CN104077580A - Pest image automatic recognition method based on high-reliability network - Google Patents
Pest image automatic recognition method based on high-reliability network Download PDFInfo
- Publication number
- CN104077580A CN104077580A CN201410337568.2A CN201410337568A CN104077580A CN 104077580 A CN104077580 A CN 104077580A CN 201410337568 A CN201410337568 A CN 201410337568A CN 104077580 A CN104077580 A CN 104077580A
- Authority
- CN
- China
- Prior art keywords
- training
- image
- belief network
- training sample
- layer
- 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
Landscapes
- Image Analysis (AREA)
Abstract
The invention provides a pest image automatic recognition method based on a high-reliability network. The method includes the following steps that preprocessing is carried out on multiple collected training images to obtain multiple training samples, and HOG feature extraction is conducted on the training samples to form joint image feature vectors of the training samples; the high-reliability network based on a restricted boltzmann machine is constructed, the joint image feature vectors of the training samples are input to the constructed high-reliability network, and training on the high-reliability network is completed; preprocessing is conducted on pest images to be tested to obtain test samples, and HOG feature extraction is carried out on the test samples to form joint image feature vectors of the test samples; the joint image feature vectors of the test samples are input to the high-reliability network after training is finished, and categories of past images to be tested are obtained through recognition. According to the method, the accuracy rate of pest recognition can be improved, and robustness of a pest recognition algorithm is enhanced.
Description
Technical field
The present invention relates to reading intelligent agriculture and mode identification technology, specifically a kind of insect automatic distinguishing method for image based on dark belief network.
Background technology
Insect is the formidable enemy in crop growth, within crops whole growth periods, has generation, can cause a large amount of underproduction of crops.The classification of existing insect, identification work mainly rely on minority plant protection expert and agriculture technical staff to complete, but pest species is various, and each plant protection expert poor its can also can only identification division insect.Increasing sign shows, increasing with the relative less contradiction of insect identification expert of insect identification demand increasingly sharpened.But existing insect automatic distinguishing method for image and system recognition rate are not high, robustness is poor, is only present in the experimental phase.Therefore, seek that a kind of discrimination is high, the insect automatic distinguishing method for image of strong robustness has very important significance.At area of pattern recognition, become the focus of numerous scholar's research based on the unsupervised degree of depth theories of learning now, be widely used in recognition of face, object identification field, and obtained good effect.
Summary of the invention
The object of the present invention is to provide that a kind of discrimination is high, the insect automatic distinguishing method for image based on dark belief network of strong robustness.
Technical scheme of the present invention is:
An insect automatic distinguishing method for image based on dark belief network, comprises the following steps:
(1) some training images of collecting are carried out to pre-service, obtain some training samples, training sample is carried out to HOG feature extraction, form the joint image proper vector of training sample;
(2) the dark belief network of structure based on limited Boltzmann machine, by the dark belief network of the joint image proper vector input structure of training sample, completes the training to dark belief network;
(3) insect image to be measured is carried out to pre-service, obtain test sample book, test sample book is carried out to HOG feature extraction, form the joint image proper vector of test sample book;
(4) the dark belief network of the input of the joint image proper vector of test sample book having been trained, identification obtains the classification of insect image to be measured.
The described insect automatic distinguishing method for image based on dark belief network, described step (1) specifically comprises:
(11) be 144 × 144 by the size normalization of all training images;
(12) training image after normalization is carried out to gray processing, gray balance and filtering processing, obtain training sample;
(13) training sample is divided into mutually overlapping piece, makes each to have 50% area overlapped with adjacent piece;
(14) adopt piece be of a size of 16 × 16 and the intrinsic dimensionality of each be that the HOG operator of 36 dimensions carries out feature extraction to training sample, obtaining intrinsic dimensionality is the HOG feature histogram of 10404 dimensions;
(15) connect with its HOG feature histogram after stretching training sample, form the joint image proper vector of training sample, the dimension of this joint image proper vector is 31140 dimensions.
