CN103226695A - Multi-scale intelligent Chinese red pine identification model based on selective vision attention mechanism - Google Patents

Abstract

The invention relates to a multi-scale intelligent Chinese red pine identification model based on a selective vision attention mechanism. Identification steps are as follows: on the basis of distinctiveness region partition, abstracting multi-dimensional features interested by a vision attention mechanism; further capturing morphological structure features of Chinese red pine under various conditions on the basis of morphological analysis; carrying out configuration modeling; and constructing the multi-scale intelligent Chinese red pine identification model based on the selective vision attention mechanism, thereby achieving accurate identification of the Chinese red pine and facilitating comprehensive monitoring, prevention and control of forest resource.

Description

Based on the multiple dimensioned Intelligent Recognition model of the masson pine of selective visual attention mechanism

One, technical field

The invention belongs to forestry remote sensing image intelligent process field, the multiple dimensioned Intelligent Recognition model of particularly a kind of masson pine based on selective visual attention mechanism is used for the Intelligent Recognition of masson pine.

Two, technical background

Raising along with remote sensing image resolution, morphological features such as the space structure of ground object target, top layer texture obtain more clearly expressing, traditional sorting technique applicability based on pixels statistics value and single yardstick is relatively poor, its main cause is shape and the textural characteristics that these traditional sorting techniques have been ignored atural object, and these information are exactly the principal character foundation that high-resolution remote sensing image is distinguished atural object just.Adopting high spatial resolution image to extract morphosis information, and carry out seeds Classification and Identification aspect in conjunction with spectral information, Chinese scholars has been made trial and has been reached good effect, but mostly need artificial too much participation, the parameter setting is too complicated, also specific data type or the survey regions of being confined to of the model of cognition of structure more.

In recent years, along with going deep into of computer vision research, people recognize the importance of selective visual attention mechanism gradually.Human can be promptly in the face of complex scene the time with oneself attention focusing on some significant targets, thereby these targets are carried out priority processing, have the mechanism of a Visual Selective Attention here.The mechanism of this most optimum distribution of resources makes the human brain visual cortex can handle the visual information of taking in well under limited neural resource.Provide reliable solution based on the target extraction method of selective visual attention mechanism for the extraction of forest tree crown.

Traditional vision significance computation model operand based on spatial domain is bigger, the parameter setting is too complicated, in order to overcome this defective, this project adopts the vision significance computing method based on frequency domain, and combining form credit analysis method has been invented a kind of multiple dimensioned Intelligent Recognition model based on selective visual attention mechanism to improving based on the remaining salient region computing method of spectrum.

Three, summary of the invention

Technology of the present invention is dealt with problems and is: a kind of multiple dimensioned Intelligent Recognition model based on selective visual attention mechanism is provided, purpose is to solve that the model automatization degree that exists in the masson pine Intelligent Recognition is low, versatility difference and the low problem of accuracy of identification, optimize the optimum factor system of selection that salient region is cut apart, satisfy the correct extraction of the masson pine conspicuousness feature under the various weather conditions; Optimize the selection parameter that appearance model is set up, reduce the susceptibility of algorithm, the get down from horse morphological feature of tail pine of different weather conditions and different plants size condition is analyzed, make up modeling method with certain versatility to the individual plant masson pine.

Technical scheme of the present invention is: based on the multiple dimensioned Intelligent Recognition model of the masson pine of selective visual attention mechanism, implementation step is as follows:

(1) chooses the multi-source remote sensing image, carry out autoregistration and fusion, figure as a result after obtaining merging, on the basis of merging figure, the vision attention mechanism of imitation human eye is in conjunction with the image contextual information, calculate the conspicuousness value of each pixel based on the remaining method of spectrum, obtain the degree size that each pixel causes that human eye vision is noted, computational resource is focused on the zone that causes that human eye vision is noted, cut apart and obtain salient region;

(2) the morphosis parameter is to understand the key factor of masson pine, the size of the imaging parameters when obtaining, image resolution and target in conjunction with image, utilize contour structure to the masson pine modeling under the different condition, in salient region, extract contour structure, carry out the form modeling;

(3) make up multiple dimensioned Intelligent Recognition model according to the result of form modeling, extract the interested multidimensional feature of vision noticing mechanism, realize the accurate identification of masson pine based on improved vision noticing mechanism.

