CN108304844A - Agricultural pest recognition methods based on deep learning binaryzation convolutional neural networks - Google Patents
Agricultural pest recognition methods based on deep learning binaryzation convolutional neural networks Download PDFInfo
- Publication number
- CN108304844A CN108304844A CN201810090151.9A CN201810090151A CN108304844A CN 108304844 A CN108304844 A CN 108304844A CN 201810090151 A CN201810090151 A CN 201810090151A CN 108304844 A CN108304844 A CN 108304844A
- Authority
- CN
- China
- Prior art keywords
- convolutional neural
- neural networks
- binaryzation
- deep learning
- recognition methods
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
- G06V10/267—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion by performing operations on regions, e.g. growing, shrinking or watersheds
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of agricultural pest recognition methods based on deep learning binaryzation convolutional neural networks, by building a deep learning convolutional neural networks model, by carrying out binary conversion treatment to convolutional neural networks, compact model reduces dependence of the model to hardware device, to keep certain accuracy of identification and speed, it is portable with real-time, the characteristics of high degree of accuracy, the blank that deep learning is agriculturally applied has been filled up, while cost of labor can be reduced.
Description
Technical field
The invention belongs to agricultural pests to identify field, especially with the convolutional neural networks mould of deep learning binaryzation
Type is come method that work is identified.
Background technology
Plant pest disaster is one of Chinese three the Nature disasters, and identification, monitoring, early warning are the decision letters of prevention and control
Cease source.FAO (Food and Agriculture Organization of the United Nation) research shows that, only diseases and pests of agronomic crop endangers natural loss rate just more than 37%.China is
Plant pest including crops endangers big country, if not taking prevention and control measure, it is contemplated that because disease pest harm every year will loss
100,000,000,000 kg of 150,000,000,000 kg of grain, fruit and vegetables, 6,800,000,000 kg of oil plant, 1.9 hundred million kg of cotton, Potential economic losses are at 500,000,000,000 yuan
More than.The method of conventional diagnostic diseases and pests of agronomic crop is manually to estimate, but this has two：On the one hand, peasant and can not
Ensure that the judgement rule of thumb made is completely correct；It on the other hand, can since professional person can not reach field diagnostic in time
The plant state of an illness can be made to be delayed or aggravate.The application of deep learning agriculturally at present is very few, and China is in agriculture disease pest
Main or by expert professional knowledge in terms of evil regulation, but the expert's quantity and energy in agricultural pest field are limited,
Even in terms of common-depth learning model is applied to agricultural, because its model is big, when mobile terminal is run, elapsed time is long, together
When requirement to hardware device it is relatively high, and cannot achieve.
Invention content
It is an object of the invention to overcome the above-mentioned deficiency in the presence of the prior art, the present invention, which uses, is based on deep learning
The agricultural pest recognition methods of binaryzation convolutional neural networks acquires some blade pictures with camera or mobile phone and puts it into
In the model of the present invention, show that this is any pest and disease damage of any plant by result of calculation, compared to traditional technology,
Model size issue is solved, so that it is reduced the dependence of hardware device, and keep higher accuracy of identification and speed, simultaneously
Decrease cost of labor.
In order to achieve the above-mentioned object of the invention, the present invention provides following technical schemes：
A kind of agricultural pest recognition methods based on deep learning binaryzation convolutional neural networks, includes the following steps：
(1) it collects and original image and is pre-processed, composing training collection and test set, calculate the equal of training set and test set
It is worth image；
(2) convolutional neural networks are built；
(3) binaryzation is carried out to the convolutional neural networks parameter, constitutes binaryzation convolutional neural networks；
(4) training set is utilized, the binaryzation convolutional neural networks are trained；
(5) test set is utilized, the binaryzation convolutional neural networks model that training is completed is made to know test sample
Not.
Preferably, in step (1), including：
(1-1) acquires RGB leaf images, and classification is marked to pest and disease damage in identification, label pest species；
(1-2) carries out gray processing delustring processing to the RGB leaf images, constitutes data set；
(1-3) is split the RGB leaf images, and the blade is made only to retain leaf image, eliminates background influence；
Gray processing delustring processing is carried out to described image again, constitutes data set；
(1-4) normalizes the size of the image of (1-2) and (1-3), respectively constitutes training set and test set；
(1-5) calculates the mean value image of the training set and all images of test set, carries out subtracting mean value training.
