CN106778896A - A kind of Cordyceps sinensis detection method based on own coding feature learning - Google Patents
A kind of Cordyceps sinensis detection method based on own coding feature learning Download PDFInfo
- Publication number
- CN106778896A CN106778896A CN201611249294.7A CN201611249294A CN106778896A CN 106778896 A CN106778896 A CN 106778896A CN 201611249294 A CN201611249294 A CN 201611249294A CN 106778896 A CN106778896 A CN 106778896A
- Authority
- CN
- China
- Prior art keywords
- cordyceps sinensis
- coding
- model
- image
- sample
- 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.)
- Granted
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
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a kind of Cordyceps sinensis detection method based on own coding feature learning,（1）A series of images comprising Cordyceps sinensis of collection；（2）The Cordyceps sinensis and other backgrounds in image are extracted, and is fabricated to the positive negative sample of size identical；（3）Own coding model is trained by the sample image for extracting, encoding model parameter is obtained from；（4）Sample is encoded by own coding model；（5）The coding and sample class of acquisition are carried out into classification based training using Nonlinear Support Vector Machines, disaggregated model parameter is obtained；（6）Collection Cordyceps sinensis image to be detected, piecemeal is carried out by it under multiple yardsticks；（7）Each block of image is encoded using own coding model, using Nonlinear Support Vector Machines classify and record position；（8）The overlapping region that rejecting is detected, and by all non-coincidence area markings on altimetric image to be checked.Can be in the Cordyceps sinensis in automatic detection environment under complex background using the method for the present invention.
Description
Technical field
The invention belongs to artificial intelligence application technical field, specifically a kind of Cordyceps sinensis based on own coding feature learning
Detection method.
Background technology
Cordyceps sinensis also known as Chinese caterpillar fungus, are China's usual a kind of rare tonic herbs among the people, and its nutritional ingredient is higher than ginseng,
Can be used as medicine, also edible, be superior delicacies, with nutritive value very high.Cordyceps sinensis main product in Jinsha jiang River, the Lancang River,
The upstream of Nujiang Three Rivers regions.To the east of the Liangshan Mountain in Sichuan Province, west to the Burang County in Tibet, North gets the Mount Min in Gansu Province, Nan Zhixi
The refined mountain of horse traction and the Yulong Xueshan in Yunnan Province.Best Chinese caterpillar fungus is grown in moist on the sunny side, the soil of 3000~5500 meters or so of height above sea level
Under the unctuous hillside of matter, grassy marshland, bushes, and excavation cycle only one of which month or so.
The excavation mode of traditional Cordyceps sinensis is using artificial search, the method for excavation, due to the rareness of Chinese caterpillar fungus, and growth
On moist grassy marshland, hillside, background is more complicated so that it is very difficult to be found, so the search of Chinese caterpillar fungus is one non-
Often time-consuming process, common Tibetan generally more than a day it can be found that 20 or so, few only has two or three, or even can
Can be gainless.Tibetan generally using kneeling the slow method searched in ground, due to the humidity on meadow, also can cause that dig grass has to body
Very big damage.So seek it is a kind of can automatic searching Chinese caterpillar fungus or reduce search area automatic mode just seem non-
It is often important.
Due to the complexity of Chinese caterpillar fungus growing environment, common image detecting method is caused to use in this regard.Depth
Learning model is a kind of neural network model, using Layered Learning mechanism, can automatically from input data Level by level learning to height
Layer abstract data.So the present invention proposes a kind of Cordyceps sinensis detection method based on own coding feature learning, by collection
The sample image for having marked, trains coding characteristic automatically, and classification and Detection is carried out using Non-linear Kernel SVMs, can compare
Accurately detect the position of Chinese caterpillar fungus.Patent retrieval and domestic and international various Indexings of Scien. and Tech. Literature are shown, not yet have based on from
The Cordyceps sinensis detection method of coding characteristic study is seen in document.