The described insect automatic distinguishing method for image based on dark belief network, described step (2) specifically comprises:
(21) the five layer depth belief networks of structure based on limited Boltzmann machine, comprise an input layer, three hidden layers and an output layer, set the nodes of each layer, wherein, the nodes of input layer is consistent with the dimension of the joint image proper vector of training sample, and the nodes of output layer is consistent with the classification number of image to be classified;
(22) adopt sdpecific dispersion algorithm is carried out to successively greedy training to input layer and three hidden layers, calculate the output valve of weights and biasing and three hidden layers of each layer;
(23) adopt softmax regression model to train output layer;
(24) adopt back-propagation algorithm to adjust whole dark belief network, optimize the parameter of dark belief network, complete overall situation training.
The dark belief network of the present invention's structure has multiple hidden layers, have the feature representation ability more excellent than shallow-layer network, original image and feature histogram are combined as the input data of dark belief network, retaining the statistical information of having given prominence to Local gradient direction on the basis of picture appearance information; The present invention, for the insect image of illumination, change of background, still can obtain good classification performance, has improved the accuracy rate of insect identification, has strengthened the robustness of insect recognizer, has reached actual application level.
Brief description of the drawings
Fig. 1 is method flow diagram of the present invention.
Embodiment
Further illustrate the present invention with specific embodiment by reference to the accompanying drawings below.
As shown in Figure 1, a kind of insect automatic distinguishing method for image based on dark belief network, comprises the following steps:
S1, collect some width images as training image, all training images carried out to pre-service, obtain several training samples, comprise the following steps:
S11, be 144 × 144 by the size normalization of every width training image.
S12, by the training image gray processing after normalization, and by the training image gray balance after gray processing.
S13, training image after adopting gaussian filtering algorithm to gray balance carry out smoothing processing, eliminate the impact of noise on training image quality.
S2, training sample is carried out to feature extraction, comprises the following steps:
S21, employing rectangular orientation histogram of gradients HOG operator are to each training sample I
gcarry out feature extraction, wherein, the piece (block) of HOG operator is of a size of 16 × 16, each is divided into nonoverlapping 4 unit (cell), the gradient direction (scope is 0-180 °) obtaining in each unit is merged into 9 intervals, 20 ° is an interval, and the intrinsic dimensionality of each unit is 9 dimensions, and the intrinsic dimensionality of each is 4*9=36 dimension.
S22, by each training sample I
gbe divided into mutually overlapping piece, each has 50% area overlapped with adjacent piece, each training sample I
gin there are ((144-8)/8) * ((144-8)/8)=289 pieces, therefore, each training sample I
gthe HOG feature histogram H obtaining
gintrinsic dimensionality be 36*289=10404 dimension.
S23, by training sample I
gstretching rear and its HOG feature histogram H
gseries connection obtains training sample I
gjoint image proper vector v
i, be a 144*144+10404=31140 dimensional vector.
S3, structure degree of deeply convinceing network C
dBN, comprise the following steps:
S31, consider insect recognition accuracy and training time expense, the 5 layer depth belief network Cs of structure based on limited Boltzmann machine RBM
dBN, comprise 1 input layer (visual layers), 3 hidden layers and 1 output layer (classification layer).
S32, appointment C
dBNinput layer number be 31140, the nodes of first hidden layer is that the nodes of 500, the second hidden layers is that the nodes of 500, the three hidden layers is 2000, the classification number that output layer nodes is image to be classified.
S4, training degree of deeply convinceing network C
dBN, comprise the following steps:
S41, start to carry out the training of first RBM (31140-500), random initializtion model parameter θ={ w, a, b}, wherein, w represents input layer weights, a represents input layer biasing, and b represents first hidden layer biasing, and sets the learning rate λ of three parameters
w=λ
a=λ
b=0.1.
S42, input v to input layer
i 0(v herein
i 0be training sample I
gjoint image proper vector v
i) carry out forward-propagating, calculate the output h of first hidden layer
j 0.
S43, output h to first hidden layer
j 0carry out backpropagation, obtain v
i 1.