Preferably, in step (2), the binaryzation convolutional neural networks include：One input layer, three convolutional layers, two
A maximum pond down-sampling layer, two full articulamentums, an output layer.
Preferably, the input of the input layer is the picture Jing Guo the step (1-4) normalized.
Preferably, the convolutional layer all uses the identical convolution kernel of size, and convolution kernel is shared in a characteristic spectrum
Parameter.
Preferably, the maximum pond down-sampling layer is operated by maximum pondization, is exported characteristic spectrum, is obtained leaf image
Feature.
Preferably, in step (3), using Binary-Weight-Networks networks, to the convolutional neural networks
Weighting parameter binaryzation, wherein not including input layer.
Preferably, using the training set, binaryzation convolutional Neural is trained in conjunction with the stochastic gradient descent algorithm based on SGD
Network.
Compared with prior art, beneficial effects of the present invention：The present invention is to be based on deep learning binaryzation convolutional Neural net
Network technology come realize agricultural pest know method for distinguishing, by build a depth convolutional neural networks model, by convolution
Neural network carries out binary conversion treatment, and compact model reduces dependence of the model to hardware device, to keep higher identification
The characteristics of accuracy and speed has real-time, portable, high degree of accuracy, has filled up deep learning and has agriculturally applied
Blank, while cost of labor can be reduced.
Description of the drawings
Fig. 1 is that the present invention is based on the agricultural pest identification process figures of deep learning binaryzation convolutional neural networks；
Fig. 2 is image normalization flow chart；
Fig. 3 is binaryzation convolutional neural networks figure.
Specific implementation mode
With reference to test example and specific implementation mode, the present invention is described in further detail.But this should not be understood
It is only limitted to embodiment below for the range of the above-mentioned theme of the present invention, it is all that this is belonged to based on the technology that the content of present invention is realized
The range of invention.
Convolutional neural networks and general neural network difference lies in, convolutional neural networks contain one by convolutional layer and
The feature extractor that sub-sampling layer is constituted.In the convolutional layer of convolutional neural networks, a neuron is only neural with part adjacent bed
Member connection.In a convolutional layer of CNN, generally comprise several characteristic planes (featureMap), each characteristic plane by
The neuron of some rectangular arrangeds forms, and the neuron of same characteristic plane shares weights, and shared weights are exactly to roll up here
Product core.Convolution kernel initializes generally in the form of random decimal matrix, and convolution kernel obtains study in the training process of network
Rational weights.The direct benefit that shared weights (convolution kernel) are brought is the connection reduced between each layer of network, while being reduced again
The risk of over-fitting.Sub-sampling is also referred to as pond (pooling), usually there is mean value sub-sampling (mean pooling) and maximum
It is worth sub-sampling (max pooling) two kinds of forms.Sub-sampling is considered as a kind of special convolution process.Convolution sum sub-sampling is big
Model complexity is simplified greatly, reduces the parameter of model.
Embodiment 1, as shown in Figure 1, a kind of agricultural pest identification based on deep learning binaryzation convolutional neural networks
Method includes the following steps：
(1) it collects and original image and is pre-processed, composing training collection and test set, calculate the equal of training set and test set
It is worth image；
(2) convolutional neural networks are built；
(3) binaryzation is carried out to the convolutional neural networks parameter, constitutes binaryzation convolutional neural networks；
(4) training set is utilized, binaryzation convolutional neural networks, warp are trained in conjunction with the stochastic gradient descent algorithm based on SGD
Cross inverse iteration update, repetition training, the size until reaching pre-set iterative value；
(5) test set is utilized, the binaryzation convolutional neural networks model that training is completed is made to know test sample
Not.
Embodiment 2, as shown in Fig. 2, to original image normalized, including：
(1-1) acquires RGB leaf images, and classification is marked to pest and disease damage in identification, label pest species；
(1-2) carries out gray processing delustring processing to the RGB leaf images, constitutes data set；
(1-3) is split the RGB leaf images, and the blade is made only to retain leaf image, eliminates background influence；
Gray processing delustring processing is carried out to described image again, constitutes data set；
(1-4) cuts original image to the image coordinate of (1-2) and (1-3) and scaling is to uniform sizes, distinguishes structure
At training set and test set；
(1-5) calculates the mean value image of the training set and all images of test set, carries out subtracting mean value training.
