,1000], and described feature packet submodule is for the model divided according to image-region by described Patch properties Enclosing and be grouped, described Patch properties matched sub-block is for carrying out the description vector representing Patch properties in each group Join；

(5) spatial alternation module, for being mapped to the coordinate of reference picture by image subject to registration by geometric transformation model In system, completing image registration, the parameter of described geometric transformation model uses RANSAC algorithm to estimate.

Wherein, described custom dimensions space is divided into many groups, and often group comprises the filter template of three different scales；Described In first group of custom dimensions space, template size corresponding to smallest dimension is 9 × 9, and marked region increment is set to 4, template Increment is 12, and second is followed successively by 21 × 21 and 33 × 33 with the 3rd template size that template is corresponding；Except custom dimensions is empty Between first group outside other groups in, first template often organized is identical with the second of previous group template size, and template increase Amount is 4 times of previous group.

Wherein, the center of gravity that the center of circle is local features in described standard round region, the radius in standard round region is local Point in edges of regions is to the average of described centroidal distance.

The image pre-processing module that the present embodiment is arranged considers visual custom and the human eye perceptibility to different color Non-linear relation with colouring intensity, it is possible to describe image the most accurately；The Patch properties detection sub-module arranged, it is possible to etc. Speed builds the metric space of image, and it can be avoided that aliasing occurs；The feature description module arranged is by setting up Weighted Grey Degree describes son and is described the Patch properties detected, it is possible to utilizes the local message in feature neighborhood to build more fully and retouches State vector；The characteristic matching module arranged first carries out the Patch properties point division that regional area coupling carries out in group again, improves The speed of images match, and the region being provided with divides the image-region after submodule will divide and is converted into standard round district Territory, reduces complexity and the impact on packet registration of the regional area extraction precision that the packet of Patch properties point judges.This reality Executing example weight factor k value is 0.09, and the image-region quantity N value of division is 400, and images match precision improves relatively 1.2%, matching speed improves 2.5%.

Embodiment 3

See Fig. 1, Fig. 2, a kind of monitoring early-warning device for natural disaster of the present embodiment, including monitoring early-warning device and Coupled speckle registration apparatus, described monitoring early-warning device includes: mounting shell, be provided with in described mounting shell for periodically, The Monitoring Data of timing acquiring natural disaster field conditions and the collection in worksite module of graphic record, for storing what expert provided Data and the expert system module of photo information, for the Monitoring Data that collection in worksite module is collected and graphic record and specially Data and photo information that family's system module provides are fitted analyzing and providing the Monitoring Data Fitting Analysis mould of analytical data Block, the analytical data be given by Monitoring Data Fitting Analysis module carry out the long-term trend prediction of disaster, disaster alarm in mid-term, Face the prediction and warning model module of calamity disaster alert, it is characterised in that described mounting shell is provided with for inputting control command Function keyboard, the display screen shown for picture and text and data, for providing the battery area of power supply for equipment, for starting device Operating switch and for showing the display lamp of early warning signal, described Monitoring Data Fitting Analysis module, prediction and warning model module By display screen display data analysis and prediction and warning result, and provide early warning signal by display lamp, also include built-in storage Card, USB interface, serial line interface and communication interface, described embedded memory card is for natural disaster basic data and monitoring and warning Process and result thereof store, and described USB interface, serial line interface and communication interface are for transmitting the number in embedded memory card According to.

Preferably, described collection in worksite module includes for taking pictures, imaging, and forms comprehensive image, attribute data loading High-definition camera.

Preferably, described mounting shell is additionally provided with it is connected with Monitoring Data Fitting Analysis module, prediction and warning model module The speech input interface connect and voice output interface.

Preferably, described speckle registration apparatus includes: pretreatment module, feature detection module, feature description module, feature Matching module and spatial alternation module；

(1) pretreatment module, for reference picture and image subject to registration are converted into gray level image, defines conversion formula For:

I (x, y)=k (G (x, y)+R (x, y)+B (x, y))+2k (G (and x, y)+R (x, y))+3k (x, y)

Wherein, (x, y) at coordinate, (x, y) grey scale pixel value at place, (x, y), (x, y), (x y) distinguishes B R G representative image I (x, y) the red, green, blue intensity level at place, k is the weight factor set, and the span of k is [0.08,0.12] to represent coordinate；

