CN108280483A - Trypetid adult image-recognizing method based on neural network - Google Patents

Trypetid adult image-recognizing method based on neural network Download PDF

Info

Publication number
CN108280483A
CN108280483A CN201810087133.5A CN201810087133A CN108280483A CN 108280483 A CN108280483 A CN 108280483A CN 201810087133 A CN201810087133 A CN 201810087133A CN 108280483 A CN108280483 A CN 108280483A
Authority
CN
China
Prior art keywords
trypetid
neural network
image
striped
region
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
Application number
CN201810087133.5A
Other languages
Chinese (zh)
Inventor
李震
吕石磊
温威
邓忠易
代秋芳
薛秀云
洪添胜
宋淑然
吴伟斌
朱余清
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
South China Agricultural University
Original Assignee
South China Agricultural University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by South China Agricultural University filed Critical South China Agricultural University
Priority to CN201810087133.5A priority Critical patent/CN108280483A/en
Publication of CN108280483A publication Critical patent/CN108280483A/en
Pending legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/06—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons
    • G06N3/061—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons using biological neurons, e.g. biological neurons connected to an integrated circuit
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • 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
    • G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
    • 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
    • G06V10/56—Extraction of image or video features relating to colour

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • General Engineering & Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Multimedia (AREA)
  • Neurology (AREA)
  • Computational Linguistics (AREA)
  • General Health & Medical Sciences (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Microelectronics & Electronic Packaging (AREA)
  • Evolutionary Biology (AREA)
  • Image Analysis (AREA)

Abstract

Trypetid adult image-recognizing method disclosed by the invention based on neural network, includes the following steps：Obtain trypetid sample image；Trypetid sample image is handled using Hough transform, adjusts image, limits striped effective coverage；Filter value processing is carried out to image using HSV color spaces, locks the scultellum region in the middle part of trypetid；It obtains trypetid characteristic area, defines 4 kinds of characterization factors of trypetid form, extract feature vector；A certain number of feature vectors are acquired, BP neural network is established and data set is trained, obtain neural network model parameter, build identification model；Feature vector is obtained to the trypetid image that need to be identified, feature vector is inputted into identification model, identification model exports recognition result；The present invention can reach good feature extraction effect when objective image quality is poor, accurately be identified from image, realize that computer automatically locks trypetid characteristic area, efficiency and accuracy rate are higher, improve the working efficiency of pest control.

Description

Trypetid adult image-recognizing method based on neural network
Technical field
The present invention relates to computer picture recognition fields, more particularly to the trypetid adult image recognition side based on neural network Method.
Background technology
Citrus fruit fly, pumpkin fruit fly, melonfly are the sociales of south China trypetid class, because its host range is wide, harm Property is strong, is included in important quarantine object by multiple countries and regions in the world.In China, above-mentioned three kinds of trypetids are mainly distributed on Southwest, south China and Taiwan, the Important Economics crop such as main harm citrus, guava, mango, pumpkin are southern citrus productions Endanger most serious and the primary pest there is an urgent need to prevention in area.Compared to the control method of traditional personal monitoring's insect pest, base High human cost is not only saved in the method for real-time of machine vision, can also reach high on work effect expires Meaning degree.Pest is accurately identified from image, be build the pest real-time monitoring system based on machine vision technique it is important before It carries.
Common insect image identification Target Recognition Algorithms using polypide configuration feature and color character as identification according to According to, such as Yang Hong treasure etc. carries out the extraction based on shape and color feature value to insect image identification, establishes radial base neural net point Class device is identified；Lou Dingfeng etc. proposes a kind of general insect image identification algorithm for pattern recognition based on shape and texture, to difference The insect of figure and color achieves good recognition effect.
Due to fruit-fly classified complexity, whole geometry feature and color character can not to its entirely accurate divide Class, the local feature by digitized processing have more excellent sort feature instead.Wang Lu etc. is with the Europe between trypetid wing punctuate Family name's distance is characteristic of division, and the identification of Diptera Bactrocera insect is realized using random forests algorithm；The uses such as Peng Yingqiong Mark point method in geometric shape surveying carries out feature extraction to drosophila wing, is realized to fruit fly in conjunction with BP neural network algorithm Classification；Zhang Lei classifies to drosophila wing and thoracic dorsal provincial characteristics using improved boost algorithms.
These researchs have high requirements to trypetid picture quality using trypetid fin structure as feature extraction object, if figure Piece is not clear enough, is not accurately captured the vein structure and pterostigma of trypetid wing, and the effect of identification can be a greater impact；In addition, Existing algorithm does not well solve the problem of automatic lock-in feature region from source images yet.
Invention content
The shortcomings that it is an object of the invention to overcome the prior art and deficiency, provide the trypetid adult based on neural network Image-recognizing method, selected characteristic is apparent and is easy to the thoracic dorsal lath line of extraction and is used as classification foundation, based on HSV color spaces with Hough transform carries out segmentation of feature regions, and proposes to be digitized place with the relevant 4 kinds of validity features of striped morphological feature Reason forms input of the feature vector as neural network, and trypetid adult has preferable recognition effect, and recognition efficiency disclosure satisfy that The requirement monitored in real time.
The purpose of the present invention is realized by the following technical solution：
Trypetid adult image-recognizing method based on neural network, includes the following steps：
S1, trypetid sample image is obtained；
S2, trypetid sample image is handled using Hough transform, modulation image trypetid dipteron linear position makes figure Trypetid rotates to be body form upward as in, while limiting striped effective coverage；
S3, filter value processing is carried out to image using HSV color spaces, locks the scultellum region in the middle part of trypetid；
S4, using trypetid dipteron linear position and the thoracic dorsal plate of trypetid scultellum area locking trypetid as trypetid characteristic area, And the center striped in the region is further processed, according to the description method of center striped shape feature, define the 4 of trypetid form Kind characterization factor, and according to characterization factor, extract feature vector；
Clarification of objective vector in S5, a certain number of trypetid images of acquisition, establishes BP neural network and is carried out to data set Training, to obtain for fruit-fly classified neural network model parameter, based on neural network model parameter, builds identification Model；Wherein acquisition threshold value is K, when collecting quantity is more than K, then carries out in next step, when collecting quantity is less than K, then returning to Step S1；
S6, S2-S4 step process is carried out to the trypetid image that needs identify, obtains feature vector；Feature vector is inputted Identification model, identification model export recognition result.
In step S2, the Hough transform process is as follows：
Y1, gaussian filtering process is carried out to trypetid sample image, and converts the true color image after gaussian filtering to ash Image is spent, trypetid profile in image is extracted by Canny algorithms；
Y2, the dipteron edge line that trypetid in trypetid profile is detected using Hough transform algorithm；
Y3, it detects and successfully then returns to straight line angle, detection failure then returns to straight-line detection failure flags；
Y4, according to straight line angular adjustment picture direction, make in picture trypetid that head-up vertical figure be presented.
In step S3, the HSV color spaces carry out filter value processing to image, and process is as follows：
Z1, the picture after Hough transform is transformed into hsv color space, and carries out the channel H, S, V filter value, conversion is such as Under：
V=max (R, G, B),
If H ＜ 0, enable H=H+360, ensure in output, 0≤V≤1,0≤S≤1,0≤H≤360；
Wherein, R is the color of red channel in rgb color space, and G is the color in rgb color space Green channel, B For the color of blue channel in rgb color space；V is the brightness of color in HSV color spaces, and H is color in HSV color spaces Tone, S be HSV color spaces in color saturation degree；
Z2, by taking the confidence interval of confidence level 0.95 to be used as threshold value respectively in the channels H, channel S and the channels V, screen Go out comprising the range including scultellum region；
Z3, intersection is done to three channels, threshold value 31~50 is taken to the channels H, obtains figure a；To channel S take threshold value 130~ 245, obtain figure b；Threshold value 110~250 is taken to obtain figure c in the channels V.Figure a, figure b, figure c are overlapped, retains and is deposited in three figures Pixel value to get to the intersection image after three channel filter values, filter value processing is carried out to high luminance pixels, is locked in picture The scultellum region of trypetid.
4 kinds of characterization factors are：Center offset D of the center striped in thoracic dorsal version region；Center striped and thoracic dorsal The length-width ratio R in version regionxAnd Ry；The area ratio S of center striped and thoracic dorsal version regions。
Center offset D of the center striped in thoracic dorsal version region is defined as fringe center point and thoracic dorsal version central point Euclidean distance, i.e. offset of the center striped in thoracic dorsal version region：
Wherein, X1And Y1Centered on striped middle point coordinates, X2And Y2For the middle point coordinates of chest backplane region；
The length-width ratio R of the center striped and thoracic dorsal version regionxAnd RyCentered on the length of striped, fineness, while with The length and width in thoracic dorsal version region makees object of reference：
Wherein, L1Centered on striped length, W1Centered on width of fringe, L2For the length and width in thoracic dorsal version region, W2For thoracic dorsal The width in version region；
The area ratio S of the center striped and thoracic dorsal version regionsFor the size of striped area, and in thoracic dorsal version region Duty ratio：
Wherein, S1Centered on striped area, S2For the area in thoracic dorsal version region；
In step S5, the BP neural network is a kind of Multi-layered Feedforward Networks trained by error backpropagation algorithm, is built Vertical BP neural network process is as follows：
When working signal forward direction is transmitted：If input layer is Ik, hidden layer neuron Hi, output layer neuron For Oj.Values of the feature vector P as input layer, the number of neuron are consistent with the dimension of feature vector；For hidden Containing layer and output layer, the output of each neuron is according to the interaction of preceding layer neuron and this layer of neuron and then by swashing Function living determines that computational methods are:
Wherein, F (x1)、F(x2) belong to activation primitive F (x), x1、x2Between expression participates in during activation primitive calculates Value, F (x) are activation primitive, two-dimentional weight vectors of the W between preceding layer and current layer neuron, WkjIndicate input layer with it is hidden A value, W are arranged containing the row k in the weight vector between layer i-thijIndicate the in the weight vector between hidden layer and output layer I row jth row value, biIndicate the bias of i-th of hidden layer neuron, bjIndicate the bias of j-th of hidden layer neuron, m For input layer number, n is hidden layer neuron number；
When error signal direction is transmitted：The weights and network of network are constantly adjusted using Widrow-Hoff learning rules Neural bias makes the error of sum square functional value of network reach minimum, and error function is defined as follows：
Wherein, E (W, b) is error of sum square function, and W is neural network weight vector, and b is BP neural network bias； yjIndicate the reality output of output layer neuron；TjIndicate the desired output of output layer neuron；
According to gradient descent method, the correction value of weighted vector is proportional to the gradient of error current function, if preceding layer is neural The output valve of member is Xi, then have：
Compared with prior art, the present invention having the following advantages that and advantageous effect：
1, the present invention is using the dipteron edge of trypetid and thoracic dorsal lath line morphological feature as classification foundation, in target image matter Measure it is poor in the case of can still reach good feature extraction effect.
2, the present invention is using shape (Hough transform detects wing straight line) and color (HSV color spaces lock scultellum area) phase In conjunction with mode, realize that computer is full automatic to lock fruit fly characteristic area, efficiency and accuracy rate are higher.
3, the present invention is based on BP neural networks to build identification model, and model parameter is obtained after mass data is trained, It only needs to import trained parameter in identification process, in conjunction with the feature extraction in the present invention and is digitized processing module, Batch identification can be carried out to target image, efficiency and accuracy rate disclosure satisfy that requirement of real time.
Description of the drawings
Fig. 1 is that the present invention is based on the image processing flow figures of the trypetid adult image-recognizing method of neural network.
Fig. 2 is that the present invention is based on the neural network models of the trypetid adult image-recognizing method of neural network to train flow Figure.
Fig. 3 is that the present invention is based on the neural network model identification process of the trypetid adult image-recognizing method of neural network Figure.
Specific implementation mode
Present invention will now be described in further detail with reference to the embodiments and the accompanying drawings, but embodiments of the present invention are unlimited In this.
Trypetid adult image-recognizing method based on neural network, includes the following steps：
The first step：Obtain trypetid sample image.
Second step：Gaussian filtering process is carried out to trypetid sample image, then using Hough transform to the reality after gaussian filtering Fly sample image is handled, as shown in Figure 1, modulation image trypetid dipteron linear position, makes trypetid in image rotate to be body Form upward, while limiting striped effective coverage；Hough transform process is as follows：
Y1, it converts the true color image after gaussian filtering to gray level image, trypetid in image is extracted by Canny algorithms Profile；
Y2, the dipteron edge line that trypetid in trypetid profile is detected using Hough transform；
Y3, it detects and successfully then returns to straight line angle, detection failure then returns to straight-line detection failure flags；
Y4, according to straight line angular adjustment picture direction, make in picture trypetid that head-up vertical figure be presented.
Third walks：Filter value processing is carried out to image using HSV color spaces, locks the scultellum region in the middle part of trypetid；HSV colors Color space carries out filter value processing to image, and process is as follows：
Picture after Hough transform is transformed into hsv color space, and carries out the channel H, S, V filter value, conversion is as follows：
V=max (R, G, B),
If H ＜ 0, enable H=H+360, ensure in output, 0≤V≤1,0≤S≤1,0≤H≤360；
Wherein, R is the color of red channel in rgb color space, and G is the color in rgb color space Green channel, B For the color of blue channel in rgb color space；V is the brightness of color in HSV color spaces, and H is color in HSV color spaces Tone, S be HSV color spaces in color saturation degree；RGB respectively has 256 grades of brightness, RGB value ranges to be：0≤R≤255, 0≤G≤255,0≤B≤255, tone H：It is measured with angle, value range is 0 °~360 °, by side counterclockwise since red To calculating, red is 0 °, and green is 120 °, and blue is 240 °.Their complementary color is：Yellow is 60 °, and cyan is 180 °, pinkish red It is 300 °；Saturation degree S：Value range is 0.0~1.0；Brightness V：Value range is 0.0 (black)~1.0 (white)；
By taking the confidence interval of confidence level 0.95 to be used as threshold value respectively in the channels H, channel S and the channels V, packet is filtered out Range including region containing scultellum；Intersection is done to three channels, filter value processing is carried out to high luminance pixels, extracts the pixel of scultellum Region locks successfully to lock the scultellum region of trypetid in picture, then carries out in next step；Success is not locked, then returns to area Domain locks failure flags；
Intersection process is as follows：Original image is transformed into HSV color spaces, does filter value processing three times respectively：To the channels H Threshold value (31,50) is taken, figure a is obtained；Threshold value (130,245) is taken to channel S, obtains figure b；Threshold value (110,250) is taken to obtain in the channels V To figure c.Figure a, figure b, figure c are overlapped, are retained in three images equal existing pixel value, you can obtain three channel filters and be worth it Intersection image afterwards.
4th step：After locking successfully, dipteron straight line is detected by Hough transform, limits the left and right range of chest backplane region And top boundary, scultellum region is detected by HSV color space filter values, limits the bottom boundary of chest backplane region, the two knot The range for closing lockable chest backplane region, the circumscribed rectangular region of thoracic dorsal plate is detached from trypetid entirety；And to the region Center striped is further processed, and according to the description method of center striped shape feature, defines 4 kinds of characterization factors of trypetid form, And according to characterization factor, extract feature vector.
4 kinds of characterization factors are：Center offset D of the center striped in thoracic dorsal version region；Center striped and thoracic dorsal version region Length-width ratio RxAnd Ry；The area ratio S of center striped and thoracic dorsal version regions；
Center offset D of the center striped in thoracic dorsal version region is defined as the European of fringe center point and thoracic dorsal version central point Distance, i.e. offset of the center striped in thoracic dorsal version region：
Wherein, X1And Y1Centered on striped middle point coordinates, X2And Y2For the middle point coordinates of chest backplane region；
The length-width ratio R of center striped and thoracic dorsal version regionxAnd RyCentered on the length of striped, fineness, while with thoracic dorsal The length and width in version region makees object of reference：
Wherein, L1Centered on striped length, W1Centered on width of fringe, L2For the length and width in thoracic dorsal version region, W2For thoracic dorsal The width in version region；
The area ratio S of center striped and thoracic dorsal version regionsFor the size of striped area, and the duty in thoracic dorsal version region Than：
Wherein, S1Centered on striped area, S2For the area in thoracic dorsal version region；
5th step：Clarification of objective vector in a certain number of trypetid images is acquired, establishes BP neural network to data set It is trained, as shown in Fig. 2, to obtain for fruit-fly classified neural network model parameter；Wherein acquisition threshold value is K, when When collecting quantity is more than K, then carry out in next step, when collecting quantity is less than K, then returning to step S1；
BP neural network is a kind of Multi-layered Feedforward Networks trained by error backpropagation algorithm, establishes BP neural network mistake Journey is as follows：
When working signal forward direction is transmitted：If input layer is Ik, hidden layer neuron Hi, output layer neuron For Oj.Values of the feature vector P as input layer, the number of neuron are consistent with the dimension of feature vector；For hidden Containing layer and output layer, the output of each neuron is according to the interaction of preceding layer neuron and this layer of neuron and then by swashing Function living determines that computational methods are:
Wherein, F (x1)、F(x2) belong to activation primitive F (x), x1、x1It indicates to participate in the centre during activation primitive calculates Value, two-dimentional weight vectors of the W between preceding layer and current layer neuron, WkjIndicate weights between input layer and hidden layer to Row k i-th in amount arranges a value, WijIndicate the i-th row jth row value in the weight vector between hidden layer and output layer, bi Indicate the bias of i-th of hidden layer neuron, bjIndicate that the bias of j-th of hidden layer neuron, m are input layer Number, n are hidden layer neuron number；
When error signal direction is transmitted：The weights and network of network are constantly adjusted using Widrow-Hoff learning rules Neural bias makes the error of sum square functional value of network reach minimum, and error function is defined as follows：
Wherein, E (W, b) is error of sum square function, and W is neural network weight vector, and b is the biasing of Current neural member Value,；yjIndicate the reality output of output layer neuron；TjIndicate the desired output of output layer neuron；
According to gradient descent method, the correction value of weighted vector is proportional to the gradient of error current function, if preceding layer is neural The output valve of member is Xi, then have：
Neural network model parameter is read, identification model is built；The present invention builds using three-layer neural network and identifies Model can complete mapping of the arbitrary M dimension to N-dimensional, and input layer includes 4 neurons, in corresponding training set data feature to The input form of the dimension of amount, neural network is vector [D, Rx, Ry, Ss]；Output layer includes 3 neurons, and the corresponding model needs Identify that the classification sum of target, that is, the classification sum for needing exist for identification target are three kinds：Citrus fruit fly, pumpkin fruit fly, melon Trypetid, activation primitive select Sigmoid functions, i.e.,A is constant, and value is 1 in experiment.
6th step：The trypetid image identified will be needed to carry out second step to the 4th step to handle, as shown in figure 3, obtaining feature Vector then carries out in next step；Feature vector is not obtained, then records failure information, identifies recognition failures；The reality that needs are identified Fly image feature vector inputs identification model, is identified, and identification model exports recognition result；
The above embodiment is a preferred embodiment of the present invention, but embodiments of the present invention are not by above-described embodiment Limitation, it is other it is any without departing from the spirit and principles of the present invention made by changes, modifications, substitutions, combinations, simplifications, Equivalent substitute mode is should be, is included within the scope of the present invention.