Embodiment 3, as shown in figure 3, the binaryzation convolutional neural networks include：One input layer, three binaryzation volumes
Lamination, two maximum pond down-sampling layers, two full articulamentums, an output layer.
The input of the input layer is the picture that pixel is 256 × 256.
The convolutional layer all uses the convolution kernel of 9 × 9 sizes, and convolution kernel slides a pixel every time, i.e. step-length is 1,
Padding is 0；The convolution kernel shared parameter in a characteristic spectrum.
The maximum pond down-sampling layer is using 2 × 2 input domain, i.e., the 4 of last layer node is as next layer 1
The input of a node, and input domain is not overlapped, i.e., 2 pixels (step-length 2) of sliding export by the operation of maximum pondization every time
Characteristic spectrum obtains leaf image feature.
The full articulamentum, it is 500 that first layer output, which is node, and it is 38 that second layer output, which is node,.
The number of nodes of the output layer is consistent with the class number of test sample.
The present invention mainly simplifies convolutional neural networks so that convolutional neural networks can come in CPU real time executions, still
Some precision can be sacrificed.The present invention proposes the weights that Binary-Weight-Networks simplifies network to convolutional neural networks
It is approximate to carry out binaryzation, so that it is accounted for memory reduces more than 32 times, enabling convolutional neural networks, and precision are run on mobile terminal
It will not sacrifice too much, such as can carry out agricultural pest with mobile phone acquisition picture by trained model transplantations to mobile phone
Classification.
Simplify network using Binary-Weight-Networks and binaryzation approximation is carried out to the weights of convolutional neural networks
When, the weighting parameter of each layer of traversal, with (+1, -1) and the approximate real weights of a scale factor.The specific method is as follows：
Uniform provisions symbol, L layers of convolutional neural networks may be defined as<I, W, *>, wherein I=IL (l=1 ..., L)Indicate volume
L layers of input tensor of product neural network,Indicate l layers in convolutional neural networks of k-th of weights square
Battle array, wherein klFor the number of weight matrix in l layer networks, * indicates the convolution operation between I and W.If realizing entire ginseng
Number binaryzation, can be by each weight matrix W beforelkIt is decomposed into a two-value weight matrix Blk∈{+1,-1}w×h×d×cWith
One scale coefficient αlk∈R+The multiplication of (or being called two finger coefficients, the i.e. coefficient of two values matrix, which is real number), formula are
Wlk≈αlkBlk.Therefore the operation of a convolution can be approximated to be formula (1).
HereinIndicate convolution operation, the operation is without multiplication.Because of two-value weight matrix BlkAll elements only-
1 and+1 two value, therefore convolution operation can be reduced to add operation and reducing, it adds and subtracts each element after convolution operation
With two-value factor alphalkIt is multiplied, then obtains the output of the two-value convolutional layer.
To find ideal two-value factor alphalkWith two-value weights Blk, then one is first trained before training this two-value network
The convolutional neural networks of a standard obtain the network weight W after optimizing according to appointed task, wherein W after traininglk∈ W, weight are
Arbitrary real number.It can make J (B to findlk,αlk) minimum of alphalkBlk, that is, allow αlkBlkThe weights optimized are characterized as far as possible
Matrix W can be acquired according to the object function of formula (2).
J(Blk,αlk)=| | Wlk-αlkBlk||2 (2)
Claims (8)
1. a kind of agricultural pest recognition methods based on deep learning binaryzation convolutional neural networks, which is characterized in that including
Following steps：
(1) it collects original image and is pre-processed, composing training collection and test set, calculate the mean value figure of training set and test set
Picture；
(2) convolutional neural networks are built；
(3) binaryzation is carried out to the convolutional neural networks parameter, obtains binaryzation convolutional neural networks；
(4) training set is utilized, the binaryzation convolutional neural networks are trained；
(5) test set is utilized, makes the binaryzation convolutional neural networks model that training is completed that test sample be identified.
2. the agricultural pest recognition methods according to claim 1 based on deep learning binaryzation convolutional neural networks,
It is characterized in that, in step (1), including：
(1-1) acquires RGB leaf images, and classification is marked to pest and disease damage in identification, label pest species；
(1-2) carries out gray processing delustring processing to the RGB leaf images, constitutes data set；
(1-3) is split the RGB leaf images, and the blade is made only to retain leaf image, eliminates background influence；It is right again
Described image carries out gray processing delustring processing, constitutes data set；
(1-4) normalizes the size of the image of (1-2) and (1-3), respectively constitutes training set and test set；
(1-5) calculates the mean value image of the training set and all images of test set, carries out subtracting mean value training.
