CN107153840A - A kind of crop pests image-recognizing method based on convolutional Neural - Google Patents
A kind of crop pests image-recognizing method based on convolutional Neural Download PDFInfo
- Publication number
- CN107153840A CN107153840A CN201710266710.2A CN201710266710A CN107153840A CN 107153840 A CN107153840 A CN 107153840A CN 201710266710 A CN201710266710 A CN 201710266710A CN 107153840 A CN107153840 A CN 107153840A
- Authority
- CN
- China
- Prior art keywords
- image
- layer
- convolutional neural
- crop pests
- input
- 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.)
- Withdrawn
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
-
- 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- 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
- G06N3/084—Backpropagation, e.g. using gradient descent
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- Computational Linguistics (AREA)
- Health & Medical Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of recognition methods of the crop pests image based on convolutional Neural, including：Crop pests image data base, image preprocessing, convolutional neural networks and model；The image of crop pests image data base is input to convolutional neural networks by pretreatment, the model of representative feature can be extracted after successively being calculated after the training of tape label image, the image that will classify is input to by pretreatment to be trained the convolutional neural networks of model and obtain the image and belong to various types of probability.The present invention gives full play to the self-teaching advantage of deep learning, by propagated forward and backpropagation training obtain that the model of representative feature can be extracted, when inputting a sub-picture, feature can accurately be extracted, progress is successively abstract to know the concept to form a certain things, and is classified with grader；And can recognize that displacement, scaling and other forms distort the X-Y scheme of consistency, the problem of being limited which solves shooting angle.
Description
Technical field
The present invention relates to machine learning field, more particularly to a kind of crop pests image recognition side based on convolutional Neural
Method.
Background technology
Traditional pest diagnosis is by the way of artificial observation, and this mode has subjectivity, limitation, ambiguity etc.
It is not enough.With Computer Image Processing and the development of artificial intelligence technology, people start with computer generation for people to carry out
The pest diagnosis of crops, it is proposed that realize the identification of pest and disease damage on computers.
Have pest and disease damage image be identified based on traditional image recognition technology at present, the technology using greyscale transformation,
Medium filtering, threshold segmentation, contour detecting, scab extract as with processing data, texture is explicitly extracted from preprocessed data special
Levy, color characteristic and shape facility.Traditional image-recognizing method is carried out based on image " point feature " or " line feature ".
It is preferable for the identification matching effect of general pattern, but when illumination condition is more complicated, photo angle changes greatly, robustness is not
It is good.Image-recognizing method based on convolutional Neural overcomes to be changed and photo angle change adaptation in traditional algorithm to illumination condition
The problem of property is not strong.
Moreover, above-mentioned traditional image-recognizing method only extracts the feature of the part of representative of image, such as SIFT
And SURF, with certain limitation, some processes also need to artificial selection；The easy over-fitting of artificial neural network, parameter is very
Hardly possible adjustment, trains slow, and effect is more excellent unlike other method when the number of plies is less.
In the 1960s, Hubel and Wiesel is used for the god of local sensitivity and set direction in research cat cortex
Find that its unique network structure can be effectively reduced the complexity of Feedback Neural Network during through member, then propose convolution god
Through net (Convolutional Neural Networks- abbreviation CNN).Now, CNN has become grinding for numerous scientific domains
One of focus is studied carefully, particularly in pattern classification field, because the network avoids the complicated early stage pretreatment to image, Ke Yizhi
Input original image is connect, thus has obtained more being widely applied.
The content of the invention
To solve shortcoming and defect of the prior art, the present invention proposes a kind of crop pests figure based on convolutional Neural
As recognition methods, convolutional Neural is combined with crop pests image recognition, using the training method successively initialized, fully
Playing the advantage of deep learning self-teaching can effectively solve that feature extraction is difficult, training difficulty is big, parameter is difficult adjustment
Problem.
It is achieved in that during technical scheme：
A kind of method of the crop pests image based on convolutional Neural, including：Crop pests image data base, image
Pretreatment, convolutional neural networks and model；The image of crop pests image data base is input to convolutional Neural by pretreatment
Network, the model of representative feature, Jiang Yaofen can be extracted after successively being calculated after the training of tape label image
The image of class is input to by pretreatment to be trained the convolutional neural networks of model and obtains the image and belong to various types of probability.