The present invention's advantage compared with prior art is:

(1) this model can identify the masson pine under the multiple conditions such as different plants size, strong robustness, and versatility is good;

(2) this model can be realized the Intelligent Recognition of masson pine under the different scale condition;

(3) this model can be realized the automatic identification of masson pine.

Four, description of drawings

Fig. 1 is technology path figure of the present invention.

Five, embodiment

As shown in Figure 1, the embodiment of the inventive method comprises four steps, and is specific as follows:

(1) cuts apart the salient region that comprises masson pine based on the spectrum remnants method of frequency domain;

A) the species growth characteristics of test site are analyzed, chosen optics and radar high-resolution remote sensing image, carry out registration, merge based on the second-order stationary Wavelet Transformation Algorithm then;

B) according to scope of experiment to the cutting of carrying out after merging, obtain the high-resolution remote sensing image in the test block;

C) vegetation index of calculating high-resolution remote sensing image obtains vegetation index figure;

D) the logarithmic spectrum L (f) of calculating vegetation index figure;

E) calculate the general type A (f) of logarithmic spectrum based on formula (1);

A(f)=h _n(f)×L(f) （1）

Wherein $h_n\left(f\right)=\frac{1}{n^2}\left(\begin{array}{cccc}1& 1& \·\·\·& 1\\ 1& 1& \·\·\·& 1\\ \·\·\·& \·\·\·& \·\·\·& \·\·\·\\ 1& 1& \·\·\·& 1\end{array}\right),$ At this n=3.

F) calculate the remaining R (f) of spectrum of image based on formula (2);

R(f)=L(f)-A(f) （2）

Wherein, the gross morphology of A (f) expression logarithmic spectrum, as the prior imformation input, R (f) is the statistical special area of input image, is defined as the spectrum remnants of image.

G) will compose residual image and carry out inverse Fourier transform, obtain the conspicuousness figure of image;

H) conspicuousness figure and high-resolution remote sensing image are carried out mask process, obtain comprising the salient region of masson pine.

The picture of (2) collect under the various weather conditions, various plant size condition being got down from horse the tail pine carries out the form modeling to masson pine;

A) masson pine under the various conditions is carried out the individual morphology analysis;

B) in conjunction with the imaging parameters of high-resolution optical remote sensing image, the spectral signature curve of masson pine is analyzed;

C) make up general appearance model;

(3) on the basis that salient region is cut apart, masson pine is carried out target prediction according to the appearance model that makes up;

(4) based on the spectrum statistical property of masson pine under the various conditions, the result of target prediction is handled, reject false-alarm, obtain final masson pine recognition result.

Claims (1)

1. multiple dimensioned Intelligent Recognition model of the masson pine based on selective visual attention mechanism, it is characterized in that, based on the interested multidimensional feature of salient region segmented extraction vision noticing mechanism, further catch the morphosis feature of masson pine under the various conditions based on the morphological analysis method, carry out the form modeling, structure is based on the multiple dimensioned Intelligent Recognition model of selective visual attention mechanism, realize the accurate identification of masson pine, made things convenient for the comprehensive monitoring and the prevention and control of the forest reserves, main experimental program comprises following three links:

(1) chooses the multi-source remote sensing image, carry out autoregistration and fusion, figure as a result after obtaining merging, on the basis of merging figure, the vision attention mechanism of imitation human eye is in conjunction with the image contextual information, calculate the conspicuousness value of each pixel based on the remaining method of spectrum, obtain the degree size that each pixel causes that human eye vision is noted, computational resource is focused on the zone that causes that human eye vision is noted, cut apart and obtain salient region;

(2) the morphosis parameter is to understand the key factor of masson pine, the size of the imaging parameters when obtaining, image resolution and target in conjunction with image, utilize contour structure to the masson pine modeling under the different condition, in salient region, extract contour structure, carry out the form modeling;

(3) make up multiple dimensioned Intelligent Recognition model according to the result of form modeling, extract the interested multidimensional feature of vision noticing mechanism, realize the accurate identification of masson pine based on improved vision noticing mechanism.

Images