Claims (10)

1. the trypetid adult image-recognizing method based on neural network, which is characterized in that include the following steps：
S1, trypetid sample image is obtained；
S2, trypetid sample image is handled using Hough transform, trypetid sample graph is adjusted according to trypetid dipteron linear position Picture makes trypetid in trypetid sample image rotate to be body form upward；
S3, filter value processing is carried out to trypetid sample image using HSV color spaces, locks the trypetid scultellum region in the middle part of trypetid；
S4, according to trypetid dipteron linear position and trypetid scultellum region, the thoracic dorsal plate for locking trypetid is trypetid characteristic area, trypetid Characteristic area defines 4 kinds of characterization factors of trypetid form, and according to 4 kinds of spies according to the description method of center striped shape feature The factor is levied, is formed and is extracted as a feature vector；
S5, acquisition trypetid sample image in feature vector, formed characteristic vector data collection, establish BP neural network to feature to Amount data set is trained, and is obtained for fruit-fly classified neural network model parameter, based on neural network model parameter, Build identification model；Wherein acquisition threshold value is K, when collecting quantity is more than K, is then carried out in next step, when collecting quantity is less than K When, then return to step S1；
S6, S2-S4 step process is carried out to the trypetid image that needs identify, obtains feature vector, feature vector is inputted and is identified Model, identification model export recognition result.
2. the trypetid adult image-recognizing method according to claim 1 based on neural network, which is characterized in that step S1 In, the trypetid sample image is the back side towards epigraph.
3. the trypetid adult image-recognizing method according to claim 1 based on neural network, which is characterized in that step S2 In, the Hough transform process is as follows：
Y1, gaussian filtering process is carried out to trypetid sample image, and converts the true color image after gaussian filtering to gray-scale map Picture extracts trypetid profile in trypetid sample image by Canny algorithms；
Y2, the dipteron edge line that trypetid in trypetid profile is detected using Hough transform algorithm；
Y3, it detects and successfully then returns to straight line angle, detection failure then returns to straight-line detection failure flags；
Y4, according to straight line angular adjustment trypetid sample image direction, so that trypetid is presented head-up vertical Figure.
4. the trypetid adult image-recognizing method according to claim 1 based on neural network, which is characterized in that step S3 In, the HSV color spaces carry out filter value processing to trypetid sample image, and process is as follows：
Z1, the picture after Hough transform is transformed into hsv color space, and carries out the channel H, S, V filter value, conversion is as follows：
V=max (R, G, B),
If H ＜ 0, enable H=H+360, ensure in output, 0≤V≤1,0≤S≤1,0≤H≤360；
Wherein, R is the color of red channel in rgb color space, and G is the color in rgb color space Green channel, B RGB The color of blue channel in color space；V is the brightness of color in HSV color spaces, and H is the color of color in HSV color spaces It adjusts, S is the saturation degree of color in HSV color spaces；
Z2, by taking the confidence interval of confidence level 0.95 to be used as threshold value respectively in the channels H, channel S and the channels V, filter out packet Range including the region of scultellum containing trypetid；
Z3, intersection is done to three channels：Threshold value 31~50 is taken to the channels H, obtains figure a；Threshold value 130~245 is taken to channel S, is obtained To figure b；Threshold value 110~250 is taken to obtain figure c in the channels V；Figure a, figure b, figure c are overlapped, existing picture in three figures is retained Element value carries out filter value processing to high luminance pixels, locks trypetid shield in picture to get to the intersection image after three channel filter values Panel region.
5. the trypetid adult image-recognizing method according to claim 1 based on neural network, which is characterized in that step S4 In, the description method of the center striped shape feature defines 4 kinds of characterization factors：Center of the center striped in thoracic dorsal version region Offset D；The length-width ratio R of center striped and thoracic dorsal version regionxAnd Ry；The area ratio S of center striped and thoracic dorsal version regions。
6. the trypetid adult image-recognizing method according to claim 5 based on neural network, which is characterized in that in described Center offset D of the heart striped in thoracic dorsal version region is defined as the Euclidean distance of fringe center point and thoracic dorsal version central point, i.e., in Offset of the heart striped in thoracic dorsal version region：
Wherein, X1And Y1Centered on striped middle point coordinates, X2And Y2For the middle point coordinates of chest backplane region；
The length-width ratio R of the center striped and thoracic dorsal version regionxAnd RyCentered on the length of striped, fineness, while with thoracic dorsal The length and width in version region makees object of reference：
Wherein, L1Centered on striped length, W1Centered on width of fringe, L2For the length and width in thoracic dorsal version region, W2For thoracic dorsal version area The width in domain；
The area ratio S of the center striped and thoracic dorsal version regionsFor the size of striped area, and the duty in thoracic dorsal version region Than：
Wherein, S1Centered on striped area, S2For the area in thoracic dorsal version region.
7. the trypetid adult image-recognizing method according to claim 1 based on neural network, which is characterized in that step S5 In, the BP neural network is a kind of Multi-layered Feedforward Networks by error backpropagation algorithm training；The BP neural network Identification model is built using three-layer neural network：Input layer, hidden layer, output layer；Input layer includes 4 neurons, corresponding instruction Practice the dimension of feature vector in collection data, the input form of neural network is vector [D, Rx, Ry, Ss]；Output layer includes 3 god Through member, the corresponding model needs to identify that the classification sum of target, activation primitive select Sigmoid functions, i.e.,：
Wherein, F (x) indicates activation primitive, and e is natural constant, e-xFor the mathematic parameter in Sigmoid functions, A is constant.
8. the trypetid adult image-recognizing method according to claim 7 based on neural network, which is characterized in that step S5 In, detailed process is as follows：
When working signal forward direction is transmitted：If input layer is Ik, hidden layer neuron Hi, output layer neuron is Oj； Values of the feature vector P as input layer, i.e. PkFor k-th of value of feature vector, IkFor the kth in input layer A neuron, the number of neuron are consistent with the dimension of feature vector；For hidden layer and output layer, each neuron it is defeated Go out and then determines that computational methods are by activation primitive according to the interaction of preceding layer neuron and this layer of neuron:
Wherein, F (x1)、F(x2) belong to activation primitive F (x), x1、x2For the median for participating in during activation primitive calculates, W is Two-dimentional weight vector between preceding layer and current layer neuron, WkjIt indicates in the weight vector between input layer and hidden layer Row k i-th arranges a value, WijIndicate the i-th row jth row value in the weight vector between hidden layer and output layer, biIndicate i-th The bias of a hidden layer neuron, bjIndicate that the bias of j-th of hidden layer neuron, m are input layer number, n For hidden layer neuron number；
When error signal direction is transmitted：Constantly adjusted using Widrow-Hoff learning rules BP neural network weight vector and Network neural bias makes the error of sum square functional value of BP neural network reach minimum, and error function is defined as follows：
Wherein, E (W, b) is error of sum square function, and W is BP neural network weight vector, and b is BP neural network bias；yjTable Show the reality output of output layer neuron；TjIndicate the desired output of output layer neuron.
9. the trypetid adult image-recognizing method according to claim 1 based on neural network, which is characterized in that step S5 In, the identification model can be to needing to identify the automatic locking that image carries out characteristic area.
10. the trypetid adult image-recognizing method according to claim 9 based on neural network, which is characterized in that described Identification model can carry out batch processing to the image that needs identify.
CN201810087133.5A 2018-01-30 2018-01-30 Trypetid adult image-recognizing method based on neural network Pending CN108280483A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201810087133.5A CN108280483A (en) 2018-01-30 2018-01-30 Trypetid adult image-recognizing method based on neural network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201810087133.5A CN108280483A (en) 2018-01-30 2018-01-30 Trypetid adult image-recognizing method based on neural network

Publications (1)

Publication Number Publication Date
CN108280483A true CN108280483A (en) 2018-07-13

Family

ID=62805652

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201810087133.5A Pending CN108280483A (en) 2018-01-30 2018-01-30 Trypetid adult image-recognizing method based on neural network

Country Status (1)

Country Link
CN (1) CN108280483A (en)

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109508721A (en) * 2018-11-08 2019-03-22 烟台职业学院 The realization of figure identification and color identification based on OpenCV
CN110516686A (en) * 2019-07-11 2019-11-29 西安电子科技大学 The mosquito recognition methods of three color RGB images
CN111354046A (en) * 2020-03-30 2020-06-30 北京芯龙德大数据科技有限公司 Indoor camera positioning method and positioning system
CN111723709A (en) * 2020-06-09 2020-09-29 大连海事大学 Fly face recognition method based on deep convolutional neural network
CN112036470A (en) * 2020-08-28 2020-12-04 扬州大学 Cloud transmission-based multi-sensor fusion cucumber bemisia tabaci identification method

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP2359257A1 (en) * 2008-11-07 2011-08-24 General Electric Company Systems and methods for automated extraction of high-content information from whole organisms
WO2017037875A1 (en) * 2015-09-01 2017-03-09 オリンパス株式会社 Method for acquiring 3-dimensional image, method for displaying 3-dimensional image, and device for acquiring 3-dimensional image
CN107463959A (en) * 2017-08-05 2017-12-12 国网江西省电力公司电力科学研究院 A kind of fruit fly recognition methods based on BP neural network

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP2359257A1 (en) * 2008-11-07 2011-08-24 General Electric Company Systems and methods for automated extraction of high-content information from whole organisms
WO2017037875A1 (en) * 2015-09-01 2017-03-09 オリンパス株式会社 Method for acquiring 3-dimensional image, method for displaying 3-dimensional image, and device for acquiring 3-dimensional image
CN107463959A (en) * 2017-08-05 2017-12-12 国网江西省电力公司电力科学研究院 A kind of fruit fly recognition methods based on BP neural network

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
李震 等: "基于神经网络的实蝇成虫图像识别算法", 《农业机械学报》 *

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109508721A (en) * 2018-11-08 2019-03-22 烟台职业学院 The realization of figure identification and color identification based on OpenCV
CN110516686A (en) * 2019-07-11 2019-11-29 西安电子科技大学 The mosquito recognition methods of three color RGB images
CN110516686B (en) * 2019-07-11 2021-03-23 西安电子科技大学 Mosquito recognition method of three-color RGB image
CN111354046A (en) * 2020-03-30 2020-06-30 北京芯龙德大数据科技有限公司 Indoor camera positioning method and positioning system
CN111723709A (en) * 2020-06-09 2020-09-29 大连海事大学 Fly face recognition method based on deep convolutional neural network
CN111723709B (en) * 2020-06-09 2023-07-11 大连海事大学 Fly face recognition method based on deep convolutional neural network
CN112036470A (en) * 2020-08-28 2020-12-04 扬州大学 Cloud transmission-based multi-sensor fusion cucumber bemisia tabaci identification method

Similar Documents

Publication Publication Date Title
CN108280483A (en) Trypetid adult image-recognizing method based on neural network
CN106204779B (en) Check class attendance method based on plurality of human faces data collection strategy and deep learning
CN104598908B (en) A kind of crops leaf diseases recognition methods
Kulkarni et al. Applying image processing technique to detect plant diseases
CN109635875A (en) A kind of end-to-end network interface detection method based on deep learning
CN109522889A (en) Hydrological ruler water level identification and estimation method based on image analysis
CN104835175B (en) Object detection method in a kind of nuclear environment of view-based access control model attention mechanism
CN107330892A (en) A kind of sunflower disease recognition method based on random forest method
CN108681692A (en) Increase Building recognition method in a kind of remote sensing images based on deep learning newly
CN108319973A (en) Citrusfruit detection method on a kind of tree
CN109657632A (en) A kind of lane detection recognition methods
CN112906550B (en) Static gesture recognition method based on watershed transformation
CN102915446A (en) Plant disease and pest detection method based on SVM (support vector machine) learning
CN107944399A (en) A kind of pedestrian's recognition methods again based on convolutional neural networks target's center model
CN108537239A (en) A kind of method of saliency target detection
CN103971126A (en) Method and device for identifying traffic signs
CN107633229A (en) Method for detecting human face and device based on convolutional neural networks
CN107767379A (en) Pcb board marks print quality inspection method
CN106934386A (en) A kind of natural scene character detecting method and system based on from heuristic strategies
CN107392929A (en) A kind of intelligent target detection and dimension measurement method based on human vision model
CN108154147A (en) The region of interest area detecting method of view-based access control model attention model
CN108388905A (en) A kind of Illuminant estimation method based on convolutional neural networks and neighbourhood context
CN114067207A (en) Vegetable seedling field weed detection method based on deep learning and image processing
CN109886146B (en) Flood information remote sensing intelligent acquisition method and device based on machine vision detection
CN108416814A (en) Quick positioning and recognition methods and the system on a kind of pineapple head

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: 20180713