(2) feature detection module, including local features detection sub-module and Patch properties detection sub-module, described office District of portion

Characteristic of field detection sub-module for detecting the local of pretreated two images by Mexican hat wavelet function Provincial characteristics, described Patch properties detection sub-module is for using the local extremum of approximation Hessian matrix at custom dimensions Space is detected the Patch properties in pretreated two images, the locus of output Patch properties and the feature chi at place Degree；

The description form of described approximation Hessian matrix is:

$H(x,y,\mathrm{\σ})=\left[\begin{array}{cc}{E}_{xx}(x,y,\mathrm{\σ})& E_{xy}(x,y,\mathrm{\σ})\\ E_{xy}(x,y,\mathrm{\σ})& E_{yy}(x,y,\mathrm{\σ})\end{array}\right]$

In formula, σ is the standard deviation of Gaussian function, i.e. scale factor；E_xx(x,y,σ)、E_xy(x,y,σ)、E_yy(x, y, σ) point Wei Gauss second-order differentialApproximation template after discretization and cutting point (x, y) place with The convolution of image；Set E_xx、E_xyAnd E_yyRepresent the convolution results of first, second, and third 9 × 9 approximation template and image, E_xx、 E_xyAnd E_YYComputing formula be defined as follows:

$E_{xx}=\frac{1}{P}(A_1-2A_2+A_3)$

$E_{xy}=\frac{1}{Q}(A_1-A_2-A_3)$

$E_{yy}=\frac{1}{P}(A_1-2A_2+A_3)$

Wherein, take the first approximation template 3 × 3 regions from left to right as the first marked region, take in the second approximation template Region, portion 2 × 2, as the second marked region, takes the 3rd approximation template 3 × 3 regions from top to bottom as the 3rd marked region, A₁、 A₂And A₃Be respectively first, second, and third marked region cover lower image pixel gray level and, P, Q are marked region area, divide Deng Yu 9 and 4；

The characteristic point receptance function of described approximation Hessian matrix is:

DET (H)=E_xxE_yy-(0.9E_xy)²

The description form of described Mexican hat wavelet function is:

$M(x,y,\mathrm{\σ})=(1-q)(2-\frac{x^2+y^2}{{\mathrm{\σ}}^2})\exp (-\frac{x^2+y^2}{2{\mathrm{\σ}}^2})$

Wherein, q is the running parameter constructing described custom dimensions space, and the relation between q and σ is σ=2^-q；

Preferably,

(3) feature description module, the Patch properties detected is described also by it by setting up weighted intensity description Formed describe vector, set up weighted intensity describe the period of the day from 11 p.m. to 1 a.m by centered by Patch properties, be perpendicular to Patch properties principal direction and Size is that the central area of l × l is divided into multiple sub-block, described weighted intensity to describe son to be:

WD={NP⁺,NP^-}

Herein

P⁺=Σ { f (D_i)×d(m_i)|d(m_i)=I (m_i)-I (m) ＞ 0, i=1,2 ... l²}

P^-=Σ { f (D_i)×d(m_i)|d(m_i)=I (m_i)-I (m) ＜ 0, i=1,2 ... l²}

In formula, WD represents that weighted intensity describes son, P⁺Represent positive gray scale difference value histogram, P^-Represent negative gray scale difference value Nogata Figure, I (m_i) and I (m) be respectively the gray average of each sub-block and the ash of whole central area using bilinear interpolation to ask for Degree average, D_iFor the distance of each sub-block Yu center, wherein, i=1,2 ... l², f (D_i) represent weighting function, NP⁺、NP^-Expression is returned The one positive and negative gray scale difference rectangular histogram changed；

(4) characteristic matching module, for mating pretreated reference picture and image subject to registration, including successively The regional area matched sub-block, the region that connect divide submodule, feature packet submodule and Patch properties matched sub-block, institute State regional area matched sub-block for the local features of pretreated reference picture and image subject to registration being carried out Joining, described region divides submodule for carrying out pretreated reference picture and image subject to registration according to local features Image-region after image-region divides and will divide is converted into standard round region, if the image-region quantity divided is N, N's Span is [200,1000], and described feature packet submodule is for the model divided according to image-region by described Patch properties Enclosing and be grouped, described Patch properties matched sub-block is for carrying out the description vector representing Patch properties in each group Join；

(5) spatial alternation module, for being mapped to the coordinate of reference picture by image subject to registration by geometric transformation model In system, completing image registration, the parameter of described geometric transformation model uses RANSAC algorithm to estimate.