The content of the invention
It is an object of the invention to provide a kind of Cordyceps sinensis detection method based on own coding feature learning, can be multiple
Cordyceps sinensis is detected under miscellaneous background.
Realize that technical solution of the invention is：A kind of Cordyceps sinensis detection method based on own coding feature learning,
Comprise the following steps：(1) a series of images comprising Cordyceps sinensis are gathered；(2) extract image in Cordyceps sinensis and other
Background, and it is fabricated to the positive negative sample of size identical；(3) sample image by extracting trains own coding model, obtains self-editing
Code model parameter；(4) sample is encoded by own coding model；(5) coding and sample class that will be obtained are using non-
Linear SVM carries out classification based training, obtains disaggregated model parameter；(6) Cordyceps sinensis image to be detected is gathered, by it
Piecemeal is carried out under multiple yardsticks；(7) each block of image is encoded using own coding model, using non-linear supporting vector
Machine classify and record position；(8) overlapping region for detecting is rejected, and by all non-coincidence area markings in mapping to be checked
As upper.
The present invention compared with prior art, its remarkable advantage：The present invention provides a kind of automatic detection Cordyceps sinensis first
Method, can on image automatic marking Cordyceps sinensis position.
Brief description of the drawings
Fig. 1 is the flow chart of Cordyceps sinensis detection method of the present invention.
Fig. 2 is stack own coding model schematic.
Fig. 3 is gaussian pyramid method schematic diagram.
Specific embodiment
The present invention is based on the Cordyceps sinensis detection method of own coding feature learning, and its step is：
1. the making of sample
1) a series of images comprising Cordyceps sinensis are gathered；
2) sub-block comprising Cordyceps sinensis, and the sub-block comprising background are extracted, withdrawal ratio is 1:10, and scale it
It is same size；
3) sample is sorted out according to positive and negative sample class, and is numbered.
2. the training of disaggregated model
1) Sample Storehouse is upset into order, is trained using two-layer stack autoencoder network, the model ginseng after being trained
Number；
2) Sample Storehouse is encoded using the model for training,
3) while coding and sample class label to be brought into Non-linear Kernel SVMs and be trained, classification is obtained
Model parameter, wherein Non-linear Kernel use Gaussian function；
3. the detection of Chinese caterpillar fungus
1) gather possibility to be detected and include Cordyceps sinensis image；
2) image is carried out into piecemeal on 5 yardsticks, the size of each block is identical with Sample Storehouse size；
3) for the image block split, encoded using the own coding model for training, then brought into what is trained
Non-linear Kernel supporting vector machine model is classified, and be would be classified as positive image block and recorded；
4) treat after the completion of the image block all classification on 5 yardsticks, the record that there will be overlapping region is purged；
Non-overlapping Domain is labeled on altimetric image to be checked by the way of rectangle frame, as the position of Cordyceps sinensis.
The present invention is described in further detail below in conjunction with the accompanying drawings.
1st, the making of sample
First, 500 width (The more the better) are no less than with picture of the camera acquisition comprising Cordyceps sinensis, if each width picture
Size be M × N；
Then, the samples pictures that 11k width size is m × n are cut out from each width picture, wherein (containing the winter containing positive sample
Worm summer grass) k width, the picture of negative sample (being free of Cordyceps sinensis) is 10k width, and wherein k is contained Cordyceps sinensis in this original image
Quantity；
Finally, positive negative sample is numbered in sequence, and classified and stored respectively, used with model training later.
2nd, the training of disaggregated model
Cordyceps sinensis coding specification model used in the present invention is own coding model and supporting vector machine model.Wherein compile
Code model is two-layer stack own coding model, and disaggregated model uses Non-linear Kernel SVMs, and core is Gaussian function.
1) feature own coding model
The extraction of feature uses two-layer stack own coding feature, as shown in Figure 2.Wherein,It is single figure
Decent, h, s, t is respectively hidden layer feature, and final own coding is characterized as s layers, W(1), W(2), W(3), W(4), b(1), b(2), b(3), b(4)It is the model parameter of autoencoder network.