Alternatively, the input layer, hidden layer and output layer, comprise the following steps：
Step a, triple channel image pixel value of the input layer input by pretreatment.
Step b, hidden layer includes：Convolutional layer, pond layer, DropOut layers and full articulamentum, hidden layer is first by forward direction
Propagate, then edible backpropagation is trained；
Step c, output layer is the layer classified, and grader softmax receives full articulamentum input, exports the image institute
Category probability of all categories.
Alternatively, the step of hidden layer uses propagated forward, specifically includes：
Step a, sample X, a Yp are taken from sample set, X is inputted into network；
Step b, calculates corresponding reality output Op.
In this stage, information, by conversion step by step, is sent to output layer from input layer.This process is also network complete
The process performed after into training during normal operation.In the process, the calculating that network is performed is (actually to input and every layer
Weight matrix phase dot product, obtain last output result)：
Op=Fn (... (F2 (F1 (X*W1+b1) W2+b2) ...) Wn+bn).
Alternatively, hidden layer uses backpropagation, and detailed process is：
Step a, calculates reality output Op and corresponding preferable output Yp difference；
Step b, weight matrix is adjusted by the method backpropagation of minimization error.
Algorithm is following：
Output=Sigmoid (Sum (convolution value)+offset) of convolutional layer.
The beneficial effects of the invention are as follows：
1st, the self-teaching advantage of deep learning is given full play to, obtains extracting by propagated forward and backpropagation training
The model of representative feature, when inputting a sub-picture, can fast and accurately extract feature, progress is successively abstract to be known
The concept of a certain things is formed, and is classified with grader；
2nd, using the architectural feature of deep learning network, the training mechanism successively initialized is taken, training is substantially reduced difficult
Degree；
3rd, the feature of each section is extracted after being divided the image into due to deep learning network, feature is then subjected to group again
Conjunction, and the indirect image for comparing view picture figure, being obtained from any angle shot, obtained local feature are all almost identical, and this is just
The problem of solving shooting angle；And can recognize that displacement, scaling and other forms distort the X-Y scheme of consistency, this is just
Solve the problem of shooting angle is limited.
4th, convolutional Neural is combined with the farming control of insect that peasant pays close attention to, and is more convenient peasant and is administered insect, it is easy to is recognized
And propagation, improve researching value.
Brief description of the drawings
Fig. 1 is convolutional neural networks structural representation of the present invention；
Fig. 2 is convolution schematic diagram of the present invention；
Fig. 3 is DropOut layers of schematic diagram of the present invention.
Embodiment
Below in conjunction with the accompanying drawing in present example, the technical scheme in the embodiment of the present invention is carried out clear, complete
Ground is described, it is clear that described example is only a part of example of the invention, rather than whole examples.Based in the present invention
Example, the every other example that those of ordinary skill in the art are obtained under the premise of creative work is not made all belongs to
In the scope of protection of the invention.
As Figure 1-3, the crop pests image-recognizing method of the invention based on convolutional Neural, including：Crops
Insect image data base, image preprocessing, convolutional neural networks and model；The image of crop pests image data base is by pre-
Processing is input to convolutional neural networks, can extract representative after successively being calculated after the training of tape label image
The model of feature, the image that will classify by pretreatment be input to train model convolutional neural networks obtain the image category
In various types of probability, i.e. classification results.
Wherein, deep learning network is divided into：Input layer, hidden layer and output layer, comprise the following steps：
Step a, triple channel image pixel value of the input layer input by pretreatment.
Step b, hidden layer is set to 7 layers, first by propagated forward, reuses backpropagation modification model parameter to enter
Row training.
Step c, the layer that output layer is classified, grader softmax receives full articulamentum input, exports belonging to the image
Probability of all categories.
The step of hidden layer uses propagated forward, specifically includes：
Step a, sample X, a Yp are taken from sample set, X is inputted into network；
Step b, calculates corresponding reality output Op.