S44, equally to v
i 1carry out forward-propagating, obtain h
j 1.
S45, in conjunction with learning rate corresponding to each parameter, Renewal model parameter θ=w, a, b}, the variable quantity of each parameter is:
△w
ij=λ
w(E[v
i 0h
j 0]-E[v
i 1h
j 1])
△a
i=λ
a(E[v
i 0]-E[v
i 1])
△b
j=λ
b(E[h
j 0]-E[h
j 1])
Wherein E[] represent to ask for mathematical expectation.
The input v of S46, change input layer
i 0, repeat S42 to S45, until convergence.
S47, input using the output of first RBM as second RBM (500-500), utilize above-mentioned training method to complete the training of second RBM; Complete the training of the 3rd RBM with same method; By stacking three RBM, form C
dBNfront four-layer network network.
S48, employing softmax regression model training C
dBNoutput layer.
S49, employing backpropagation BP algorithm are to whole C
dBNadjust, optimize C
dBNparameter, complete C
dBNoverall situation training.
S5, insect image to be measured is classified, comprises the following steps:
S51, employing mobile phone or camera are taken the insect image in field;
S52, insect image to be measured, after normalization, gray processing, gray balance, smoothing processing, obtain the test sample book with training sample of the same size 144 × 144.
The HOG feature of S53, extraction test sample book, forms joint image proper vector v
t(concrete steps are with the HOG feature extraction of training sample), by the joint image proper vector v of test sample book
tthe C that input has been trained
dBNin, carry out the automatic identification of insect image.
The above embodiment is only that the preferred embodiment of the present invention is described; not scope of the present invention is limited; design under the prerequisite of spirit not departing from the present invention; various distortion and improvement that those of ordinary skill in the art make technical scheme of the present invention, all should fall in the definite protection domain of claims of the present invention.
Claims (3)
1. the insect automatic distinguishing method for image based on dark belief network, is characterized in that, comprises the following steps:
(1) some training images of collecting are carried out to pre-service, obtain some training samples, training sample is carried out to HOG feature extraction, form the joint image proper vector of training sample;
(2) the dark belief network of structure based on limited Boltzmann machine, by the dark belief network of the joint image proper vector input structure of training sample, completes the training to dark belief network;
(3) insect image to be measured is carried out to pre-service, obtain test sample book, test sample book is carried out to HOG feature extraction, form the joint image proper vector of test sample book;
(4) the dark belief network of the input of the joint image proper vector of test sample book having been trained, identification obtains the classification of insect image to be measured.
2. the insect automatic distinguishing method for image based on dark belief network according to claim 1, is characterized in that, described step (1) specifically comprises:
(11) be 144 × 144 by the size normalization of all training images;
(12) training image after normalization is carried out to gray processing, gray balance and filtering processing, obtain training sample;
(13) training sample is divided into mutually overlapping piece, makes each to have 50% area overlapped with adjacent piece;
(14) adopt piece be of a size of 16 × 16 and the intrinsic dimensionality of each be that the HOG operator of 36 dimensions carries out feature extraction to training sample, obtaining intrinsic dimensionality is the HOG feature histogram of 10404 dimensions;
(15) connect with its HOG feature histogram after stretching training sample, form the joint image proper vector of training sample, the dimension of this joint image proper vector is 31140 dimensions.
3. the insect automatic distinguishing method for image based on dark belief network according to claim 1, is characterized in that, described step (2) specifically comprises:
(21) the five layer depth belief networks of structure based on limited Boltzmann machine, comprise an input layer, three hidden layers and an output layer, set the nodes of each layer, wherein, the nodes of input layer is consistent with the dimension of the joint image proper vector of training sample, and the nodes of output layer is consistent with the classification number of image to be classified;
(22) adopt sdpecific dispersion algorithm is carried out to successively greedy training to input layer and three hidden layers, calculate the output valve of weights and biasing and three hidden layers of each layer;
(23) adopt softmax regression model to train output layer;
(24) adopt back-propagation algorithm to adjust whole dark belief network, optimize the parameter of dark belief network, complete overall situation training.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410337568.2A CN104077580A (en) | 2014-07-15 | 2014-07-15 | Pest image automatic recognition method based on high-reliability network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410337568.2A CN104077580A (en) | 2014-07-15 | 2014-07-15 | Pest image automatic recognition method based on high-reliability network |
Publications (1)
Publication Number | Publication Date |
---|---|
CN104077580A true CN104077580A (en) | 2014-10-01 |
Family
ID=51598828
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201410337568.2A Pending CN104077580A (en) | 2014-07-15 | 2014-07-15 | Pest image automatic recognition method based on high-reliability network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN104077580A (en) |
Cited By (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104484684A (en) * | 2015-01-05 | 2015-04-01 | 苏州大学 | Handwriting identifying method and system |
CN104636732A (en) * | 2015-02-12 | 2015-05-20 | 合肥工业大学 | Sequence deeply convinced network-based pedestrian identifying method |
CN105447441A (en) * | 2015-03-19 | 2016-03-30 | 北京天诚盛业科技有限公司 | Face authentication method and device |
CN105844333A (en) * | 2016-03-18 | 2016-08-10 | 厦门大学 | Immunity chromatography test strip quantitation detection method based on deep reliability network |
CN106408037A (en) * | 2015-07-30 | 2017-02-15 | 阿里巴巴集团控股有限公司 | Image recognition method and apparatus |
CN106548155A (en) * | 2016-10-28 | 2017-03-29 | 安徽四创电子股份有限公司 | A kind of detection method of license plate based on depth belief network |
CN107194418A (en) * | 2017-05-10 | 2017-09-22 | 中国科学院合肥物质科学研究院 | A kind of Aphids in Rice Field detection method based on confrontation feature learning |
CN108629370A (en) * | 2018-04-28 | 2018-10-09 | 广东工业大学 | A kind of classification and identification algorithm and device based on depth confidence network |
CN111160416A (en) * | 2019-12-12 | 2020-05-15 | 杭州电子科技大学 | Aedes mosquito identification method based on support vector machine |
CN117475304A (en) * | 2023-10-24 | 2024-01-30 | 北京佳格天地科技有限公司 | Agricultural pest identification method based on artificial intelligence |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20040199482A1 (en) * | 2002-04-15 | 2004-10-07 | Wilson Scott B. | Systems and methods for automatic and incremental learning of patient states from biomedical signals |
CN102930249A (en) * | 2012-10-23 | 2013-02-13 | 四川农业大学 | Method for identifying and counting farmland pests based on colors and models |
CN103778432A (en) * | 2014-01-08 | 2014-05-07 | 南京邮电大学 | Human being and vehicle classification method based on deep belief net |
-
2014
- 2014-07-15 CN CN201410337568.2A patent/CN104077580A/en active Pending
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20040199482A1 (en) * | 2002-04-15 | 2004-10-07 | Wilson Scott B. | Systems and methods for automatic and incremental learning of patient states from biomedical signals |
CN102930249A (en) * | 2012-10-23 | 2013-02-13 | 四川农业大学 | Method for identifying and counting farmland pests based on colors and models |
CN103778432A (en) * | 2014-01-08 | 2014-05-07 | 南京邮电大学 | Human being and vehicle classification method based on deep belief net |
Cited By (17)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104484684A (en) * | 2015-01-05 | 2015-04-01 | 苏州大学 | Handwriting identifying method and system |
CN104484684B (en) * | 2015-01-05 | 2018-11-02 | 苏州大学 | A kind of Manuscripted Characters Identification Method and system |