Cordyceps sinensis sample in training sample database is coloured image, comprising three passages, so the α=3mn in model,
And in the present invention, h being set, t layers of its characteristic node number is γ=β/2=α/4.
After ensureing that all of positive and negative sample data is upset order by the generality of training, the present invention, instruction is all brought into
Practice model, be the parameter W that can obtain own coding characteristic model using the training method of feedback neural network(1), W(2), b(1), b(2)。
Finally, the feature own coding model for being completed using training, is brought all of training sample into this model and obtains s layers
Own coding feature.
2) disaggregated model：
Bringing own coding feature obtained in the previous step into two classification kernel support vectors machines carries out classification based training, is bringing training into
When, the tag along sort of positive sample is+1, and the tag along sort of negative sample is -1.Wherein, core uses non-linear Gaussian kernel
The model parameter for obtaining kernel support vectors machine may finally be trained.
3rd, the detection of Chinese caterpillar fungus
First, one width of collection may include the scene image of Cordyceps sinensis, if its original size is M × N, it is contemplated that single
The image of one yardstick may result in detection not exclusively, so the present invention is carried out in the case of multiple dimensioned, original M × N be schemed
As carrying out down-sampling using gaussian pyramid as shown in Figure 3, each layer of sampling ratio is the 3/4 of last layer, that is, after sampling
Image size beWherein i is pyramid level number, and ground floor is 0.
Same is the size of m × n by each layer of image block, brings each piecemeal into model shown in Fig. 2, is used
The parameter W for training(1), W(2), b(1), b(2)S layers of feature of correspondence piecemeal is calculated, this feature is brought into kernel support vectors machine mould
Type is classified, you can obtain the classification of this fritter.
If the classification of the fritter is made just, to record the yardstick sequence number and block serial number (i, c, d) of gaussian pyramid
It is labeling position, is negative if classification, then does not process.Wherein i is yardstick sequence number, and c, d is respectively on row and column direction
Block sequence number.
The calculating of all of 5 layers of yardstick is performed, all labeling positions for detecting and size has been counted, using four-tuple square
Shape frame carries out position mark, and the position of rectangle frame isWherein the first two
Numeral is original position, and final two digits are the length and width of rectangle frame.
Finally, it is contemplated that be possible to all detect same target under different scale, by calculating between rectangle frame
By inclusion relation, by it is all of by comprising rectangle frame position rejected, and according to result of calculation, in the figure of original M × N
Corresponding rectangle frame, the as position of Cordyceps sinensis are drawn out on picture.
Claims (2)
1. a kind of Cordyceps sinensis detection method based on own coding feature learning, it is characterised in that comprise the following steps：(1) gather
A series of images comprising Cordyceps sinensis；(2) Cordyceps sinensis and other backgrounds in image are extracted, and it is identical to be fabricated to size
Positive negative sample；(3) sample image by extracting trains own coding model, is obtained from encoding model parameter；(4) sample is led to
Own coding model is crossed to be encoded；(5) coding and sample class of acquisition are classified using Nonlinear Support Vector Machines
Training, obtains disaggregated model parameter；(6) Cordyceps sinensis image to be detected is gathered, it is carried out into piecemeal under multiple yardsticks；
(7) each block of image is encoded using own coding model, using Nonlinear Support Vector Machines classify and record position；
(8) overlapping region for detecting is rejected, and by all non-coincidence area markings on altimetric image to be checked.
