CN108960310A - A kind of agricultural pest recognition methods based on artificial intelligence - Google Patents
A kind of agricultural pest recognition methods based on artificial intelligence Download PDFInfo
- Publication number
- CN108960310A CN108960310A CN201810660699.2A CN201810660699A CN108960310A CN 108960310 A CN108960310 A CN 108960310A CN 201810660699 A CN201810660699 A CN 201810660699A CN 108960310 A CN108960310 A CN 108960310A
- Authority
- CN
- China
- Prior art keywords
- image
- neural network
- layer
- scale
- obtains
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- 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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/30—Noise filtering
Abstract
Step 1 a kind of agricultural pest recognition methods based on artificial intelligence pre-processes collected leaf image, leaf image is transformed to frequency domain using Fast Fourier Transform (FFT), removes high frequency section, reverts to spatial domain picture;Step 2, obtained image is subjected to the processing of fragment multiscalization, obtains the image slice of different scale, forms the training dataset of different scale;Step 3, building deep neural network model carries out parameter initialization and tuning, it is made to be more suitable for crops data;Step 4, Analysis On Multi-scale Features depth network model obtained carry out MLP Fusion Features, are identified using the feature training classifier of fusion.
Description
Technical field
The present invention relates to a kind of recognition methods more particularly to a kind of agricultural pest recognition methods based on artificial intelligence.
Background technique
In currently available technology, there are following three kinds for the Intelligent prevention and cure method of pest and disease damage, first is that based on deep learning
Crop pest control scheme recommender system, but this system not only needs to input cultivar name, but also there is faces
To small data set, the problems such as deep neural network is difficult to train, frequent over-fitting.Second is that the crop of fusion spectrum and image information
Pest and disease damage identifies and distinguishes between method, but the method needs the high spectrum image using EO-1 hyperion camera herborization blade, therefore
This method higher cost, can not promote civilian;Third is that the sunflower leaf diseases judgment method based on support vector machines, but its
It is voted using multiple two classifiers, needs to train a fairly large number of classifier when in face of a variety of diseases, therefore work as disease
This method is difficult to work normally when class is more.
Summary of the invention
The present invention proposes a kind of agricultural pest recognition methods based on artificial intelligence.Acquired image is carried out first
Leaf image is transformed to frequency domain using Fast Fourier Transform (FFT) by pretreatment, and setting average threshold removes high frequency section, then
Spatial domain picture is reverted to, the background of the noise of leaf image and complexity is removed at this time, only the remaining blade for needing to be concerned
Main body.Then obtained image is subjected to the processing of fragment multiscalization, obtains the image slice of different scale, carries out forming difference
The training dataset of scale.Use 16 network model of vgg crossed via million image data set Image net pre-training
Parameter carries out parameter initialization to the deep neural network model that we construct.Then using above-mentioned training dataset to obtaining
Deep neural network carries out tuning, it is made to be more suitable for crops data.The multiple dimensioned spy that finally depth network model is obtained
Sign carries out MLP Fusion Features, using feature one classifier of training of fusion, completes pest and disease damage identification mission using classifier.
Detailed description of the invention
Fig. 1 is the overall flow figure of recognition methods of the present invention;
The step of Fig. 2 is recognition methods of the present invention in an embodiment is schemed;
Specific embodiment
In order to make the objectives, technical solutions, and advantages of the present invention clearer, with reference to the accompanying drawings and embodiments, right
The present invention is further elaborated.It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention, and
It is not used in the restriction present invention.As long as in addition, technical characteristic involved in the various embodiments of the present invention described below
Not constituting a conflict with each other can be combined with each other.
The present invention proposes a kind of agricultural pest recognition methods based on artificial intelligence, is as shown in Figure 1 present invention identification
The overall flow figure of method;
Acquired image is pre-processed first, leaf image is transformed into frequency domain using Fast Fourier Transform (FFT),
Setting average threshold removes high frequency section, then reverts to spatial domain picture, at this time the noise of leaf image and the back of complexity
Scape is removed, only the remaining blade body for needing to be concerned.
Then obtained image is subjected to the processing of fragment multiscalization, obtains the image slice of different scale, is formed
The training dataset of different scale.
Using the 16 network model parameter of vgg crossed via million image data set Image net pre-training to us
The deep neural network model of building carries out parameter initialization.Then using above-mentioned training dataset to obtained depth nerve net
Network carries out tuning, it is made to be more suitable for crops data.
The Analysis On Multi-scale Features that finally depth network model is obtained carry out MLP Fusion Features, use the feature training of fusion
One classifier completes pest and disease damage identification mission using classifier.
The step of being illustrated in figure 2 recognition methods of the present invention in embodiment figure, first to collected leaf image into
Row pretreatment, leaf image is transformed into frequency domain using Fast Fourier Transform (FFT), to having a size of M*N leaf image function f (x,
Y) discrete Fourier transform is
U and v is the frequency variable of leaf image, and M and N are positive number, and F (u, v) is the frequency distribution of leaf image
Map, wherein the position frequency closer to center is lower, the position of brighter (gray value is higher) represents the signal amplitude of the frequency
It is bigger, thus using centre be it is true, remaining be that false masking-out is multiplied with frequency distribution map and to filter high frequency.
Masking-out
Spatial domain picture f ' (x, y) is converted by frequency distribution map using inverse Fourier transform
Wherein
F '=F × G
Then the image f'(x, y that will be obtained) multiple dimensioned fragmentation processing is carried out, it is that size is cut with (M/2, N/2)
Piece obtains 4 (M/2 × N/2) sectioning images;It is that size is sliced with (M/3, N/3), obtains 9 (M/3 × N/3) slices
Image;All images are all done into above-mentioned processing, obtain the training set of three kinds of scales, scale is respectively (M × N) (M/2 × N/2)
(M/3×N/3)。
Then neural network is constructed according to deep neural network VGG-16 network model, and trained using Image-Net
The parameter arrived is its initiation parameter, the data set handled using multiple dimensioned fragmentation and its label training network model, study
Model parameter.Freeze convolution layer parameter in learning process, convolutional layer learning rate is set to 0, only full connection layer parameter is carried out
It optimizes and revises, learning rate is set to 0.001.If former label is y, the output that deep neural network model obtains is a, then obtain
Cross entropy is C, and wherein x indicates that sample, n indicate total sample number.
Then tuning is carried out to obtained deep neural network, the optimization method is
Wherein, σ ' (z)=σ (z) (1- σ (z)), σ (z) are activation primitives, and exporting as a=σ (z), z is neural network node
Input XjWith network weight ωjInner product, j is positive integer.
After deep neural network study convergence, network the last layer is removed, image input network is obtained into 4096 dimensions
Sparse features vector.Due to having input the image of three groups of different scales, multiple dimensioned image characteristics of image is obtained.By three kinds of rulers
The merging features of degree are to the vector for obtaining 4096*3=12288 dimension together, and using this vector as input, building MLP feature is melted
Close frame.
MLP Fusion Features frame includes an input layer, two hidden layers and an output layer.Wherein input layer includes
12288 neurons, first hidden layer include the full articulamentum of 4096 neurons, and second hidden layer is to include 4096
The full articulamentum of neuron, output layer are Softmax layers, and neuron number is output classification number.Finally using under stochastic gradient
It drops algorithm and carries out parameter update, the feature merged after convergence and recognition result.
Finally, it should be noted that the above embodiments are merely illustrative of the technical solutions of the present invention, rather than its limitations;Although
Present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that: it still may be used
To modify to technical solution documented by previous embodiment or equivalent replacement of some of the technical features;And
These are modified or replaceed, the spirit and model of technical solution of various embodiments of the present invention that it does not separate the essence of the corresponding technical solution
It encloses.
Claims (5)
1. a kind of agricultural pest recognition methods based on artificial intelligence, which is characterized in that step 1, to collected leaf figure
As being pre-processed, leaf image is transformed into frequency domain using Fast Fourier Transform (FFT), removes high frequency section, reverts to airspace figure
Picture;Step 2, the obtained spatial domain picture is subjected to the processing of fragment multiscalization, obtains the image slice of different scale, formed
The training dataset of different scale;Step 3, building deep neural network model carries out parameter initialization and tuning;Step 4, will
The Analysis On Multi-scale Features that depth network model obtains carry out MLP Fusion Features, are carried out using the feature training classifier of the fusion
Identification.
2. the method as described in claim 1, which is characterized in that in the step 1, first to collected leaf image into
Row pretreatment, leaf image is transformed into frequency domain using Fast Fourier Transform (FFT), to having a size of M*N leaf image function f (x,
Y) discrete Fourier transform is
The u and v is the frequency variable of leaf image, and F (u, v) is the frequency distribution map of leaf image, wherein closer to
The position frequency at center is lower, and the signal amplitude that the higher position of gray value represents the frequency is bigger, using centre be it is true, remaining
It is that false masking-out is multiplied with frequency distribution map and filters high frequency, the masking-out is
Spatial domain picture f ' (x, y) is converted by frequency distribution map using inverse Fourier transform,
Wherein
F '=F × G
3. method according to claim 2, which is characterized in that in the step 2, the image f'(x, y that will obtain) it carries out
Multiple dimensioned fragmentation processing is that size is sliced with (M/2, N/2), obtains 4 (M/2 × N/2) sectioning images;With (M/3,
N/3 it) is sliced for size, obtains 9 (M/3 × N/3) sectioning images;All images are all done into above-mentioned processing, obtain three kinds
The training set of scale, scale are respectively (M × N) (M/2 × N/2) (M/3 × N/3).
4. method as claimed in claim 3, which is characterized in that in the step 3, be based on deep neural network VGG-16 net
Network model construction neural network is the neural network initiation parameter using the parameter that Image-Net training obtains, using more
The data set and its label training network model, learning model parameter of scale fragmentation processing freeze in the learning process
Convolution layer parameter only optimizes adjustment to full connection layer parameter, if former label is y, deep neural network model is obtained defeated
It is out a, wherein x indicates that sample, n indicate total sample number, and obtained cross entropy is C
Then tuning is carried out to obtained deep neural network, the optimization method is
Wherein, σ ' (z)=σ (z) (1- σ (z)), σ (z) are activation primitives, and z is the input X of neural network nodejWith network weight
ωjInner product.
5. method as claimed in claim 4, which is characterized in that in the step 4, after deep neural network study convergence,
Network the last layer is removed, the image input network of three groups of different scales is obtained into the sparse features vector of 4096 dimensions, by institute
Vector is stated as input, constructs MLP Fusion Features frame, the MLP Fusion Features frame includes an input layer, and two are hidden
Layer and an output layer, wherein input layer includes 12288 neurons, and first hidden layer includes connecting entirely for 4096 neurons
Layer is connect, second hidden layer is the full articulamentum comprising 4096 neurons, and output layer is flexible maximum value transfer function layer, mind
It is output classification number through first number, uses the cross entropy as loss function, carried out more using stochastic gradient descent algorithm
Newly, convergence obtains fusion feature and recognition result.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810660699.2A CN108960310A (en) | 2018-06-25 | 2018-06-25 | A kind of agricultural pest recognition methods based on artificial intelligence |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810660699.2A CN108960310A (en) | 2018-06-25 | 2018-06-25 | A kind of agricultural pest recognition methods based on artificial intelligence |
Publications (1)
Publication Number | Publication Date |
---|---|
CN108960310A true CN108960310A (en) | 2018-12-07 |
Family
ID=64486321
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810660699.2A Pending CN108960310A (en) | 2018-06-25 | 2018-06-25 | A kind of agricultural pest recognition methods based on artificial intelligence |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108960310A (en) |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110084145A (en) * | 2019-04-08 | 2019-08-02 | 华北水利水电大学 | A kind of multiple dimensioned identifying system of pest and disease damage time-frequency domain and operating method based on TensorFlow |
CN110263644A (en) * | 2019-05-21 | 2019-09-20 | 华南师范大学 | Classifying Method in Remote Sensing Image, system, equipment and medium based on triplet's network |