CN104636732A (en) * | 2015-02-12 | 2015-05-20 | 合肥工业大学 | Sequence deeply convinced network-based pedestrian identifying method |
CN104636732B (en) * | 2015-02-12 | 2017-11-07 | 合肥工业大学 | A kind of pedestrian recognition method based on the deep belief network of sequence |
WO2016145940A1 (en) * | 2015-03-19 | 2016-09-22 | 北京天诚盛业科技有限公司 | Face authentication method and device |
CN105447441A (en) * | 2015-03-19 | 2016-03-30 | 北京天诚盛业科技有限公司 | Face authentication method and device |
CN105447441B (en) * | 2015-03-19 | 2019-03-29 | 北京眼神智能科技有限公司 | Face authentication method and device |
CN106408037A (en) * | 2015-07-30 | 2017-02-15 | 阿里巴巴集团控股有限公司 | Image recognition method and apparatus |
CN106408037B (en) * | 2015-07-30 | 2020-02-18 | 阿里巴巴集团控股有限公司 | Image recognition method and device |
CN105844333A (en) * | 2016-03-18 | 2016-08-10 | 厦门大学 | Immunity chromatography test strip quantitation detection method based on deep reliability network |
CN106548155A (en) * | 2016-10-28 | 2017-03-29 | 安徽四创电子股份有限公司 | A kind of detection method of license plate based on depth belief network |
CN107194418A (en) * | 2017-05-10 | 2017-09-22 | 中国科学院合肥物质科学研究院 | A kind of Aphids in Rice Field detection method based on confrontation feature learning |
CN108629370A (en) * | 2018-04-28 | 2018-10-09 | 广东工业大学 | A kind of classification and identification algorithm and device based on depth confidence network |
CN111160416A (en) * | 2019-12-12 | 2020-05-15 | 杭州电子科技大学 | Aedes mosquito identification method based on support vector machine |
CN111160416B (en) * | 2019-12-12 | 2023-03-21 | 杭州电子科技大学 | Aedes mosquito identification method based on support vector machine |
CN117475304A (en) * | 2023-10-24 | 2024-01-30 | 北京佳格天地科技有限公司 | Agricultural pest identification method based on artificial intelligence |
CN117475304B (en) * | 2023-10-24 | 2024-07-26 | 北京佳格天地科技有限公司 | Agricultural pest identification method based on artificial intelligence |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN104077580A (en) | Pest image automatic recognition method based on high-reliability network | |
CN107393542B (en) | Bird species identification method based on two-channel neural network | |
CN104143079B (en) | The method and system of face character identification | |
CN101763502B (en) | High-efficiency method and system for sensitive image detection | |
CN110097003A (en) | Check class attendance method, equipment, storage medium and device neural network based | |
CN108198154A (en) | Image de-noising method, device, equipment and storage medium | |
CN104850836A (en) | Automatic insect image identification method based on depth convolutional neural network | |
CN110728656A (en) | Meta-learning-based no-reference image quality data processing method and intelligent terminal | |
CN105512676A (en) | Food recognition method at intelligent terminal | |
CN108197669B (en) | Feature training method and device of convolutional neural network | |
CN109145717A (en) | A kind of face identification method of on-line study | |
CN114092769B (en) | Transformer substation multi-scene inspection analysis method based on federal learning | |
CN104636755A (en) | Face beauty evaluation method based on deep learning | |
CN111382808A (en) | Vehicle detection processing method and device | |
CN103971106A (en) | Multi-view human facial image gender identification method and device | |
CN104091181A (en) | Injurious insect image automatic recognition method and system based on deep restricted Boltzmann machine | |
CN110852795A (en) | Mobile phone valuation system and valuation method based on computer vision | |
CN110048978A (en) | A kind of signal modulate method | |
CN111401169A (en) | Power supply business hall service personnel behavior identification method based on monitoring video information | |
CN110852271A (en) | Micro-expression recognition method based on peak frame and deep forest | |
CN110348434A (en) | Camera source discrimination method, system, storage medium and calculating equipment | |
CN116958961A (en) | Lightweight pomegranate identification method based on improved YOLOv8s | |
CN115953643A (en) | Knowledge distillation-based model training method and device and electronic equipment | |
CN104484679B (en) | Non- standard rifle shooting warhead mark image automatic identifying method | |
CN115879817A (en) | Regional carbon reduction amount evaluation method and device, electronic equipment and storage medium |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into 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: 20141001 |