2. the Cordyceps sinensis detection method based on own coding feature learning according to claim 1, it is characterised in that：Specifically
Step is as follows：
1. the making of sample
1) using picture of the camera acquisition comprising Cordyceps sinensis；
2) samples pictures that size is m × n are cut out from each width picture, wherein positive sample contains Cordyceps sinensis, and negative sample is not
Containing Cordyceps sinensis, positive and negative sample proportion is 1:10；
3) positive negative sample is numbered in sequence, and classified and stored respectively；
2. the training of disaggregated model
1) feature own coding：The extraction of feature uses two-layer stack own coding feature；α=3mn in model, set h, t layers its
Characteristic node number is γ=β/2=α/4, and after all of positive and negative sample data is upset into order, all bringing training pattern into is carried out
Training, obtains the parameter W of own coding characteristic model(1), W(2), b(1), b(2)；
The feature own coding model completed using training, brings all of training sample into own coding spy that this model obtains s layers
Levy；
2) disaggregated model：Bringing own coding feature obtained in the previous step into two classification kernel support vectors machines carries out classification based training, its
In, core uses non-linear Gaussian kernelFinal training obtains the model parameter of kernel support vectors machine；
3. the detection of Chinese caterpillar fungus
(1) one width of collection may include the scene image of Cordyceps sinensis；
(2) carried out in the case of multiple dimensioned, down-sampling carried out using gaussian pyramid to original M × N images, each layer is adopted
Sample ratio is the 3/4 of last layer, and each tomographic image size is after sampling
(3) it is the size of m × n by each layer of image block；
(4) bring own coding model into and supporting vector machine model is classified, if the classification of the fritter is just, record Gauss
Pyramidal yardstick sequence number and block serial number (i, c, d) are used as labeling position；
(5) calculating of be of five storeys yardstick is carried out in this way, all labeling positions for detecting and size is counted, and is labeled as
(6) by calculate between rectangle frame by inclusion relation, by it is all of by comprising rectangle frame position rejected, and root
According to result of calculation, corresponding rectangle frame, the as position of Cordyceps sinensis are drawn out on the image of original M × N.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611249294.7A CN106778896B (en) | 2016-12-29 | 2016-12-29 | Cordyceps sinensis detection method based on self-coding feature learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611249294.7A CN106778896B (en) | 2016-12-29 | 2016-12-29 | Cordyceps sinensis detection method based on self-coding feature learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106778896A true CN106778896A (en) | 2017-05-31 |
| CN106778896B CN106778896B (en) | 2020-12-25 |
Family
ID=58928383
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201611249294.7A Active CN106778896B (en) | 2016-12-29 | 2016-12-29 | Cordyceps sinensis detection method based on self-coding feature learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106778896B (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107562787A (en) * | 2017-07-31 | 2018-01-09 | 北京三快在线科技有限公司 | A kind of POI coding methods and device, POI recommend method, electronic equipment |
| CN108573226A (en) * | 2018-04-08 | 2018-09-25 | 浙江大学 | The drosophila larvae body segment key independent positioning method returned based on cascade posture |
| CN108694994A (en) * | 2018-05-11 | 2018-10-23 | 浙江大学 | Noninvasive cardiac infarction disaggregated model construction method based on stack self-encoding encoder and support vector machines |
| CN109948488A (en) * | 2019-03-08 | 2019-06-28 | 上海达显智能科技有限公司 | A kind of intelligence smoke eliminating equipment and its control method |
| CN114092489A (en) * | 2021-11-02 | 2022-02-25 | 清华大学 | Porous medium seepage channel extraction and model training method, device and equipment |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20050251347A1 (en) * | 2004-05-05 | 2005-11-10 | Pietro Perona | Automatic visual recognition of biological particles |
| CN104850836A (en) * | 2015-05-15 | 2015-08-19 | 浙江大学 | Automatic insect image identification method based on depth convolutional neural network |
-
2016
- 2016-12-29 CN CN201611249294.7A patent/CN106778896B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20050251347A1 (en) * | 2004-05-05 | 2005-11-10 | Pietro Perona | Automatic visual recognition of biological particles |