CN110414615A (en) * | 2019-08-02 | 2019-11-05 | 中国科学院合肥物质科学研究院 | Image is repaired based on interim depth and improves the corn Spodopterafrugiperda detection method of Double-DQN technology |
CN111783901A (en) * | 2020-07-13 | 2020-10-16 | 征图新视(江苏)科技股份有限公司 | Printing quality classification method based on deep learning |
CN112861712A (en) * | 2021-02-06 | 2021-05-28 | 郑州师范学院 | Agricultural pest and disease monitoring method based on artificial intelligence and multi-temporal remote sensing |
CN117576467A (en) * | 2023-11-22 | 2024-02-20 | 安徽大学 | Crop disease image identification method integrating frequency domain and spatial domain information |
Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN1557090A (en) * | 2001-09-21 | 2004-12-22 | ������������ʽ���� | Threshold value matrix creating method, image output system, storage medium, gradation reproducing method, threshold value matrix, image processing method, image processing apparatus, image forming ap |
WO2015030689A2 (en) * | 2013-08-27 | 2015-03-05 | Altun Halis | A tool and method for robust, scale and orientation invariant object detection and classification |
CN105825235A (en) * | 2016-03-16 | 2016-08-03 | 博康智能网络科技股份有限公司 | Image identification method based on deep learning of multiple characteristic graphs |
CN106708782A (en) * | 2016-12-22 | 2017-05-24 | 贾翔 | Regional pest detection diagnosis judging method based on wavelet analysis |
CN106991666A (en) * | 2017-02-24 | 2017-07-28 | 中国科学院合肥物质科学研究院 | A kind of disease geo-radar image recognition methods suitable for many size pictorial informations |
CN107798356A (en) * | 2017-11-24 | 2018-03-13 | 郑州大学西亚斯国际学院 | Crop leaf disease recognition method based on depth convolutional neural networks |
CN107886127A (en) * | 2017-11-10 | 2018-04-06 | 深圳市唯特视科技有限公司 | A kind of histopathology image classification method based on convolutional neural networks |
CN107967454A (en) * | 2017-11-24 | 2018-04-27 | 武汉理工大学 | Take the two-way convolutional neural networks Classification in Remote Sensing Image method of spatial neighborhood relation into account |
-
2018
- 2018-06-25 CN CN201810660699.2A patent/CN108960310A/en active Pending
Patent Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN1557090A (en) * | 2001-09-21 | 2004-12-22 | ������������ʽ���� | Threshold value matrix creating method, image output system, storage medium, gradation reproducing method, threshold value matrix, image processing method, image processing apparatus, image forming ap |
WO2015030689A2 (en) * | 2013-08-27 | 2015-03-05 | Altun Halis | A tool and method for robust, scale and orientation invariant object detection and classification |
CN105825235A (en) * | 2016-03-16 | 2016-08-03 | 博康智能网络科技股份有限公司 | Image identification method based on deep learning of multiple characteristic graphs |
CN106708782A (en) * | 2016-12-22 | 2017-05-24 | 贾翔 | Regional pest detection diagnosis judging method based on wavelet analysis |
CN106991666A (en) * | 2017-02-24 | 2017-07-28 | 中国科学院合肥物质科学研究院 | A kind of disease geo-radar image recognition methods suitable for many size pictorial informations |
CN107886127A (en) * | 2017-11-10 | 2018-04-06 | 深圳市唯特视科技有限公司 | A kind of histopathology image classification method based on convolutional neural networks |
CN107798356A (en) * | 2017-11-24 | 2018-03-13 | 郑州大学西亚斯国际学院 | Crop leaf disease recognition method based on depth convolutional neural networks |
CN107967454A (en) * | 2017-11-24 | 2018-04-27 | 武汉理工大学 | Take the two-way convolutional neural networks Classification in Remote Sensing Image method of spatial neighborhood relation into account |
Non-Patent Citations (9)
Title |
---|
YONG HOU 等: "SVM Based MLP Neural Network Algorithm and Application in Intrusion Detection", 《SPRINGER》 * |
李向荣 等: "基于快速傅里叶变换的条纹图像处理研究", 《光学技术》 * |
杨东旭 等: "基于改进DCNN结合迁移学习的图像分类方法", 《新疆大学学报(自然科学版)》 * |
王俊杰: "优化BP神经网络在手写体数字识别中的性能研究", 《电子设计工程》 * |
王文朋 等: "基于深度迁移学习的烟雾识别方法", 《计算机应用》 * |
王柯力 等: "基于迁移学习的水产动物图像识别方法", 《计算机应用》 * |
王粲 等: "基于卷积神经网络提取多尺度分层特征识别玉米杂草", 《农业工程学报》 * |
邹修国: "基于计算机视觉的农作物病虫害识别研究现状", 《计算机系统应用》 * |
陈慧岩 等主编: "《无人驾驶车辆理论与设计》", 31 March 2018 * |
Cited By (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110084145A (en) * | 2019-04-08 | 2019-08-02 | 华北水利水电大学 | A kind of multiple dimensioned identifying system of pest and disease damage time-frequency domain and operating method based on TensorFlow |
CN110084145B (en) * | 2019-04-08 | 2023-05-12 | 华北水利水电大学 | TensorFlow-based pest and disease damage time-frequency domain multi-scale identification system and operation method |
CN110263644A (en) * | 2019-05-21 | 2019-09-20 | 华南师范大学 | Classifying Method in Remote Sensing Image, system, equipment and medium based on triplet's network |
CN110263644B (en) * | 2019-05-21 | 2021-08-10 | 华南师范大学 | Remote sensing image classification method, system, equipment and medium based on triplet network |
CN110414615A (en) * | 2019-08-02 | 2019-11-05 | 中国科学院合肥物质科学研究院 | Image is repaired based on interim depth and improves the corn Spodopterafrugiperda detection method of Double-DQN technology |
CN110414615B (en) * | 2019-08-02 | 2021-12-10 | 中国科学院合肥物质科学研究院 | Corn Spodoptera frugiperda detection method based on staged depth restoration image |
CN111783901A (en) * | 2020-07-13 | 2020-10-16 | 征图新视(江苏)科技股份有限公司 | Printing quality classification method based on deep learning |
CN112861712A (en) * | 2021-02-06 | 2021-05-28 | 郑州师范学院 | Agricultural pest and disease monitoring method based on artificial intelligence and multi-temporal remote sensing |
CN117576467A (en) * | 2023-11-22 | 2024-02-20 | 安徽大学 | Crop disease image identification method integrating frequency domain and spatial domain information |
CN117576467B (en) * | 2023-11-22 | 2024-04-26 | 安徽大学 | Crop disease image identification method integrating frequency domain and spatial domain information |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108960310A (en) | A kind of agricultural pest recognition methods based on artificial intelligence | |
Yalcin | Plant phenology recognition using deep learning: Deep-Pheno | |
Ashqar et al. | Image-based tomato leaves diseases detection using deep learning | |
CN110148120B (en) | Intelligent disease identification method and system based on CNN and transfer learning | |
CN110309856A (en) | Image classification method, the training method of neural network and device | |
CN110929029A (en) | Text classification method and system based on graph convolution neural network | |
CN105469376B (en) | The method and apparatus for determining picture similarity | |
CN108615010A (en) | Facial expression recognizing method based on the fusion of parallel convolutional neural networks characteristic pattern | |
CN109241817A (en) | A kind of crops image-recognizing method of unmanned plane shooting | |
CN108416318A (en) | Diameter radar image target depth method of model identification based on data enhancing | |
CN112308152B (en) | Hyperspectral image ground object classification method based on spectrum segmentation and homogeneous region detection | |
CN111652247A (en) | Diptera insect identification method based on deep convolutional neural network | |
CN110472530B (en) | Retina OCT image classification method based on wavelet transformation and migration learning | |
CN111611889B (en) | Miniature insect pest recognition device in farmland based on improved convolutional neural network | |
Lim et al. | Performance effect analysis for insect classification using convolutional neural network | |
CN107545279A (en) | Image-recognizing method based on convolutional neural networks Yu Weighted Kernel signature analysis | |
KR20220004491A (en) | Artificial intelligence based tree data management system and tree data management method | |
CN110096976A (en) | Human behavior micro-Doppler classification method based on sparse migration network | |
Chakraborty et al. | Pneumonia and eye disease detection using convolutional neural networks | |
CN112380917A (en) | A unmanned aerial vehicle for crops plant diseases and insect pests detect | |
Zhu et al. | Indoor scene segmentation algorithm based on full convolutional neural network | |
CN111626969B (en) | Corn disease image processing method based on attention mechanism | |
US11150657B2 (en) | Lossy data compressor for vehicle control systems | |
Zulkeflie et al. | Evaluation of basic convolutional neural network, alexnet and bag of features for indoor object recognition | |
Sehree et al. | Olive trees cases classification based on deep convolutional neural network from unmanned aerial vehicle imagery |
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: 20181207 |
|
RJ01 | Rejection of invention patent application after publication |