3. the agricultural pest recognition methods according to claim 1 based on deep learning binaryzation convolutional neural networks,
It is characterized in that, in step (2), the convolutional neural networks include：One input layer, three convolutional layers, two maximum ponds
Down-sampling layer, two full articulamentums, an output layer.
4. the agricultural pest recognition methods according to claim 3 based on deep learning binaryzation convolutional neural networks,
It is characterized in that, the input of the input layer is the picture Jing Guo the step (1-4) normalized.
5. the agricultural pest recognition methods according to claim 3 based on deep learning binaryzation convolutional neural networks,
It is characterized in that, the convolutional layer all uses the identical convolution kernel of size, and the convolution kernel shared parameter in a characteristic spectrum.
6. the agricultural pest recognition methods according to claim 5 based on deep learning binaryzation convolutional neural networks,
It is characterized in that, the maximum pond down-sampling layer is operated by maximum pondization, characteristic spectrum is exported, obtains leaf image spy
Sign.
7. the agricultural pest recognition methods according to claim 1 based on deep learning binaryzation convolutional neural networks,
It is characterized in that, in step (3), using Binary-Weight-Networks networks, to the weights of the convolutional neural networks
Parameter binaryzation.
8. the agricultural pest recognition methods according to claim 1 based on deep learning binaryzation convolutional neural networks,
It is characterized in that, in step (4), using the training set, in conjunction with the stochastic gradient descent algorithm training binaryzation volume based on SGD
Product neural network.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810090151.9A CN108304844A (en) | 2018-01-30 | 2018-01-30 | Agricultural pest recognition methods based on deep learning binaryzation convolutional neural networks |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810090151.9A CN108304844A (en) | 2018-01-30 | 2018-01-30 | Agricultural pest recognition methods based on deep learning binaryzation convolutional neural networks |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108304844A true CN108304844A (en) | 2018-07-20 |
Family
ID=62866940
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810090151.9A Pending CN108304844A (en) | 2018-01-30 | 2018-01-30 | Agricultural pest recognition methods based on deep learning binaryzation convolutional neural networks |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108304844A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110633668A (en) * | 2019-09-09 | 2019-12-31 | 合肥飞扬机电有限公司 | Railway shunting signal lamp detection method and system based on binary convolution neural network |
| CN111105393A (en) * | 2019-11-25 | 2020-05-05 | 长安大学 | Grape disease and pest identification method and device based on deep learning |
| CN112528726A (en) * | 2020-10-14 | 2021-03-19 | 石河子大学 | Aphis gossypii insect pest monitoring method and system based on spectral imaging and deep learning |
| CN112560644A (en) * | 2020-12-11 | 2021-03-26 | 四川大学 | Crop disease and insect pest automatic identification method suitable for field |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104850858A (en) * | 2015-05-15 | 2015-08-19 | 华中科技大学 | Injection-molded product defect detection and recognition method |
| CN106971160A (en) * | 2017-03-23 | 2017-07-21 | 西京学院 | Winter jujube disease recognition method based on depth convolutional neural networks and disease geo-radar image |
| CN107230202A (en) * | 2017-05-16 | 2017-10-03 | 淮阴工学院 | The automatic identifying method and system of pavement disease image |
| CN107563389A (en) * | 2017-09-11 | 2018-01-09 | 合肥工业大学 | A kind of corps diseases recognition methods based on deep learning |
-
2018
- 2018-01-30 CN CN201810090151.9A patent/CN108304844A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104850858A (en) * | 2015-05-15 | 2015-08-19 | 华中科技大学 | Injection-molded product defect detection and recognition method |
| CN106971160A (en) * | 2017-03-23 | 2017-07-21 | 西京学院 | Winter jujube disease recognition method based on depth convolutional neural networks and disease geo-radar image |
| CN107230202A (en) * | 2017-05-16 | 2017-10-03 | 淮阴工学院 | The automatic identifying method and system of pavement disease image |
| CN107563389A (en) * | 2017-09-11 | 2018-01-09 | 合肥工业大学 | A kind of corps diseases recognition methods based on deep learning |
Non-Patent Citations (5)
| Title |
|---|
| COURBARIAUX M: "Binarized neural networks:training deep neural networks with weights and activations constrained to + 1 or -1", 《HTTPS: //ARXIV.ORG/ABS/1602.02830》 * |
| MOHAMMAD RASTEGARI: "XNOR-Net: ImageNet Classification Using Binary Convolutional Neural Networks", 《EUROPEAN CONFERENCE ON COMPUTER VISION》 * |
| 张善文: "基于物联网和深度卷积神经网络的冬枣病害识别方法", 《浙江农业学报》 * |
| 梁万杰: "基于卷积神经网络的水稻虫害识别", 《江苏农业科学》 * |
| 雷印杰: "四元数奇异值分解与彩色图像去噪", 《四川大学学报》 * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110633668A (en) * | 2019-09-09 | 2019-12-31 | 合肥飞扬机电有限公司 | Railway shunting signal lamp detection method and system based on binary convolution neural network |
| CN111105393A (en) * | 2019-11-25 | 2020-05-05 | 长安大学 | Grape disease and pest identification method and device based on deep learning |
| CN111105393B (en) * | 2019-11-25 | 2023-04-18 | 长安大学 | Grape disease and pest identification method and device based on deep learning |
| CN112528726A (en) * | 2020-10-14 | 2021-03-19 | 石河子大学 | Aphis gossypii insect pest monitoring method and system based on spectral imaging and deep learning |
| CN112528726B (en) * | 2020-10-14 | 2022-05-13 | 石河子大学 | Cotton aphid pest monitoring method and system based on spectral imaging and deep learning |
| CN112560644A (en) * | 2020-12-11 | 2021-03-26 | 四川大学 | Crop disease and insect pest automatic identification method suitable for field |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108304844A (en) | Agricultural pest recognition methods based on deep learning binaryzation convolutional neural networks | |
| Suryawati et al. | Deep structured convolutional neural network for tomato diseases detection | |
| CN110309856A (en) | Image classification method, the training method of neural network and device | |
| CN105956150B (en) | A kind of method and device generating user's hair style and dressing collocation suggestion | |
| CN110378381A (en) | Object detecting method, device and computer storage medium | |
| CN106845418A (en) | A kind of hyperspectral image classification method based on deep learning | |
| CN110462680A (en) | System and method for improving image texture | |
| CN106778701A (en) | A kind of fruits and vegetables image-recognizing method of the convolutional neural networks of addition Dropout | |
| CN109344738A (en) | The recognition methods of crop diseases and pest crop smothering and device | |
| Sharma et al. | KrishiMitr (Farmer’s Friend): using machine learning to identify diseases in plants | |
| CN113221913A (en) | Agriculture and forestry disease and pest fine-grained identification method and device based on Gaussian probability decision-level fusion | |
| Ma et al. | Research on fish image classification based on transfer learning and convolutional neural network model | |
| Sehree et al. | Olive trees cases classification based on deep convolutional neural network from unmanned aerial vehicle imagery | |
| Younis et al. | Robust optimization of mobilenet for plant disease classification with fine tuned parameters | |
| Ratha et al. | Papaya fruit maturity estimation using wavelet and ConvNET | |
| CN110956201A (en) | Image distortion type classification method based on convolutional neural network | |
| Ram et al. | Olive oil content prediction models based on image processing | |
| Luz et al. | Boron deficiency precisely identified on growth stage V4 of maize crop using texture image analysis | |
| Pratondo et al. | Classification of sweet potato leaf variants using transfer learning | |
| Saravanan et al. | Prediction of mango leaf diseases using convolutional neural network | |
| Altınbilek et al. | Identification of paddy rice diseases using deep convolutional neural networks | |
| CN115035511A (en) | ResNet residual error network-based pest and disease identification method and device | |
| CN108460426A (en) | A kind of image classification method based on histograms of oriented gradients combination pseudoinverse learning training storehouse self-encoding encoder | |
| CN113128525A (en) | Control device and method for desert grassland population patch identification | |
| Iparraguirre-Villanueva et al. | Disease identification in crop plants based on convolutional neural networks |
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 |
Application publication date: 20180720 |
|
| RJ01 | Rejection of invention patent application after publication |