In this stage, information, by conversion step by step, is sent to output layer from input layer.This process is also network complete
The process performed after into training during normal operation.In the process, the calculating that network is performed is (actually to input and every layer
Weight matrix phase dot product, obtain last output result)：
Op=Fn (... (F2 (F1 (X*W1+b1) W2+b2) ...) Wn+bn)
Hidden layer uses backpropagation, and detailed process is：
Step a, calculates propagated forward output Op and corresponding preferable output Yp difference；
Step b, weight matrix is adjusted by the method backpropagation of minimization error.
Wherein, in above-mentioned steps b, adjustment weight matrix process is：Changing weight w and deviant b by gradient descent method makes
Propagated forward output valve Op restrains.
Algorithm is following：
Output=Sigmoid (Sum (convolution value)+offset) of convolutional layer；
Fig. 2 is the analysis diagram of part convolutional neural networks in the present invention, and as seen from Figure 2, input layer is aobvious layer, input
By the image pixel value of pretreatment.
The present invention gives full play to the self-teaching advantage of deep learning, and energy is obtained by propagated forward and backpropagation training
The model of representative feature is extracted, when inputting a sub-picture, feature can be fast and accurately extracted, carry out successively abstract
Know the concept to form a certain things, and classified with grader；And displacement, scaling and other forms distortion can be recognized
The X-Y scheme of consistency, the problem of being limited which solves shooting angle.
The feature that each section is extracted after deep learning network is divided the image into is additionally, since, then again feature is carried out
Combination, and the indirect image for comparing entire image, being obtained from any angle shot, obtained local feature are all almost identical,
The problem of being limited which solves shooting angle.
Also, the yield and quality of crops are more paid close attention in the identification of crop pests image, beneficial to popularity, it incite somebody to action both
With reference to very high researching value and application prospect.
The foregoing is only the preferred embodiments of the present invention, be not intended to limit the invention, it is all in spirit of the invention and
Within principle, made any modification, equivalent substitution and improvements etc. should be included in the scope of the protection.
Claims (10)
1. a kind of crop pests image-recognizing method based on convolutional Neural, it is characterised in that including：Crop pests image
Database, image preprocessing, convolutional neural networks and model；The image of crop pests image data base is inputted by pretreatment
To convolutional neural networks, the mould of representative feature can be extracted after successively being calculated after the training of tape label image
Type, the image that will classify is input to by pretreatment to be trained the convolutional neural networks of model and obtains the image and belong to various types of
The probability of type, i.e. classification results.
2. a kind of crop pests image-recognizing method based on convolutional Neural as claimed in claim 1, it is characterised in that institute
Image preprocessing is stated, including：Image pixel size normalization, pixel value size normalization.
3. a kind of crop pests image-recognizing method based on convolutional Neural as claimed in claim 1, it is characterised in that institute
Stating convolutional neural networks includes input layer, hidden layer and output layer, and specific steps include：
Step a, triple channel image pixel value of the input layer input by pretreatment；
Step b, hidden layer includes：Convolutional layer, pond layer, DropOut layers and full articulamentum, hidden layer are passed first by forward direction
Broadcast, then edible backpropagation is trained；
Step c, output layer is the layer classified, and grader softmax receives full articulamentum input, exports each belonging to the image
The probability of classification.
4. a kind of crop pests image-recognizing method based on convolutional Neural as claimed in claim 1, it is characterised in that institute
One group of data that model is one neutral net of correspondence as obtained by training tape label image data model are stated, input is included
Every layer is used in layer, hidden layer, output layer weight w and offset b.
5. a kind of crop pests image-recognizing method based on convolutional Neural as claimed in claim 2, it is characterised in that figure
As pixel size normalization refers to：Using bicubic interpolation method image scaling to unified size, bicubic interpolation method formula is such as
Under：
6. a kind of crop pests image-recognizing method based on convolutional Neural as claimed in claim 3, it is characterised in that volume
Lamination refers to：Convolution is carried out with the feature maps of a convolution kernel that can learn and last layer, then by an activation primitive, just
Output characteristic map can be obtained, each output map is the value for combining the multiple input maps of convolution, and convolution function is as follows：
Op=Fn (F2 (F1 (X*W1+b1).
7. a kind of crop pests image-recognizing method based on convolutional Neural as claimed in claim 3, it is characterised in that pond
Change layer to refer to：After convolution feature extraction is completed, feature in the region is calculated on each region of the matrix obtained in convolution
Average or maximum, DropOut refer to:Allow the weight of some nodes of the full articulamentum of network not work at random, play the extensive energy of enhancing
The effect of power, full articulamentum refers to：All features of upper strata input are connected, output layer is given by output valve.
8. a kind of crop pests image-recognizing method based on convolutional Neural as claimed in claim 3, it is characterised in that defeated
Go out layer to refer to：The layer of output category result, grader softmax receives full articulamentum input, exports of all categories belonging to the image
Probability, propagated forward, step includes：
Step a, a sample (X, Yp) is taken from sample set, X is inputted into network；
Step b, calculates corresponding reality output Op；
In this stage, information, by conversion step by step, is sent to output layer from input layer.This process is also that network completes to instruct
The process performed after white silk during normal operation.In the process, network perform calculating be：
Op=Fn (... (F2 (F1 (X*W1+b1) W2+b2) ...) Wn+bn).
9. a kind of crop pests image-recognizing method based on convolutional Neural as described in right 3, it is characterised in that reversely pass
Broadcast, step includes：
Step a, calculates reality output Op and corresponding preferable output Yp difference；
Step b, weight matrix is adjusted by the method backpropagation of minimization error；
Algorithm is following：
Output=Sigmoid (Sum (convolution value)+offset) of convolutional layer.
10. a kind of crop pests image-recognizing method based on convolutional Neural as claimed in claim 4, it is characterised in that
Model training, be specially：Trained by the data of tape label, by propagated forward and backpropagation, the parameter to each layer is entered
Row fine setting, the parameter of Accurate classification can be carried out to all kinds of images by obtaining one group.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710266710.2A CN107153840A (en) | 2017-04-21 | 2017-04-21 | A kind of crop pests image-recognizing method based on convolutional Neural |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710266710.2A CN107153840A (en) | 2017-04-21 | 2017-04-21 | A kind of crop pests image-recognizing method based on convolutional Neural |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107153840A true CN107153840A (en) | 2017-09-12 |
Family
ID=59794111
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710266710.2A Withdrawn CN107153840A (en) | 2017-04-21 | 2017-04-21 | A kind of crop pests image-recognizing method based on convolutional Neural |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107153840A (en) |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107665355A (en) * | 2017-09-27 | 2018-02-06 | 重庆邮电大学 | A kind of agricultural pests detection method based on region convolutional neural networks |
| CN107886077A (en) * | 2017-11-17 | 2018-04-06 | 苏州健雄职业技术学院 | A kind of crop pests recognition methods and its system based on wechat public number |
| CN110009043A (en) * | 2019-04-09 | 2019-07-12 | 广东省智能制造研究所 | A kind of pest and disease damage detection method based on depth convolutional neural networks |
| CN110427922A (en) * | 2019-09-03 | 2019-11-08 | 陈�峰 | One kind is based on machine vision and convolutional neural networks pest and disease damage identifying system and method |
| CN110825118A (en) * | 2019-12-23 | 2020-02-21 | 河南大学 | Multi-unmanned aerial vehicle cooperative farmland spraying method based on deep learning algorithm |
| CN111104830A (en) * | 2018-10-29 | 2020-05-05 | 富士通株式会社 | Deep learning model for image recognition, training device and method of deep learning model |
| CN112116143A (en) * | 2020-09-14 | 2020-12-22 | 贵州大学 | Forest pest occurrence probability calculation processing method based on neural network |
| CN113627274A (en) * | 2021-07-20 | 2021-11-09 | 南京信大卫星应用研究院有限公司 | Visual pest and disease damage identification equipment based on image identification |
| WO2023107023A1 (en) * | 2021-12-06 | 2023-06-15 | Onur Yolay | Artificial intelligence based predictive decision support system in disease, pest and weed fighting |
-
2017
- 2017-04-21 CN CN201710266710.2A patent/CN107153840A/en not_active Withdrawn
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107665355A (en) * | 2017-09-27 | 2018-02-06 | 重庆邮电大学 | A kind of agricultural pests detection method based on region convolutional neural networks |
| CN107886077A (en) * | 2017-11-17 | 2018-04-06 | 苏州健雄职业技术学院 | A kind of crop pests recognition methods and its system based on wechat public number |
| CN111104830A (en) * | 2018-10-29 | 2020-05-05 | 富士通株式会社 | Deep learning model for image recognition, training device and method of deep learning model |
| CN110009043A (en) * | 2019-04-09 | 2019-07-12 | 广东省智能制造研究所 | A kind of pest and disease damage detection method based on depth convolutional neural networks |
| CN110009043B (en) * | 2019-04-09 | 2021-08-17 | 广东省智能制造研究所 | Disease and insect pest detection method based on deep convolutional neural network |
| CN110427922A (en) * | 2019-09-03 | 2019-11-08 | 陈�峰 | One kind is based on machine vision and convolutional neural networks pest and disease damage identifying system and method |
| CN110825118A (en) * | 2019-12-23 | 2020-02-21 | 河南大学 | Multi-unmanned aerial vehicle cooperative farmland spraying method based on deep learning algorithm |
| CN112116143A (en) * | 2020-09-14 | 2020-12-22 | 贵州大学 | Forest pest occurrence probability calculation processing method based on neural network |
| CN112116143B (en) * | 2020-09-14 | 2023-06-13 | 贵州大学 | Forest pest occurrence probability calculation processing method based on neural network |
| CN113627274A (en) * | 2021-07-20 | 2021-11-09 | 南京信大卫星应用研究院有限公司 | Visual pest and disease damage identification equipment based on image identification |
| WO2023107023A1 (en) * | 2021-12-06 | 2023-06-15 | Onur Yolay | Artificial intelligence based predictive decision support system in disease, pest and weed fighting |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107153840A (en) | A kind of crop pests image-recognizing method based on convolutional Neural | |
| CN108615010A (en) | Facial expression recognizing method based on the fusion of parallel convolutional neural networks characteristic pattern | |
| CN110084794B (en) | Skin cancer image identification method based on attention convolution neural network | |
| Milton | Automated skin lesion classification using ensemble of deep neural networks in isic 2018: Skin lesion analysis towards melanoma detection challenge | |
| CN108648191B (en) | Pest image recognition method based on Bayesian width residual error neural network | |
| CN110084318B (en) | Image identification method combining convolutional neural network and gradient lifting tree | |
| Burhan et al. | Comparative study of deep learning algorithms for disease and pest detection in rice crops | |
| CN108304826A (en) | Facial expression recognizing method based on convolutional neural networks | |
| CN111860330A (en) | Apple leaf disease identification method based on multi-feature fusion and convolutional neural network | |
| CN107451565B (en) | Semi-supervised small sample deep learning image mode classification and identification method | |
| CN110363253A (en) | A kind of Surfaces of Hot Rolled Strip defect classification method based on convolutional neural networks | |
| Xu et al. | Recurrent convolutional neural network for video classification | |
| CN104077612B (en) | A kind of insect image-recognizing method based on multiple features rarefaction representation technology | |
| CN110472530B (en) | Retina OCT image classification method based on wavelet transformation and migration learning | |
| CN107944399A (en) | A kind of pedestrian's recognition methods again based on convolutional neural networks target's center model | |
| CN111127423B (en) | Rice pest and disease identification method based on CNN-BP neural network algorithm | |
| CN108053398A (en) | A kind of melanoma automatic testing method of semi-supervised feature learning | |
| CN111178177A (en) | Cucumber disease identification method based on convolutional neural network | |
| CN107665352A (en) | A kind of pearl sorting technique based on multichannel residual error network | |
| CN108334901A (en) | A kind of flowers image classification method of the convolutional neural networks of combination salient region | |
| CN108268890A (en) | A kind of hyperspectral image classification method | |
| CN115331104A (en) | Crop planting information extraction method based on convolutional neural network | |
| CN113344077A (en) | Anti-noise solanaceae disease identification method based on convolution capsule network structure | |
| CN110751271B (en) | Image traceability feature characterization method based on deep neural network | |
| CN111563542A (en) | Automatic plant classification method based on convolutional neural network |
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 | ||
| WW01 | Invention patent application withdrawn after publication |
Application publication date: 20170912 |
|
| WW01 | Invention patent application withdrawn after publication |