| CN104850836A (en) * | 2015-05-15 | 2015-08-19 | 浙江大学 | Automatic insect image identification method based on depth convolutional neural network |
Non-Patent Citations (1)
| Title |
|---|
| 张文达等: "基于多尺度分块卷积神经网络的图像目标识别算法", 《计算机应用》 * |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107562787A (en) * | 2017-07-31 | 2018-01-09 | 北京三快在线科技有限公司 | A kind of POI coding methods and device, POI recommend method, electronic equipment |
| CN107562787B (en) * | 2017-07-31 | 2020-11-13 | 北京三快在线科技有限公司 | POI (point of interest) encoding method and device, POI recommendation method and electronic equipment |
| CN108573226A (en) * | 2018-04-08 | 2018-09-25 | 浙江大学 | The drosophila larvae body segment key independent positioning method returned based on cascade posture |
| CN108573226B (en) * | 2018-04-08 | 2021-10-08 | 浙江大学 | Drosophila larva body node key point positioning method based on cascade posture regression |
| CN108694994A (en) * | 2018-05-11 | 2018-10-23 | 浙江大学 | Noninvasive cardiac infarction disaggregated model construction method based on stack self-encoding encoder and support vector machines |
| CN108694994B (en) * | 2018-05-11 | 2021-06-22 | 浙江大学 | Noninvasive cardiac infarction classification model construction method based on stack type self-encoder and support vector machine |
| CN109948488A (en) * | 2019-03-08 | 2019-06-28 | 上海达显智能科技有限公司 | A kind of intelligence smoke eliminating equipment and its control method |
| CN114092489A (en) * | 2021-11-02 | 2022-02-25 | 清华大学 | Porous medium seepage channel extraction and model training method, device and equipment |
| CN114092489B (en) * | 2021-11-02 | 2023-08-29 | 清华大学 | Porous medium seepage channel extraction and model training method, device and equipment |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106778896B (en) | 2020-12-25 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106778896A (en) | A kind of Cordyceps sinensis detection method based on own coding feature learning | |
| Liu et al. | Tomato diseases and pests detection based on improved Yolo V3 convolutional neural network | |
| CN107563381B (en) | Multi-feature fusion target detection method based on full convolution network | |
| Zhou et al. | Using colour features of cv.‘Gala’apple fruits in an orchard in image processing to predict yield | |
| Wang et al. | Deep learning approach for apple edge detection to remotely monitor apple growth in orchards | |
| CN109961024A (en) | Wheat weeds in field detection method based on deep learning | |
| CN107492095A (en) | Medical image pulmonary nodule detection method based on deep learning | |
| CN106683091A (en) | Target classification and attitude detection method based on depth convolution neural network | |
| CN107016405A (en) | A kind of insect image classification method based on classification prediction convolutional neural networks | |
| CN106778902A (en) | Milk cow individual discrimination method based on depth convolutional neural networks | |
| CN109086826A (en) | Wheat Drought recognition methods based on picture depth study | |
| CN107067043A (en) | A kind of diseases and pests of agronomic crop detection method | |
| CN111340826A (en) | Single tree crown segmentation algorithm for aerial image based on superpixels and topological features | |
| CN107464035A (en) | Chinese medicine performance rating method and system | |
| CN110033015A (en) | A kind of plant disease detection method based on residual error network | |
| CN110321956B (en) | Grass pest control method and device based on artificial intelligence | |
| CN113191334B (en) | Plant canopy dense leaf counting method based on improved CenterNet | |
| CN107704878A (en) | A kind of high-spectral data storehouse semi-automation method for building up based on deep learning | |
| CN109871905A (en) | A kind of plant leaf identification method based on attention mechanism depth model | |
| Huang et al. | Depth semantic segmentation of tobacco planting areas from unmanned aerial vehicle remote sensing images in plateau mountains | |
| Zheng et al. | YOLOv4-lite–based urban plantation tree detection and positioning with high-resolution remote sensing imagery | |
| CN107766810A (en) | A kind of cloud, shadow detection method | |
| Qi et al. | Related study based on otsu watershed algorithm and new squeeze-and-excitation networks for segmentation and level classification of tea buds | |
| Gavhale et al. | Identification of medicinal plant using Machine learning approach | |
| CN109615610A (en) | A kind of medical band-aid flaw detection method based on YOLO v2-tiny |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |