CN108564092A - Sunflower disease recognition method based on SIFT feature extraction algorithm - Google Patents
Sunflower disease recognition method based on SIFT feature extraction algorithm Download PDFInfo
- Publication number
- CN108564092A CN108564092A CN201810325111.8A CN201810325111A CN108564092A CN 108564092 A CN108564092 A CN 108564092A CN 201810325111 A CN201810325111 A CN 201810325111A CN 108564092 A CN108564092 A CN 108564092A
- Authority
- CN
- China
- Prior art keywords
- image
- disease
- point
- scale
- sunflower
- Prior art date
- Legal status (The legal status 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 status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
- G06V10/462—Salient features, e.g. scale invariant feature transforms [SIFT]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/74—Image or video pattern matching; Proximity measures in feature spaces
- G06V10/75—Organisation of the matching processes, e.g. simultaneous or sequential comparisons of image or video features; Coarse-fine approaches, e.g. multi-scale approaches; using context analysis; Selection of dictionaries
- G06V10/758—Involving statistics of pixels or of feature values, e.g. histogram matching
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20016—Hierarchical, coarse-to-fine, multiscale or multiresolution image processing; Pyramid transform
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20048—Transform domain processing
- G06T2207/20056—Discrete and fast Fourier transform, [DFT, FFT]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30181—Earth observation
- G06T2207/30188—Vegetation; Agriculture
Abstract
The sunflower disease recognition method based on SIFT feature extraction algorithm that the present invention relates to a kind of, including：Sunflower leaf portion disease geo-radar image to be identified is obtained, image enhancement, denoising, smooth, sharpening pretreatment are carried out to sunflower leaf portion to be identified disease geo-radar image；Sunflower leaf portion disease geo-radar image to be identified is subjected to SIFT feature vector with all kinds of disease reference pictures in sample database respectively and carries out matching degree analysis, if the quantity that the SIFT feature vector of sunflower leaf portion disease geo-radar image to be identified and a kind of disease reference picture matches is greater than or equal to predetermined threshold, judge that the corresponding sunflower leaf of the sunflower leaf portion disease geo-radar image to be identified is such disease sunflower leaf；If the quantity that the SIFT feature vector of all kinds of disease reference pictures in sunflower leaf portion disease geo-radar image to be identified and sample database matches is both less than predetermined threshold, judge that the corresponding sunflower leaf of the sunflower leaf portion disease geo-radar image to be identified is not any kind disease in sample database.
Description
Technical field
The present invention relates to agricultural pests to identify field, more particularly to a kind of sunflower based on SIFT feature extraction algorithm
Disease recognition method.
Background technology
It is identified by naked eyes it is well known that traditional disease screening method is mainly plant protection personnel, and combines plant disease
The form of pathogen judged that this method diagnosis efficiency is low, it is difficult in time, accurately judge Damage Types.Accurate agriculture
Industry " provides new thinking for agriculture cultivator, by quickly and effectively identifying plant disease with information technology, relative to tradition
Recognition methods, recognition speed is fast, accuracy rate is high, also have timeliness.It is right by taking the crop diseases such as apple, cucumber and capsicum as an example
The image of plant leaf diseases is analyzed, and using level set and C-V scale-model investigations, experience are improved for the colouring information of image
Demonstrate,prove the discrimination that this method substantially increases disease.
Invention content
In order to solve the above technical problems, the object of the present invention is to provide a kind of sunflowers based on SIFT feature extraction algorithm
Disease recognition method.
The present invention is based on the sunflower disease recognition methods of SIFT feature extraction algorithm, including：
Sunflower leaf portion disease geo-radar image to be identified is obtained, image increasing is carried out to the sunflower leaf portion disease geo-radar image to be identified
By force, denoising, smooth, sharpening pretreatment；
The SIFT feature vector that the sunflower leaf portion disease geo-radar image to be identified is extracted based on SIFT feature extraction algorithm, will
Sunflower leaf portion disease geo-radar image to be identified carries out SIFT feature vector progress with all kinds of disease reference pictures in sample database respectively
Matching degree is analyzed,
If the number that the SIFT feature vector of sunflower leaf portion disease geo-radar image to be identified and a kind of disease reference picture matches
Amount is greater than or equal to predetermined threshold, then judges that the corresponding sunflower leaf of the sunflower leaf portion disease geo-radar image to be identified is such disease
Sunflower leaf；
If the SIFT feature vector of sunflower leaf portion disease geo-radar image to be identified and all kinds of disease reference pictures in sample database
The quantity to match is both less than predetermined threshold, then judges that the corresponding sunflower leaf of the sunflower leaf image to be identified is not sample database
In any kind disease.
Further, the pretreatment includes：The equalization of histogram, homomorphic filtering, wherein
The equalization of the histogram specifically includes：
To its histogram of the pending image statistics of input, find out
In formula, L is gray level；pr(rk) it is the probability for taking kth grade gray value；nkIt is to occur kth grade gray scale in the picture
Number；N is sum of all pixels in image；
It is converted with cumulative distribution function according to the histogram come out, finds out the new gray scale after transformation, cumulative distribution
Function is as follows：
Old gray scale is replaced with new gray scale, finds out PS(s), gray value is equal or be approximately merged together；
The homomorphic filtering specifically includes：
Logarithm, purpose is taken exactly to convert multiplying to add operation original image f (x, y)：
Z (x, y)=ln f (x, y)=lni (x, y)+lnr (x, y)
Fourier transformation is done to logarithmic function, image is exactly transformed into frequency domain by purpose：
F (z (x, y))=F [lni (x, y)]+F [lnr (x, y)] i.e. Z=I+R
Transmission function appropriate, the variation range of press irradiation component i (x, y) is selected to weaken I (u, v), enhancing r (x, y)
Contrast, promoted R (u, v), enhance high fdrequency component, that is, determine a H (u, v)；
Selection one homomorphic filter function H (u, v) come handle the logarithm of original image f (x, y) Fourier transformation Z (u,
V), it obtains
S (u, v)=Z (u, v) H (u, v)=I (u, v) H (u, v)+R (u, v) H (u, v),
Inversion changes to spatial domain and obtains s (x, y)=F-1(S(u,v))；
Final result g (x, y)=exp (s (x, y)) i.e. is obtained to fetching number, is equivalent to high-pass filtering.
Further, the extraction step of SIFT feature vector includes：Scale space extreme value is extracted；Positioning feature point；Feature
Direction assignment；Extract feature point description；
The scale space L (x, y, σ) of one width two dimensional image indicate can by a change of scale gaussian kernel function G (x, y,
It σ) is obtained with original image I (x, y) convolution, such as following formula：
L (x, y, σ)=G (x, y, σ) * I (x, y)
Wherein, (x, y) indicates that the pixel coordinate of picture point, I (x, y) indicate that image data, σ are known as the scale space factor, *
Indicate convolution algorithm；
Wherein, scale space factor sigma is the variance of Gauss normal distribution, shows the degree that image is smoothed, and σ values are smaller
Indicate that the degree that image is smoothed is smaller, corresponding scale is also smaller；
Gaussian difference scale space DoG, DoG operator is the difference of two different Gaussian functions of scale, as gray scale
The enhancing operator of image can be effectively used to the edge of enhancing image, i.e., constantly be obscured to original-gray image, retain
Marginal portion on original image can also remove picture noise, the image with a large amount of random noises is particularly suitable for, if k
For the scale factor between two adjacent scales, then DoG operator definitions are as follows：
D (x, y, σ)=(G (x, y, σ)-G (x, y, σ)) * I (x, y)
=L (x, y, k σ)-L (x, y, σ)
DoG operators calculate simply, are the approximations of the LoG operators of dimension normalization；The ruler of first layer in DoG pyramids
The degree factor is consistent with the first layer in gaussian pyramid；
DoG spatial extremas detect, and need the DoG values for detecting 26 pixels, therefrom select DoG values maximum or minimum value is made
For candidate feature point, its position and corresponding scale are write down；This 26 pixels are chosen such that the spaces DoG any layer (most bottom
Layer and top except) arbitrary pixel same layer in adjacent 8 pixels and its last layer and the 9 of next layer it is adjacent
Pixel；
After obtaining candidate feature point, precise positioning feature point position is next wanted, namely is rejected in candidate feature point low
Contrast point and unstable skirt response point, the value of the pixel of characteristic point must have apparent difference with the point of surrounding；It is special
Sign point cannot be marginal point；
It is each key point assigned direction number using the gradient direction distribution characteristic of key point neighborhood territory pixel, operator is made to have
Rotational invariance；For each Gaussian image；The mould m (x, y) of the gradient of each point L (x, y) passes through as follows with direction θ (x, y)
Formula is calculated：
In above formula, scale used in L (x, y) is scale where key point；
It is sampled in neighborhood window centered on key point, the gradient direction of statistics with histogram neighborhood territory pixel is used in combination；Gradient
The range of histogram is 360 degree, wherein every 10 degree of columns, 36 columns in total；The peak value of histogram then represents the key point
The principal direction for locating neighborhood gradient, i.e., as the direction of the key point；
In gradient orientation histogram, when there are the peak value that another is equivalent to 80% energy of main peak value, then by this
Direction is considered the auxiliary direction of the key point；One key point may be designated with multiple directions, a principal direction, one
The above auxiliary direction, so far, the key point of image, which has detected, to be finished, and there are three information for each key point：Position, residing scale, side
To；Thereby determine that a SIFT feature region；
The direction that reference axis is rotated to be to key point, to ensure rotational invariance, the generation of feature point description.
Further, it after the SIFT feature vector of sunflower leaf image to be identified and disease reference picture generates, waits knowing
The nearest the first two key point of Euclidean distance in key point and disease reference picture in other sunflower leaf image, in the two passes
In key point, the similarity determination for being used as key point in two images using the Euclidean distance of key point feature vector is measured, such as
The nearest distance of fruit divided by secondary close distance are less than some proportion threshold value, then receive this pair of of match point.
Further, the disease reference picture of four class diseases is stored in the sample database, four class diseases are respectively bacterium
Property leaf spot, powdery mildew, black spot, downy mildew.
Compared with prior art, the present invention is based on the sunflower disease recognition method of SIFT feature extraction algorithm at least just like
Lower advantage：
The present invention identifies sunflower leaf portion different diseases using SIFT algorithms, obtains sunflower graph picture in big field first；
When secondly in view of shooting image, inevitably influenced to lead to figure by factors such as various weather, environment, shooting angle
As situations such as of poor quality and smudgy, image is pre-processed before carrying out image recognition, select histogram equalization
Defogging processing is carried out to image, denoising is carried out to image using homomorphic filtering method；Again, pre- place is extracted using SIFT algorithms
Image feature vector after reason carries out the extraction of scale space extreme value, positioning feature point, characteristic direction tax to images to be recognized
Value, extraction feature point description this several step, obtain image feature vector；Finally using the Euclidean distance of key point feature vector come
Similarity determination as the key point of image in images to be recognized and picture library is measured, and the knowledge of sunflower leaf portion disease geo-radar image is completed
Not, discrimination reaches 93.33% or more, and recognition effect is preferable.
Above description is only the general introduction of technical solution of the present invention, in order to better understand the technical means of the present invention,
And can be implemented in accordance with the contents of the specification, below with presently preferred embodiments of the present invention and after coordinating attached drawing to be described in detail such as.
Description of the drawings
Fig. 1 is the block diagram of the sunflower disease recognition method the present invention is based on SIFT feature extraction algorithm；
Fig. 2 is the acquisition system figure of the sunflower disease recognition method the present invention is based on SIFT feature extraction algorithm；
Fig. 3 is the image histogram equalization of the sunflower disease recognition method the present invention is based on SIFT feature extraction algorithm
Processing, a original image histograms, b treated image histograms；
Fig. 4 be the shape of the H (u, v) of the sunflower disease recognition method the present invention is based on SIFT feature extraction algorithm substantially
Figure；
Fig. 5 is the homomorphic filtering principle frame of the sunflower disease recognition method the present invention is based on SIFT feature extraction algorithm
Figure；
Fig. 6 be the sunflower disease recognition method the present invention is based on SIFT feature extraction algorithm extraction SIFT feature and
The flow chart of its feature vector；
Fig. 7 is the gaussian pyramid of the sunflower disease recognition method the present invention is based on SIFT feature extraction algorithm；
Fig. 8 is the Gaussian image and DoG gold of the sunflower disease recognition method the present invention is based on SIFT feature extraction algorithm
Word tower；
Fig. 9 is the spaces the DOG local extremum of the sunflower disease recognition method the present invention is based on SIFT feature extraction algorithm
Detection；
Figure 10 is the true by histogram of gradients of the sunflower disease recognition method the present invention is based on SIFT feature extraction algorithm
Fixed main gradient direction；
Figure 11 is the sunflower disease recognition method the present invention is based on SIFT feature extraction algorithm by key point field ladder
It spends information and generates feature vector；
Figure 12 is that the sunflower black spot of the sunflower disease recognition method the present invention is based on SIFT feature extraction algorithm exists
Image under different scale space；
Figure 13 is four kinds of the sunflower leaf portion of the sunflower disease recognition method the present invention is based on SIFT feature extraction algorithm
Disease recognition situation.
Specific implementation mode
With reference to the accompanying drawings and examples, the specific implementation mode of the present invention is described in further detail.Implement below
Example is not limited to the scope of the present invention for illustrating the present invention.
People are when identifying an object, no matter this object is close or remote, can correctly be identified to it, this
" scale invariability " described in being exactly；Equally, when this object rotates, the mankind still can correctly identify it, this
" rotational invariance " described in being exactly ... how so that machine has this ability as the mankind, here it is image locals not
The problem of denaturation is solved.And SIFT feature extracting method is the feature detection side based on scale and angle invariability technological improvement
Method, it has invariance to image scaling, rotation even radiation transformation, there is higher robustness and higher computational stability.
So the present invention using the different types of disease of four kinds of SIFT feature extracting method identification sunflower leaf portion (bacterial leaf spot,
Powdery mildew, black spot, downy mildew).
Join and present embodiments provides a kind of sunflower disease knowledge based on SIFT feature extraction algorithm shown in Fig. 1 to Figure 13
Other method, including：Sunflower leaf portion disease geo-radar image to be identified is obtained, figure is carried out to the sunflower leaf portion disease geo-radar image to be identified
Image intensifying, denoising, smooth, sharpening pretreatment；
The SIFT feature vector that the sunflower leaf image to be identified is extracted based on SIFT feature extraction algorithm, will be to be identified
Sunflower leaf portion disease geo-radar image carries out SIFT feature vector with all kinds of disease reference pictures in sample database respectively and carries out matching degree
Analysis,
If the number that the SIFT feature vector of sunflower leaf portion disease geo-radar image to be identified and a kind of disease reference picture matches
Amount is greater than or equal to predetermined threshold, then judges that the corresponding sunflower leaf of the sunflower leaf portion disease geo-radar image to be identified is such disease
Sunflower leaf；
If the SIFT feature vector of sunflower leaf portion disease geo-radar image to be identified and all kinds of disease reference pictures in sample database
The quantity to match is both less than predetermined threshold, then judges that the corresponding sunflower leaf of the sunflower leaf image to be identified is not sample database
In any kind disease.
The acquisition of sunflower leaf portion disease geo-radar image is specific to be introduced：
The external information about 60% that the mankind obtain is from visual pattern, is regarded so how to obtain and handle
Feel that information just seems to be even more important.In early days, people are typically all and use conventional methods, and mainly pass through black and white to the acquisition of image
Or obtained from the shooting of colour film camera, but the image quantity that this method not only stores, and captured image
It is analog image.Later, with the development of science and technology, the especially development of multimedia technology, high-resolution digital camera
It is increasingly popularized with DV.Now, as the fast development of mobile communication technology, high-resolution and high-definition are taken pictures
The use of mobile phone becomes more and more popular in people's lives, therefore the present invention uses Android phone conduct herein
The sampling instrument of sunflower leaf portion disease geo-radar image, this trifle mainly introduce the various information of disease geo-radar image acquisition.
Under the environment of crop field, natural lighting and complex background etc. all can identify have an impact to subsequent image, so making
When shooting image with cell phone cameras, free white background board is placed in below blade, simplifies background, keeps mobile phone and blade as possible
It is parallel, obtain the higher image pattern of quality.Sunflower leaf portion disease geo-radar image acquisition system is as shown in Figure 2.
The sample number of Image Acquisition：Four kinds of 480 width of sunflower leaf portion disease geo-radar image are acquired altogether.Wherein, bacillary leaf
120 width of pinta image, 120 width of downy mildew image, 120 width of black spot image, 120 width of powdery mildew image.
The pretreatment of sunflower leaf portion disease geo-radar image is specific to be introduced：
Since sunflower leaf portion disease geo-radar image is collected under natural lighting by mobile phone, so inevitably
Can be influenced to cause by various factors poor image quality and it is smudgy situations such as, for the correct knowledge of image disease below
Not, it before carrying out follow-up disease recognition work, first has to pre-process image, is easy to sunflower disease recognition to obtain
Ideal image.
The purpose of image preprocessing is exactly to improve picture quality and highlight characteristics of image.It to the information in image into
The inhibition or reinforcement of row selectivity, eliminate unnecessary information, to which enhancing is to the reliability of subsequent image disease recognition.Figure
Picture pretreatment includes mainly the contents such as image enhancement, denoising, smooth, sharpening.
The equalization of histogram：The equalization (Histogram Equalization) of histogram is that common histogram is repaiied
Correction method, also known as histogram equalization.It is adjusted to contrast using image histogram in image processing field
A kind of method is typically used for increasing the local contrast of image, especially when the contrast of the useful data of image very close to
When.By the method for histogram equalization, the brightness of image can be good at being distributed on the histogram, in this way, both enhancing
The contrast of image local, and do not influence the contrast of image entirety.Histogram equalization is bright by effective expanded images
It spends to realize the effect of image enhancement.
The core concept of histogram equalization is exactly the grey level histogram of original image from some gray scale for comparing concentration
Section becomes being uniformly distributed in whole tonal ranges.Nonlinear extension substantially is carried out to original image, is redistributed
The pixel value of original image keeps the pixel quantity in certain tonal range roughly the same.In this way, among the histogram of original image
Crest portion contrast is just enhanced, and the part contrast of both sides the lowest point reduces, and the histogram of treated image is just
Become the histogram of " uniform " distribution.Histogram equalization processing algorithm, which sums up, is broadly divided into following three steps：
(1) it to its histogram of the pending image statistics of input, finds out
In formula 3-1, L is gray level；pr(rk) it is the probability for taking kth grade gray value；nkIt is to occur kth grade ash in the picture
The number of degree；N is sum of all pixels in image.
(2) it is converted with cumulative distribution function according to the histogram come out, cumulative distribution function is as follows：
Find out the new gray scale after transformation.
(3) old gray scale is replaced with new gray scale, finds out PS(s), this step is approximation, is done as possible according to the purpose of processing
To reasonable while gray value is equal or be approximately merged together.
After sunflower leaf portion disease geo-radar image is handled (by taking black spot as an example) with histogram equalizing method shown in figure figure below：
As can be seen that the gray level of original image concentrates on a narrow model in histogram equalization processing design sketch
In enclosing, dynamic range is relatively narrow, and the intensity value ranges occupied are small.After histogram equalization processing, histogram occupies whole
The range that a gray value of image allows, increases gradation of image dynamic range, also increases the contrast of image, entire brightness region
The distribution that domain has comparison average, image effect is more smooth after processing, as seen from the figure, has reached good defog effect.Cause
This, selects histogram equalization to carry out defogging processing to image herein.
Homomorphic filtering：Homomorphic filtering is a kind of one of method of image enhancement carried out in a frequency domain.It can reduce low
Frequently and increase high frequency, so as to reduce illumination variation and sharpen edge details.Gradation of image range is adjusted, is eliminated
Unevenness is illuminated on image, can protect details well, takes out the noise in signal.Homomorphic filter is exactly a kind of nonlinear filtering
Wave device, is a kind of method of the contrast enhancing of feature based, and uneven caused image deterioration is made for reducing radiance, and
Interested part can effectively be enhanced.
If piece image, illumination function is expressed as with image function f (x, y), can with its incident light component i (x,
Y) it is indicated with the product of reflected light component r (x, y), then relational expression can be expressed as
F (x, y)=i (x, y) r (x, y) 0<r(x,y)<∞, 0<i(x,y)<∞ (3-3)
Homomorphic filtering is carried out, there are following several steps：
(1) it first has to take logarithm, purpose exactly to convert multiplying to addition fortune original image f (x, y)
It calculates：Z (x, y)=lnf (x, y)=lni (x, y)+lnr (x, y) (3-4)
(2) Fourier transformation is done to logarithmic function again, image is exactly transformed into frequency domain by purpose：
F (z (x, y))=F [lni (x, y)]+F [lnr (x, y)] i.e. Z=I+R
Transmission function appropriate, the variation range of press irradiation component i (x, y) is selected to weaken I (u, v), enhancing r (x, y)
Contrast, promoted R (u, v), enhance high fdrequency component, that is, determine a suitable H (u, v).
(3) one homomorphic filter function H (u, v) of selection is assumed to handle the logarithm of original image f (x, y)
Fourier transformation Z (u, v), obtains
S (u, v)=Z (u, v) H (u, v)=I (u, v) H (u, v)+R (u, v) H (u, v)
Another mistake transforms to spatial domain and obtains s (x, y)=F-1(S(u,v))
(4) final result g (x, y)=exp (s (x, y)) i.e. is obtained to fetching number again, is equivalent to high-pass filtering.
The background luminance of original image is weakened, and the contrast at scab and its edge enhances, and has been effectively maintained scab
Details, effectively enhance image useful information contrast, there is preferable denoising effect.Therefore, homomorphic filtering method is selected herein
Denoising is carried out to image, enhances picture contrast.
SIFT (Scale-invariant feature transform, scale invariant feature conversion) is a kind of based on meter
The recognizer of calculation machine vision finds extreme value for detecting and describing the handle feature in image in its space scale
Point, and its position, scale, rotational invariants are extracted, which is the method for summarizing existing feature extraction by Lowe
A kind of new and effective feature detection description method proposed afterwards.SIFT algorithms are all obtained at military, industry and civilian aspect at present
Arrived different degrees of application, many fields have been permeated in application, typical application as object identification, robot localization with
Navigation, image mosaic, three-dimensional modeling, gesture identification, video tracking, notes identification, fingerprint and recognition of face, crime scene features
Extraction etc..
The main thought of SIFT algorithms is to initially set up the scale space of image to indicate, then search graph in scale space again
The extreme point of picture has extreme point to resettle feature description vector, finally carries out similarity mode with feature description vector.The algorithm
There is its uniqueness：
(1) SIFT algorithms are the local features of image, are maintained the invariance to image rotation, scaling, brightness change,
Also there is certain tolerance to visual angle change, radiation variation, noise.
(2) unique good, SIFT feature informative, is suitble to quick and precisely match in high-volume database.
(3) object of volume, a small number of several texture-rich can also generate a large amount of SIFT feature vector.
(4) high speed, the SIFT algorithms by optimization even can reach real-time requirement.
The realization of SIFT feature matching algorithm includes two stages, and the first stage is the generation of SIFT feature vector, is exactly
From needing to extract the feature vector unrelated to scaling, rotation, brightness change in matched image；Second stage is exactly
The matching of SIFT feature vector.
In first stage, the generating algorithm of the SIFT feature vector of piece image is divided into four steps：Scale space extreme value
Extraction；Positioning feature point；Characteristic direction assignment；Extract feature point description.
(1) scale space extremum extracting
Scale-space theory is the basis for detecting invariance.Witkin (1983) proposes Scale-space theory,
This theory is expanded to two dimensional image by Koenderink (1984), and it is to realize change of scale only to demonstrate Gaussian convolution core
One not core.This scale invariability is obtained according to Scale-space theory.
The scale space L (x, y, σ) of one width two dimensional image indicate can by a change of scale gaussian kernel function G (x, y,
It σ) is obtained with original image I (x, y) convolution, such as following formula：
L (x, y, σ)=G (x, y, σ) * I (x, y) (4-44)
Wherein, (x, y) indicates that the pixel coordinate of picture point, I (x, y) indicate that image data, σ are known as the scale space factor, *
Indicate convolution algorithm.
Wherein, scale space factor sigma is the variance of Gauss normal distribution, shows the degree that image is smoothed, and σ values are smaller
Indicate that the degree that image is smoothed is smaller, corresponding scale is also smaller.The minutia of small scale correspondence image, large scale correspond to
The general picture feature of image.
The realization of scale space uses the expression of gaussian pyramid.The gaussian pyramid model of image is exactly by original image
Constantly depression of order samples, and obtains a series of images not of uniform size, descending, constitutes a tower-like model from top to bottom.
Original image is as pyramidal first layer, and each down-sampled obtained new image is as pyramidal one layer (one every layer
Image), each pyramid shares n-layer.The pyramidal number of plies is determined jointly according to the size of original image and the size of tower top image
Fixed.
In order to effectively detect stable key point in scale space, it is proposed that Gaussian difference scale space (DoG).
DoG operators (Difference of Gaussians Operator) are the differences of two different Gaussian functions of scale, can also
As the enhancing operator of gray level image, it can be effectively used to the edge of enhancing image, i.e., constantly to original-gray image
It is obscured, retains the marginal portion on original image, can also remove picture noise, be particularly suitable for carrying and largely make an uproar at random
The image of sound.
If k is the scale factor between two adjacent scales, then DoG operator definitions are as follows：
D (x, y, σ)=(G (x, y, σ)-G (x, y, σ)) * I (x, y)
=L (x, y, k σ)-L (x, y, σ) (4-47)
DoG operators calculate simply, are the approximations of the LoG operators of dimension normalization.The ruler of first layer in DoG pyramids
Spend the factor and Gauss
DoG spatial extremas detect, and need the DoG values for detecting 26 pixels, therefrom select DoG values maximum or minimum value is made
For candidate feature point, its position and corresponding scale are write down.This 26 pixels are chosen such that the spaces DoG any layer (most bottom
Layer and top except) arbitrary pixel same layer in adjacent 8 pixels and its last layer and the 9 of next layer it is adjacent
Pixel.
(2) precise positioning feature point position
After obtaining candidate feature point, precise positioning feature point position is next wanted, that is, rejected in candidate feature point
Low contrast point and unstable skirt response point.Characteristic point cannot be the point of low contrast, that is to say, that the pixel of characteristic point
Value must have apparent difference with the point of surrounding.Characteristic point cannot be marginal point, because DOG operators will produce stronger side
Edge responds, and removal marginal point helps to enhance matched stability, improves noise resisting ability.
(3) key point direction determines：
Next it is each key point assigned direction number using the gradient direction distribution characteristic of key point neighborhood territory pixel, makes calculation
Son has rotational invariance.For each Gaussian image.The mould m (x, y) and direction θ (x, y) of the gradient of each point L (x, y) are logical
Following formula is crossed to be calculated：
In above formula, scale used in L (x, y) is scale where key point.
When actually calculating, the present invention samples in neighborhood window centered on by key point, is used in combination statistics with histogram adjacent
The gradient direction of domain pixel.The range of histogram of gradients is 360 degree, wherein every 10 degree of columns, 36 columns in total.Histogram
Peak value then represents the principal direction of neighborhood gradient at the key point, i.e., as the direction of the key point.
In gradient orientation histogram, when there are the peak value that another is equivalent to 80% energy of main peak value, then by this
Direction is considered the auxiliary direction of the key point.One key point may be designated with multiple directions (principal direction, one
The above auxiliary direction), this can enhance matched robustness.So far, the key point of image, which has detected, finishes, and each key point has three
A information：Position, residing scale, direction.Thereby determine that a SIFT feature region.
(4) SIFT feature vector is generated
It is finally the generation of feature point description, reference axis is rotated to be to the direction of key point first, ensures rotation not
Denaturation.
In practical calculating process, in order to enhance matched robustness, to each key point, using 4x4, totally 16 seed points are come
Description, can generate 128 data for a key point in this way, that is, ultimately form the SIFT feature vector of 128 dimensions.At this time
SIFT feature vector has had been removed the influence of the geometry deformations factor such as dimensional variation, rotation, is further continued for the length of feature vector
Degree normalization, then can further remove the influence of illumination variation.
(1) by taking sunflower leaf portion black spot as an example, the SIFT feature vector of image is extracted.
(2) under different scale space, image shows Figure 12.
As seen from the figure, with the increase of space scale, image is increasingly fuzzyyer, while can be seen that in large scale
Profile point seeks to the position of the characteristic point of extraction in space.
(3) characteristic point and form 128 key point description tieed up that image zooming-out arrives.
Sunflower leaf portion black spot image extracts 43 SIFT feature vectors altogether, then can generate the key of 43 128 dimensions
Point description.
After the SIFT feature vector of two images generates, some key point in image l (images to be recognized) is taken, and look for
Go out its first two key point nearest with Euclidean distance in image 2 (image in sample database), in the two key points, the present invention
It is used as the similarity determination measurement of key point in two images using the Euclidean distance of key point feature vector, if nearest
Distance divided by secondary close distance are less than some proportion threshold value, then receive this pair of of match point.It is by inspection information and largely real
It tests, is chosen herein when matching points are more than or equal to 5, it is believed that successful match.
(1) extraction images to be recognized and the SIFT feature of image in sample database are vectorial respectively.
It extracts 43 feature vectors altogether in image to be matched, then can generate key point description of 43 128 dimensions.
98 feature vectors are extracted altogether in sunflower leaf portion black spot image, then can generate the key of 98 128 dimensions
Point description.
(2) images to be recognized carries out characteristic matching with image in sample database.
Image to be matched and sunflower leaf portion black spot Image Feature Matching obtain 8 characteristic matching points, receive at this time this 8
To match point, matching points are more than 5, it is believed that images to be recognized is sunflower leaf portion black spot image.
The sunflower leaf portion disease geo-radar image that Android phone shooting, collecting arrives is chosen herein：Bacterial leaf spot (is denoted as
1), powdery mildew (being denoted as 2), black spot (being denoted as 3), each 120 width of downy mildew (being denoted as 4) image, wherein 60 width are as training sample,
60 width identify matching sunflower four kinds of different diseases of leaf portion, identification situation such as Figure 13 institutes as test sample, using SIFT algorithms
Show：
Count simultaneously analysis and identification result such as following table：
Table 1 sunflower leaf portion, four kinds of disease recognition situations
As can be seen from the above table, identify that sunflower four kinds of diseases of leaf portion, discrimination can reach using SIFT matching algorithms
To 93.33% or more, recognition effect is preferable.
The above is only a preferred embodiment of the present invention, it is not intended to restrict the invention, it is noted that for this skill
For the those of ordinary skill in art field, without departing from the technical principles of the invention, can also make it is several improvement and
Modification, these improvements and modifications also should be regarded as protection scope of the present invention.
Claims (5)
1. a kind of sunflower disease recognition method based on SIFT feature extraction algorithm, which is characterized in that including：
Obtain sunflower leaf portion disease geo-radar image to be identified, to the sunflower leaf portion disease geo-radar image to be identified carry out image enhancement,
Denoising, smooth, sharpening pretreatment；
The SIFT feature vector that the sunflower leaf portion disease geo-radar image to be identified is extracted based on SIFT feature extraction algorithm, will wait knowing
Other sunflower leaf image carries out SIFT feature vector with all kinds of disease reference pictures in sample database respectively and carries out matching degree analysis,
If the quantity that the SIFT feature vector of sunflower leaf portion disease geo-radar image to be identified and a kind of disease reference picture matches is big
In or be equal to predetermined threshold, then judge that the corresponding sunflower leaf of the sunflower leaf portion disease geo-radar image to be identified is such disease to day
Certain herbaceous plants with big flowers leaf；
If the SIFT feature vector phase of sunflower leaf portion disease geo-radar image to be identified and all kinds of disease reference pictures in sample database
The quantity matched is both less than predetermined threshold, then judges that the corresponding sunflower leaf of the sunflower leaf portion disease geo-radar image to be identified is not sample
Any kind disease in library.
2. the sunflower disease recognition method according to claim 1 based on SIFT feature extraction algorithm, which is characterized in that
The pretreatment includes：The equalization of histogram, homomorphic filtering, wherein
The equalization of the histogram specifically includes：
To its histogram of the pending image statistics of input, find out
In formula, L is gray level；pr(rk) it is the probability for taking kth grade gray value；nkIt is time for occurring kth grade gray scale in the picture
Number；N is sum of all pixels in image；
It is converted with cumulative distribution function according to the histogram come out, finds out the new gray scale after transformation, cumulative distribution function
It is as follows：
Old gray scale is replaced with new gray scale, finds out PS(s), gray value is equal or be approximately merged together；
The homomorphic filtering specifically includes：
Logarithm, purpose is taken exactly to convert multiplying to add operation original image f (x, y)：
Z (x, y)=lnf (x, y)=lni (x, y)+lnr (x, y)
Fourier transformation is done to logarithmic function, image is exactly transformed into frequency domain by purpose：
F (z (x, y))=F [lni (x, y)]+F [lnr (x, y)] i.e. Z=I+R
Transmission function appropriate, the variation range of press irradiation component i (x, y) is selected to weaken I (u, v), pair of enhancing r (x, y)
Than degree, R (u, v) is promoted, enhances high fdrequency component, that is, determines a H (u, v)；
A homomorphic filter function H (u, v) is selected to handle the Fourier transformation of the logarithm of original image f (x, y)
Z (u, v), obtains
S (u, v)=Z (u, v) H (u, v)=I (u, v) H (u, v)+R (u, v) H (u, v),
Inversion changes to spatial domain and obtains s (x, y)=F-1(S(u,v))；
Final result g (x, y)=exp (s (x, y)) i.e. is obtained to fetching number, is equivalent to high-pass filtering.
3. the sunflower disease recognition method according to claim 1 based on SIFT feature extraction algorithm, which is characterized in that
The extraction step of SIFT feature vector includes：Scale space extreme value is extracted；Positioning feature point；Characteristic direction assignment；Extract feature
Point description；
The scale space L (x, y, σ) of one width two dimensional image indicate can by a change of scale gaussian kernel function G (x, y, σ) with
Original image I (x, y) convolution obtains, such as following formula：
L (x, y, σ)=G (x, y, σ) * I (x, y)
Wherein, (x, y) indicates that the pixel coordinate of picture point, I (x, y) indicate that image data, σ are known as the scale space factor, and * is indicated
Convolution algorithm；
Wherein, scale space factor sigma is the variance of Gauss normal distribution, shows the degree that image is smoothed, the smaller expression of σ values
The degree that image is smoothed is smaller, and corresponding scale is also smaller；
Gaussian difference scale space DoG, DoG operator is the difference of two different Gaussian functions of scale, as gray level image
Enhancing operator, can be effectively used to enhancing image edge, i.e., constantly original-gray image is obscured, retain it is original
Marginal portion on image can also remove picture noise, be particularly suitable for the image with a large amount of random noises, if k is two
Scale factor between a adjacent scale, then DoG operator definitions are as follows：
D (x, y, σ)=(G (x, y, σ)-G (x, y, σ)) * I (x, y)
=L (x, y, k σ)-L (x, y, σ)
DoG operators calculate simply, are the approximations of the LoG operators of dimension normalization；The scale of first layer in DoG pyramids because
Son is consistent with the first layer in gaussian pyramid；
DoG spatial extremas detect, and need the DoG values for detecting 26 pixels, therefrom select DoG values maximum or minimum value as time
Characteristic point is selected, its position and corresponding scale are write down；This 26 pixels be chosen such that the spaces DoG any layer (bottom and
Except top) arbitrary pixel same layer in adjacent 8 pixels and its last layer and the 9 of next layer adjacent pictures
Element；
After obtaining candidate feature point, precise positioning feature point position is next wanted, namely low comparison is rejected in candidate feature point
Degree point and unstable skirt response point, the value of the pixel of characteristic point must have apparent difference with the point of surrounding；Characteristic point
It cannot be marginal point；
It is each key point assigned direction number using the gradient direction distribution characteristic of key point neighborhood territory pixel, operator is made to have rotation
Invariance；For each Gaussian image；The mould m (x, y) and direction θ (x, y) of the gradient of each point L (x, y) pass through following formula
It is calculated：
In above formula, scale used in L (x, y) is scale where key point；
It is sampled in neighborhood window centered on key point, the gradient direction of statistics with histogram neighborhood territory pixel is used in combination；Gradient histogram
The range of figure is 360 degree, wherein every 10 degree of columns, 36 columns in total；The peak value of histogram then represents adjacent at the key point
The principal direction of domain gradient, i.e., as the direction of the key point；
In gradient orientation histogram, when there are the peak value that another is equivalent to 80% energy of main peak value, then by this direction
It is considered the auxiliary direction of the key point；One key point may be designated with multiple directions, a principal direction, more than one
Auxiliary direction, so far, the key point of image, which has detected, to be finished, and there are three information for each key point：Position, residing scale, direction；By
This determines a SIFT feature region；
The direction that reference axis is rotated to be to key point, to ensure rotational invariance, the generation of feature point description.
4. the sunflower disease recognition method according to claim 1 based on SIFT feature extraction algorithm, which is characterized in that
After the SIFT feature vector of sunflower leaf image to be identified and disease reference picture generates, in sunflower leaf image to be identified
The key point the first two key point nearest with Euclidean distance in disease reference picture, in the two key points, using key point
The Euclidean distance of feature vector is used as the similarity determination measurement of key point in two images, if nearest distance divided by secondary
Close distance is less than some proportion threshold value, then receives this pair of of match point.
5. the sunflower disease recognition method according to claim 1 based on SIFT feature extraction algorithm, which is characterized in that
It is stored with the disease reference picture of four class diseases in the sample database, four class diseases are respectively bacterial leaf spot, powdery mildew, black
Pinta, downy mildew.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810325111.8A CN108564092A (en) | 2018-04-12 | 2018-04-12 | Sunflower disease recognition method based on SIFT feature extraction algorithm |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810325111.8A CN108564092A (en) | 2018-04-12 | 2018-04-12 | Sunflower disease recognition method based on SIFT feature extraction algorithm |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108564092A true CN108564092A (en) | 2018-09-21 |
Family
ID=63534784
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810325111.8A Pending CN108564092A (en) | 2018-04-12 | 2018-04-12 | Sunflower disease recognition method based on SIFT feature extraction algorithm |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108564092A (en) |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106780440A (en) * | 2016-11-29 | 2017-05-31 | 北京邮电大学 | Destruction circuit plate relic image automatic comparison recognition methods |
| CN109409387A (en) * | 2018-11-06 | 2019-03-01 | 深圳增强现实技术有限公司 | The acquisition direction of image capture device determines method, apparatus and electronic equipment |
| CN111652285A (en) * | 2020-05-09 | 2020-09-11 | 济南浪潮高新科技投资发展有限公司 | Tea cake category identification method, equipment and medium |
| CN113155293A (en) * | 2021-04-06 | 2021-07-23 | 内蒙古工业大学 | Human body remote sensing temperature measurement monitoring and recognition system based on unmanned aerial vehicle |
| CN113343870A (en) * | 2021-06-17 | 2021-09-03 | 南京金盾公共安全技术研究院有限公司 | Identification and enabling method for mobile equipment based on Android system |
| CN113409253A (en) * | 2021-06-02 | 2021-09-17 | 南京公诚节能新材料研究院有限公司 | Agricultural condition monitoring key technical method |
| CN113420633A (en) * | 2021-06-18 | 2021-09-21 | 桂林电子科技大学 | Traffic sign identification method based on UM enhancement and SIFT feature extraction |
| CN113723157A (en) * | 2020-12-15 | 2021-11-30 | 京东数字科技控股股份有限公司 | Crop disease identification method and device, electronic equipment and storage medium |
| CN115131683A (en) * | 2022-08-25 | 2022-09-30 | 金乡县林业保护和发展服务中心(金乡县湿地保护中心、金乡县野生动植物保护中心、金乡县国有白洼林场) | Forestry information identification method based on high-resolution remote sensing image |
| CN116740378A (en) * | 2023-07-03 | 2023-09-12 | 南通黄海药械有限公司 | Garden plant diseases and insect pests evaluation system based on image processing |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20070110306A1 (en) * | 2005-11-14 | 2007-05-17 | Haibin Ling | Diffusion distance for histogram comparison |
| CN102722731A (en) * | 2012-05-28 | 2012-10-10 | 南京航空航天大学 | Efficient image matching method based on improved scale invariant feature transform (SIFT) algorithm |
| CN104899834A (en) * | 2015-03-04 | 2015-09-09 | 苏州大学 | Blurred image recognition method and apparatus based on SIFT algorithm |
| CN106778807A (en) * | 2016-11-22 | 2017-05-31 | 天津大学 | The fine granularity image classification method of dictionary pair is relied on based on public dictionary pair and class |
| CN107330892A (en) * | 2017-07-24 | 2017-11-07 | 内蒙古工业大学 | A kind of sunflower disease recognition method based on random forest method |
-
2018
- 2018-04-12 CN CN201810325111.8A patent/CN108564092A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20070110306A1 (en) * | 2005-11-14 | 2007-05-17 | Haibin Ling | Diffusion distance for histogram comparison |
| CN102722731A (en) * | 2012-05-28 | 2012-10-10 | 南京航空航天大学 | Efficient image matching method based on improved scale invariant feature transform (SIFT) algorithm |
| CN104899834A (en) * | 2015-03-04 | 2015-09-09 | 苏州大学 | Blurred image recognition method and apparatus based on SIFT algorithm |
| CN106778807A (en) * | 2016-11-22 | 2017-05-31 | 天津大学 | The fine granularity image classification method of dictionary pair is relied on based on public dictionary pair and class |
| CN107330892A (en) * | 2017-07-24 | 2017-11-07 | 内蒙古工业大学 | A kind of sunflower disease recognition method based on random forest method |
Non-Patent Citations (2)
| Title |
|---|
| 张书霞等: "SIFT特征匹配算法研究", 《现代计算机》 * |
| 汪秦峰: "基于直方图均衡化和Retinex的图像去雾算法研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106780440B (en) * | 2016-11-29 | 2019-05-31 | 北京邮电大学 | Destruction circuit plate relic image automatic comparison recognition methods |
| CN106780440A (en) * | 2016-11-29 | 2017-05-31 | 北京邮电大学 | Destruction circuit plate relic image automatic comparison recognition methods |
| CN109409387A (en) * | 2018-11-06 | 2019-03-01 | 深圳增强现实技术有限公司 | The acquisition direction of image capture device determines method, apparatus and electronic equipment |
| CN109409387B (en) * | 2018-11-06 | 2022-03-15 | 深圳增强现实技术有限公司 | Acquisition direction determining method and device of image acquisition equipment and electronic equipment |
| CN111652285A (en) * | 2020-05-09 | 2020-09-11 | 济南浪潮高新科技投资发展有限公司 | Tea cake category identification method, equipment and medium |
| CN113723157A (en) * | 2020-12-15 | 2021-11-30 | 京东数字科技控股股份有限公司 | Crop disease identification method and device, electronic equipment and storage medium |
| CN113723157B (en) * | 2020-12-15 | 2024-02-09 | 京东科技控股股份有限公司 | Crop disease identification method and device, electronic equipment and storage medium |
| CN113155293A (en) * | 2021-04-06 | 2021-07-23 | 内蒙古工业大学 | Human body remote sensing temperature measurement monitoring and recognition system based on unmanned aerial vehicle |
| CN113409253A (en) * | 2021-06-02 | 2021-09-17 | 南京公诚节能新材料研究院有限公司 | Agricultural condition monitoring key technical method |
| CN113343870A (en) * | 2021-06-17 | 2021-09-03 | 南京金盾公共安全技术研究院有限公司 | Identification and enabling method for mobile equipment based on Android system |
| CN113343870B (en) * | 2021-06-17 | 2024-02-23 | 南京金盾公共安全技术研究院有限公司 | Identification enabling method based on Android system mobile equipment |
| CN113420633A (en) * | 2021-06-18 | 2021-09-21 | 桂林电子科技大学 | Traffic sign identification method based on UM enhancement and SIFT feature extraction |
| CN113420633B (en) * | 2021-06-18 | 2022-04-12 | 桂林电子科技大学 | Traffic sign identification method based on UM enhancement and SIFT feature extraction |
| CN115131683A (en) * | 2022-08-25 | 2022-09-30 | 金乡县林业保护和发展服务中心(金乡县湿地保护中心、金乡县野生动植物保护中心、金乡县国有白洼林场) | Forestry information identification method based on high-resolution remote sensing image |
| CN116740378A (en) * | 2023-07-03 | 2023-09-12 | 南通黄海药械有限公司 | Garden plant diseases and insect pests evaluation system based on image processing |
| CN116740378B (en) * | 2023-07-03 | 2024-04-02 | 南通黄海药械有限公司 | Garden plant diseases and insect pests evaluation system based on image processing |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108564092A (en) | Sunflower disease recognition method based on SIFT feature extraction algorithm | |
| Gao et al. | Automatic change detection in synthetic aperture radar images based on PCANet | |
| CN103077512B (en) | Based on the feature extracting and matching method of the digital picture that major component is analysed | |
| CN111369605B (en) | Infrared and visible light image registration method and system based on edge features | |
| CN106557740B (en) | The recognition methods of oil depot target in a kind of remote sensing images | |
| Hussin et al. | Plant species identification by using scale invariant feature transform (sift) and grid based colour moment (gbcm) | |
| CN105654122B (en) | Based on the matched spatial pyramid object identification method of kernel function | |
| CN112308873B (en) | Edge detection method for multi-scale Gabor wavelet PCA fusion image | |
| CN111259756A (en) | Pedestrian re-identification method based on local high-frequency features and mixed metric learning | |
| CN110348289A (en) | A kind of finger vein identification method based on binary map | |
| CN111814690B (en) | Target re-identification method, device and computer readable storage medium | |
| CN110766657B (en) | Laser interference image quality evaluation method | |
| George et al. | Leaf identification using Harris corner detection, SURF feature and FLANN matcher | |
| CN105825215B (en) | It is a kind of that the instrument localization method of kernel function is embedded in based on local neighbor and uses carrier | |
| Jubair et al. | Scale invariant feature transform based method for objects matching | |
| CN110321890B (en) | Digital instrument identification method of power inspection robot | |
| CN115035281B (en) | Rapid infrared panoramic image stitching method | |
| CN108960285B (en) | Classification model generation method, tongue image classification method and tongue image classification device | |
| Kim et al. | Improving faster r-cnn framework for multiscale Chinese character detection and localization | |
| Liu et al. | SAR image matching based on speeded up robust feature | |
| Wu et al. | An accurate feature point matching algorithm for automatic remote sensing image registration | |
| CN114926635A (en) | Method for segmenting target in multi-focus image combined with deep learning method | |
| CN113269020A (en) | Fingertip vein image feature identification positioning method and system | |
| CN107239754B (en) | Automobile logo identification method based on sparse sampling intensity profile and gradient distribution | |
| WO2015168362A1 (en) | Method and apparatus for processing block to be processed of urine sediment image |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180921 |