Wherein, described custom dimensions space is divided into many groups, and often group comprises the filter template of three different scales；Described In first group of custom dimensions space, template size corresponding to smallest dimension is 9 × 9, and marked region increment is set to 4, template Increment is 12, and second is followed successively by 21 × 21 and 33 × 33 with the 3rd template size that template is corresponding；Except custom dimensions is empty Between first group outside other groups in, first template often organized is identical with the second of previous group template size, and template increase Amount is 4 times of previous group.

Wherein, the center of gravity that the center of circle is local features in described standard round region, the radius in standard round region is local Point in edges of regions is to the average of described centroidal distance.

The image pre-processing module that the present embodiment is arranged considers visual custom and the human eye perceptibility to different color Non-linear relation with colouring intensity, it is possible to describe image the most accurately；The Patch properties detection sub-module arranged, it is possible to etc. Speed builds the metric space of image, and it can be avoided that aliasing occurs；The feature description module arranged is by setting up Weighted Grey Degree describes son and is described the Patch properties detected, it is possible to utilizes the local message in feature neighborhood to build more fully and retouches State vector；The characteristic matching module arranged first carries out the Patch properties point division that regional area coupling carries out in group again, improves The speed of images match, and the region being provided with divides the image-region after submodule will divide and is converted into standard round district Territory, reduces complexity and the impact on packet registration of the regional area extraction precision that the packet of Patch properties point judges.This reality Executing example weight factor k value is 0.10, and the image-region quantity N value of division is 600, and images match precision improves relatively 1.8%, matching speed improves 2.1%.

Embodiment 4

See Fig. 1, Fig. 2, a kind of monitoring early-warning device for natural disaster of the present embodiment, including monitoring early-warning device and Coupled speckle registration apparatus, described monitoring early-warning device includes: mounting shell, be provided with in described mounting shell for periodically, The Monitoring Data of timing acquiring natural disaster field conditions and the collection in worksite module of graphic record, for storing what expert provided Data and the expert system module of photo information, for the Monitoring Data that collection in worksite module is collected and graphic record and specially Data and photo information that family's system module provides are fitted analyzing and providing the Monitoring Data Fitting Analysis mould of analytical data Block, the analytical data be given by Monitoring Data Fitting Analysis module carry out the long-term trend prediction of disaster, disaster alarm in mid-term, Face the prediction and warning model module of calamity disaster alert, it is characterised in that described mounting shell is provided with for inputting control command Function keyboard, the display screen shown for picture and text and data, for providing the battery area of power supply for equipment, for starting device Operating switch and for showing the display lamp of early warning signal, described Monitoring Data Fitting Analysis module, prediction and warning model module By display screen display data analysis and prediction and warning result, and provide early warning signal by display lamp, also include built-in storage Card, USB interface, serial line interface and communication interface, described embedded memory card is for natural disaster basic data and monitoring and warning Process and result thereof store, and described USB interface, serial line interface and communication interface are for transmitting the number in embedded memory card According to.

Preferably, described collection in worksite module includes for taking pictures, imaging, and forms comprehensive image, attribute data loading High-definition camera.

Preferably, described mounting shell is additionally provided with it is connected with Monitoring Data Fitting Analysis module, prediction and warning model module The speech input interface connect and voice output interface.

Preferably, described speckle registration apparatus includes: pretreatment module, feature detection module, feature description module, feature Matching module and spatial alternation module；

(1) pretreatment module, for reference picture and image subject to registration are converted into gray level image, defines conversion formula For:

I (x, y)=k (G (x, y)+R (x, y)+B (x, y))+2k (G (and x, y)+R (x, y))+3k (x, y)

Wherein, (x, y) at coordinate, (x, y) grey scale pixel value at place, (x, y), (x, y), (x y) distinguishes B R G representative image I (x, y) the red, green, blue intensity level at place, k is the weight factor set, and the span of k is [0.08,0.12] to represent coordinate；

(2) feature detection module, including local features detection sub-module and Patch properties detection sub-module, described office District of portion

Characteristic of field detection sub-module for detecting the local of pretreated two images by Mexican hat wavelet function Provincial characteristics, described Patch properties detection sub-module is for using the local extremum of approximation Hessian matrix at custom dimensions Space is detected the Patch properties in pretreated two images, the locus of output Patch properties and the feature chi at place Degree；

The description form of described approximation Hessian matrix is:

$H(x,y,\mathrm{\σ})=\left[\begin{array}{cc}{E}_{xx}(x,y,\mathrm{\σ})& E_{xy}(x,y,\mathrm{\σ})\\ E_{xy}(x,y,\mathrm{\σ})& E_{yy}(x,y,\mathrm{\σ})\end{array}\right]$

In formula, σ is the standard deviation of Gaussian function, i.e. scale factor；E_xx(x,y,σ)、E_xy(x,y,σ)、E_yy(x, y, σ) point Wei Gauss second-order differentialApproximation template after discretization and cutting point (x, y) place with The convolution of image；Set E_xx、E_xyAnd E_yyRepresent the convolution results of first, second, and third 9 × 9 approximation template and image, E_xx、 E_xyAnd E_yyComputing formula be defined as follows:

$E_{xx}=\frac{1}{P}(A_1-2A_2+A_3)$

$E_{xy}=\frac{1}{Q}(A_1-A_2-A_3)$

$E_{yy}=\frac{1}{P}(A_1-2A_2+A_3)$

Wherein, take the first approximation template 3 × 3 regions from left to right as the first marked region, take in the second approximation template Region, portion 2 × 2, as the second marked region, takes the 3rd approximation template 3 × 3 regions from top to bottom as the 3rd marked region, A₁、 A₂And A₃Be respectively first, second, and third marked region cover lower image pixel gray level and, P, Q are marked region area, divide Deng Yu 9 and 4；

The characteristic point receptance function of described approximation Hessian matrix is:

DET (H)=E_xxE_yy-(0.9E_xy)²

The description form of described Mexican hat wavelet function is:

$M(x,y,\mathrm{\σ})=(1-q)(2-\frac{x^2+y^2}{{\mathrm{\σ}}^2})\exp (-\frac{x^2+y^2}{2{\mathrm{\σ}}^2})$

Wherein, q is the running parameter constructing described custom dimensions space, and the relation between q and σ is σ=2^-q；

Preferably,

(3) feature description module, the Patch properties detected is described also by it by setting up weighted intensity description Formed describe vector, set up weighted intensity describe the period of the day from 11 p.m. to 1 a.m by centered by Patch properties, be perpendicular to Patch properties principal direction and Size is that the central area of l × l is divided into multiple sub-block, described weighted intensity to describe son to be:

WD={NP⁺,NP^-}

Herein

P⁺=Σ { f (D_i)×d(m_i)|d(m_i)=I (m_i)-I (m) ＞ 0, i=1,2 ... l²}

P^-=Σ { f (D_i)×d(m_i)|d(m_i)=I (m_i)-I (m) ＜ 0, i=1,2 ... l²}

In formula, WD represents that weighted intensity describes son, P⁺Represent positive gray scale difference value histogram, P^-Represent negative gray scale difference value Nogata Figure, I (m_i) and I (m) be respectively the gray average of each sub-block and the ash of whole central area using bilinear interpolation to ask for Degree average, D_iFor the distance of each sub-block Yu center, wherein, i=1,2 ... l², f (D_i) represent weighting function, NP⁺、NP^-Expression is returned The one positive and negative gray scale difference rectangular histogram changed；

(4) characteristic matching module, for mating pretreated reference picture and image subject to registration, including successively The regional area matched sub-block, the region that connect divide submodule, feature packet submodule and Patch properties matched sub-block, institute State regional area matched sub-block for the local features of pretreated reference picture and image subject to registration being carried out Joining, described region divides submodule for carrying out pretreated reference picture and image subject to registration according to local features Image-region after image-region divides and will divide is converted into standard round region, if the image-region quantity divided is N, N's Span is [200,1000], and described feature packet submodule is for the model divided according to image-region by described Patch properties Enclosing and be grouped, described Patch properties matched sub-block is for carrying out the description vector representing Patch properties in each group Join；

(5) spatial alternation module, for being mapped to the coordinate of reference picture by image subject to registration by geometric transformation model In system, completing image registration, the parameter of described geometric transformation model uses RANSAC algorithm to estimate.

Wherein, described custom dimensions space is divided into many groups, and often group comprises the filter template of three different scales；Described In first group of custom dimensions space, template size corresponding to smallest dimension is 9 × 9, and marked region increment is set to 4, template Increment is 12, and second is followed successively by 21 × 21 and 33 × 33 with the 3rd template size that template is corresponding；Except custom dimensions is empty Between first group outside other groups in, first template often organized is identical with the second of previous group template size, and template increase Amount is 4 times of previous group.

Wherein, the center of gravity that the center of circle is local features in described standard round region, the radius in standard round region is local Point in edges of regions is to the average of described centroidal distance.

The image pre-processing module that the present embodiment is arranged considers visual custom and the human eye perceptibility to different color Non-linear relation with colouring intensity, it is possible to describe image the most accurately；The Patch properties detection sub-module arranged, it is possible to etc. Speed builds the metric space of image, and it can be avoided that aliasing occurs；The feature description module arranged is by setting up Weighted Grey Degree describes son and is described the Patch properties detected, it is possible to utilizes the local message in feature neighborhood to build more fully and retouches State vector；The characteristic matching module arranged first carries out the Patch properties point division that regional area coupling carries out in group again, improves The speed of images match, and the region being provided with divides the image-region after submodule will divide and is converted into standard round district Territory, reduces complexity and the impact on packet registration of the regional area extraction precision that the packet of Patch properties point judges.This reality Executing example weight factor k value is 0.11, and the image-region quantity N value of division is 800, and images match precision improves relatively 1.5%, matching speed improves 1.5%.

Embodiment 5

See Fig. 1, Fig. 2, a kind of monitoring early-warning device for natural disaster of the present embodiment, including monitoring early-warning device and Coupled speckle registration apparatus, described monitoring early-warning device includes: mounting shell, be provided with in described mounting shell for periodically, The Monitoring Data of timing acquiring natural disaster field conditions and the collection in worksite module of graphic record, for storing what expert provided Data and the expert system module of photo information, for the Monitoring Data that collection in worksite module is collected and graphic record and specially Data and photo information that family's system module provides are fitted analyzing and providing the Monitoring Data Fitting Analysis mould of analytical data Block, the analytical data be given by Monitoring Data Fitting Analysis module carry out the long-term trend prediction of disaster, disaster alarm in mid-term, Face the prediction and warning model module of calamity disaster alert, it is characterised in that described mounting shell is provided with for inputting control command Function keyboard, the display screen shown for picture and text and data, for providing the battery area of power supply for equipment, for starting device Operating switch and for showing the display lamp of early warning signal, described Monitoring Data Fitting Analysis module, prediction and warning model module By display screen display data analysis and prediction and warning result, and provide early warning signal by display lamp, also include built-in storage Card, USB interface, serial line interface and communication interface, described embedded memory card is for natural disaster basic data and monitoring and warning Process and result thereof store, and described USB interface, serial line interface and communication interface are for transmitting the number in embedded memory card According to.

Preferably, described collection in worksite module includes for taking pictures, imaging, and forms comprehensive image, attribute data loading High-definition camera.

Preferably, described mounting shell is additionally provided with it is connected with Monitoring Data Fitting Analysis module, prediction and warning model module The speech input interface connect and voice output interface.

Preferably, described speckle registration apparatus includes: pretreatment module, feature detection module, feature description module, feature Matching module and spatial alternation module；

(1) pretreatment module, for reference picture and image subject to registration are converted into gray level image, defines conversion formula For:

I (x, y)=k (G (x, y)+R (x, y)+B (x, y))+2k (G (and x, y)+R (x, y))+3k (x, y)

Wherein, (x, y) at coordinate, (x, y) grey scale pixel value at place, (x, y), (x, y), (x y) distinguishes B R G representative image I (x, y) the red, green, blue intensity level at place, k is the weight factor set, and the span of k is [0.08,0.12] to represent coordinate；

(2) feature detection module, including local features detection sub-module and Patch properties detection sub-module, described office District of portion

Characteristic of field detection sub-module for detecting the local of pretreated two images by Mexican hat wavelet function Provincial characteristics, described Patch properties detection sub-module is for using the local extremum of approximation Hessian matrix at custom dimensions Space is detected the Patch properties in pretreated two images, the locus of output Patch properties and the feature chi at place Degree；

The description form of described approximation Hessian matrix is:

$H(x,y,\mathrm{\σ})=\left[\begin{array}{cc}{E}_{xx}(x,y,\mathrm{\σ})& E_{xy}(x,y,\mathrm{\σ})\\ E_{xy}(x,y,\mathrm{\σ})& E_{yy}(x,y,\mathrm{\σ})\end{array}\right]$

In formula, σ is the standard deviation of Gaussian function, i.e. scale factor；E_xx(x,y,σ)、E_xy(x,y,σ)、E_yy(x, y, σ) point Wei Gauss second-order differentialApproximation template after discretization and cutting point (x, y) place with The convolution of image；Set E_xx、E_xyAnd E_yyRepresent the convolution results of first, second, and third 9 × 9 approximation template and image, E_xx、 E_xyAnd E_yyComputing formula be defined as follows:

$E_{xx}=\frac{1}{P}(A_1-2A_2+A_3)$

$E_{xy}=\frac{1}{Q}(A_1-A_2-A_3)$

$E_{yy}=\frac{1}{P}(A_1-2A_2+A_3)$

Wherein, take the first approximation template 3 × 3 regions from left to right as the first marked region, take in the second approximation template Region, portion 2 × 2, as the second marked region, takes the 3rd approximation template 3 × 3 regions from top to bottom as the 3rd marked region, A₁、 A₂And A₃Be respectively first, second, and third marked region cover lower image pixel gray level and, P, Q are marked region area, divide Deng Yu 9 and 4；

The characteristic point receptance function of described approximation Hessian matrix is:

DET (H)=E_xxE_yy-(0.9E_xy)²

The description form of described Mexican hat wavelet function is:

$M(x,y,\mathrm{\σ})=(1-q)(2-\frac{x^2+y^2}{{\mathrm{\σ}}^2})\exp (-\frac{x^2+y^2}{2{\mathrm{\σ}}^2})$

Wherein, q is the running parameter constructing described custom dimensions space, and the relation between q and σ is σ=2^-q；

Preferably,

(3) feature description module, the Patch properties detected is described also by it by setting up weighted intensity description Formed describe vector, set up weighted intensity describe the period of the day from 11 p.m. to 1 a.m by centered by Patch properties, be perpendicular to Patch properties principal direction and Size is that the central area of l × l is divided into multiple sub-block, described weighted intensity to describe son to be:

WD={NP⁺,NP^-}

Herein

P⁺=Σ { f (D_i)×d(m_i)|d(m_i)=I (m_i)-I (m) ＞ 0, i=1,2 ... l²}

P^-=Σ { f (D_i)×d(m_i)|d(m_i)=I (m_i)-I (m) ＜ 0, i=1,2 ... l²}

In formula, WD represents that weighted intensity describes son, P⁺Represent positive gray scale difference value histogram, P^-Represent negative gray scale difference value Nogata Figure, I (m_i) and I (m) be respectively the gray average of each sub-block and the ash of whole central area using bilinear interpolation to ask for Degree average, D_iFor the distance of each sub-block Yu center, wherein, i=1,2 ... l², f (D_i) represent weighting function, NP⁺、NP^-Expression is returned The one positive and negative gray scale difference rectangular histogram changed；

(4) characteristic matching module, for mating pretreated reference picture and image subject to registration, including successively The regional area matched sub-block, the region that connect divide submodule, feature packet submodule and Patch properties matched sub-block, institute State regional area matched sub-block for the local features of pretreated reference picture and image subject to registration being carried out Joining, described region divides submodule for carrying out pretreated reference picture and image subject to registration according to local features Image-region after image-region divides and will divide is converted into standard round region, if the image-region quantity divided is N, N's Span is [200,1000], and described feature packet submodule is for the model divided according to image-region by described Patch properties Enclosing and be grouped, described Patch properties matched sub-block is for carrying out the description vector representing Patch properties in each group Join；

(5) spatial alternation module, for being mapped to the coordinate of reference picture by image subject to registration by geometric transformation model In system, completing image registration, the parameter of described geometric transformation model uses RANSAC algorithm to estimate.

Wherein, described custom dimensions space is divided into many groups, and often group comprises the filter template of three different scales；Described In first group of custom dimensions space, template size corresponding to smallest dimension is 9 × 9, and marked region increment is set to 4, template Increment is 12, and second is followed successively by 21 × 21 and 33 × 33 with the 3rd template size that template is corresponding；Except custom dimensions is empty Between first group outside other groups in, first template often organized is identical with the second of previous group template size, and template increase Amount is 4 times of previous group.

Wherein, the center of gravity that the center of circle is local features in described standard round region, the radius in standard round region is local Point in edges of regions is to the average of described centroidal distance.

The image pre-processing module that the present embodiment is arranged considers visual custom and the human eye perceptibility to different color Non-linear relation with colouring intensity, it is possible to describe image the most accurately；The Patch properties detection sub-module arranged, it is possible to etc. Speed builds the metric space of image, and it can be avoided that aliasing occurs；The feature description module arranged is by setting up Weighted Grey Degree describes son and is described the Patch properties detected, it is possible to utilizes the local message in feature neighborhood to build more fully and retouches State vector；The characteristic matching module arranged first carries out the Patch properties point division that regional area coupling carries out in group again, improves The speed of images match, and the region being provided with divides the image-region after submodule will divide and is converted into standard round district Territory, reduces complexity and the impact on packet registration of the regional area extraction precision that the packet of Patch properties point judges.This reality Executing example weight factor k value is 0.12, and the image-region quantity N value of division is 1000, and images match precision improves relatively 1.5%, matching speed improves 1.2%.

Last it should be noted that, above example is only in order to illustrate technical scheme, rather than the present invention is protected Protecting the restriction of scope, although having made to explain to the present invention with reference to preferred embodiment, those of ordinary skill in the art should Work as understanding, technical scheme can be modified or equivalent, without deviating from the reality of technical solution of the present invention Matter and scope.

Claims (8)

1., for a monitoring early-warning device for natural disaster, registrate dress including monitoring early-warning device and coupled speckle Putting, described monitoring early-warning device includes: mounting shell, is provided with for regular, timing acquiring natural disaster scene in described mounting shell The Monitoring Data of situation and the collection in worksite module of graphic record, for storing data and the expert of photo information that expert provides System module, the data provided for Monitoring Data that collection in worksite module is collected and graphic record and expert system module It is fitted analyzing and providing the Monitoring Data Fitting Analysis module of analytical data, divided by Monitoring Data matching with photo information The analysis analytical data that is given of module carry out the long-term trend prediction of disaster, disaster alarm in mid-term, face calamity disaster alert prediction pre- Alert model module, it is characterised in that described mounting shell is provided with the function keyboard for inputting control command, for picture and text and number According to the display screen of display, for providing the battery area of power supply for equipment, showing pre-for the operating switch and being used for of starting device The display lamp of alert signal, described Monitoring Data Fitting Analysis module, prediction and warning model module are divided by display screen video data Analysis and prediction and warning result, and provide early warning signal by display lamp, also include embedded memory card, USB interface, serial line interface and Communication interface, described embedded memory card for natural disaster basic data and monitoring and warning process and result thereof are stored, Described USB interface, serial line interface and communication interface are for transmitting the data in embedded memory card.

A kind of monitoring early-warning device for natural disaster the most according to claim 1, is characterized in that, described collection in worksite Module includes for taking pictures, imaging, and forms comprehensive image, the high-definition camera of attribute data loading.

A kind of monitoring early-warning device for natural disaster the most according to claim 2, is characterized in that, on described mounting shell It is additionally provided with the speech input interface being connected with Monitoring Data Fitting Analysis module, prediction and warning model module and voice output connects Mouthful.

A kind of monitoring early-warning device for natural disaster the most according to claim 3, is characterized in that, described speckle registrates Device includes: pretreatment module, feature detection module, feature description module, characteristic matching module and spatial alternation module:

(1) pretreatment module, for reference picture and image subject to registration are converted into gray level image, definition conversion formula is:

I (x, y)=k (G (x, y)+R (x, y)+B (x, y))+2k (G (and x, y)+R (x, y))+3k (x, y)

Wherein, (x, y) at coordinate, (x, y) grey scale pixel value at place, (x, y), (x, y), (x y) represents B R G representative image I respectively (x, y) the red, green, blue intensity level at place, k is the weight factor set to coordinate, and the span of k is [0.08,0.12]；

(2) feature detection module, including local features detection sub-module and Patch properties detection sub-module, described office

Portion's provincial characteristics detection sub-module for detecting the local of pretreated two images by Mexican hat wavelet function Provincial characteristics, described Patch properties detection sub-module is for using the local extremum of approximation Hessian matrix at custom dimensions Space is detected the Patch properties in pretreated two images, the locus of output Patch properties and the feature chi at place Degree；

The description form of described approximation Hessian matrix is:

In formula, σ is the standard deviation of Gaussian function, i.e. scale factor；E_xx(x, y, σ), E_xy(x, y, σ), E_yy(x, y, σ) is respectively Gauss second-order differentialApproximation template after discretization and cutting is at point (x, y) place and image Convolution；Set E_xx、E_xyAnd E_yyRepresent the convolution results of first, second, and third 9 × 9 approximation template and image, E_xx、E_xyWith E_yyComputing formula be defined as follows:

Wherein, take the first approximation template 3 × 3 regions from left to right as the first marked region, take in the middle part of the second approximation template 2 × 2 regions, as the second marked region, take the 3rd approximation template 3 × 3 regions from top to bottom as the 3rd marked region, A₁、A₂And A₃ Be respectively first, second, and third marked region cover lower image pixel gray level with, P, Q are marked region area, respectively etc. In 9 and 4；

The characteristic point receptance function of described approximation Hessian matrix is:

DET (H)=E_xxE_yy-(0.9E_xy)²

The description form of described Mexican hat wavelet function is:

Wherein, q is the running parameter constructing described custom dimensions space, and the relation between q and σ is σ=2^-q。

A kind of monitoring early-warning device for natural disaster the most according to claim 4, is characterized in that,

(3) feature description module, the Patch properties detected is described and is formed by setting up weighted intensity description by it Vector is described, set up weighted intensity describe the period of the day from 11 p.m. to 1 a.m by centered by Patch properties, be perpendicular to principal direction and the size of Patch properties Be 1 × 1 central area be divided into multiple sub-block, described weighted intensity to describe son to be:

WD={NP⁺, NP^-}

Herein

P⁺=Σ { f (D_i)×d(m_i)|d(m_i)=I (m_i)-I (m) ＞ 0, i=1,2 ... l²}

P^-=Σ { f (D_i)×d(m_i)|d(m_i)=I (m_i)-I (m) ＜ 0, i=1,2 ... l²}

In formula, WD represents that weighted intensity describes son, P⁺Represent positive gray scale difference value histogram, P^-Represent negative gray scale difference value histogram, I (m_i) and I (m) to be respectively the gray scale of the gray average of each sub-block and the whole central area using bilinear interpolation to ask for equal Value, D_iFor the distance of each sub-block Yu center, wherein, i=1,2 ... l², f (D_i) represent weighting function, NP⁺、NP^-Represent normalization Positive and negative gray scale difference rectangular histogram；

(4) characteristic matching module, for mating pretreated reference picture and image subject to registration, including being sequentially connected with Regional area matched sub-block, region divide submodule, feature packet submodule and Patch properties matched sub-block, described office Portion's Region Matching submodule is used for mating the local features of pretreated reference picture and image subject to registration, institute State region and divide submodule for pretreated reference picture and image subject to registration being carried out image according to local features Image-region after region divides and will divide is converted into standard round region, if the value that image-region quantity is N, N divided Scope is [200,1000], and described feature packet submodule is for entering the scope that described Patch properties divides according to image-region Row packet, described Patch properties matched sub-block is for mating the description vector representing Patch properties in each group；

(5) spatial alternation module, for image subject to registration is mapped in the coordinate system of reference picture by geometric transformation model, Completing image registration, the parameter of described geometric transformation model uses RANSAC algorithm to estimate.

A kind of monitoring early-warning device for natural disaster the most according to claim 5, is characterized in that, described self-defined chi Degree space is divided into many groups, and often group comprises the filter template of three different scales.

A kind of monitoring early-warning device for natural disaster the most according to claim 6, is characterized in that, described self-defined chi In degree first group of space, template size corresponding to smallest dimension is 9 × 9, and marked region increment is set to 4, and template increment is 12, Second is followed successively by 21 × 21 and 33 × 33 with the 3rd template size that template is corresponding；Except custom dimensions space first group In other outer groups, first template often organized is identical with the second of previous group template size, and template increment is previous group 4 times.

A kind of monitoring early-warning device for natural disaster the most according to claim 7, is characterized in that, described standard round district The center of circle in territory is the center of gravity of local features, and the radius in standard round region is that the point on part-area edge is to described distance of centre of gravity From average.