CN109522797A - Rice seedling and Weeds at seedling recognition methods and system based on convolutional neural networks - Google Patents
Rice seedling and Weeds at seedling recognition methods and system based on convolutional neural networks Download PDFInfo
- Publication number
- CN109522797A CN109522797A CN201811199948.9A CN201811199948A CN109522797A CN 109522797 A CN109522797 A CN 109522797A CN 201811199948 A CN201811199948 A CN 201811199948A CN 109522797 A CN109522797 A CN 109522797A
- Authority
- CN
- China
- Prior art keywords
- seedling
- convolutional neural
- neural networks
- weeds
- model
- 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
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a kind of rice seedling based on convolutional neural networks and Weeds at seedling recognition methods and systems, which comprises obtains the color catalog image of rice seedling and Weeds at seedling；Wherein, the color catalog image of the rice seedling and Weeds at seedling has corresponding type label；The color catalog image of rice seedling and Weeds at seedling is expanded, training set and test set are formed；Rice seedling and Weeds at seedling identification model based on convolutional neural networks are constructed, the convolutional neural networks model of three kinds of heterogeneous networks depth is designed；The network parameter for adjusting the convolutional neural networks model of three kinds of network depths is trained by convolutional neural networks model of the training set to three kinds of network depth heterogeneous networks parameters, selects the highest convolutional neural networks model of recognition accuracy.The present invention can learn from the color catalog image of rice seedling and Weeds at seedling and extract to obtain the feature of strong robustness, and obtain preferable recognition effect.
Description
Technical field
It is especially a kind of based on convolutional neural networks the present invention relates to a kind of rice seedling and Weeds at seedling recognition methods
Rice seedling and Weeds at seedling recognition methods, system, computer equipment and storage medium, belong to image procossing and deep learning is led
Domain.
Background technique
Precisely spraying for pesticide can be under the premise of not influencing weeds control effect, and effectively save 40~60% pesticide is used
Amount.Rice seedling and Weeds at seedling identification are the foundations that herbicide sprays object and drug variety selection, are that weeds in paddy field is accurate
Therefore how the basis of prevention and control management quick and precisely carries out rice seedling and Weeds at seedling automatic identification is of great significance.
Due to often relying on hand-designed feature in previous rice seedling and Weeds at seedling identification process, need to enrich
Professional knowledge and devote a tremendous amount of time.The quality of feature largely will also rely on experience and fortune, often whole
The test and adjusting work of a algorithm all concentrate on this, need to have been manually done, cause high effort for.In contrast, in recent years by wide
An important insight in the deep learning theory of general concern is exactly first that the Feature Descriptor of hand design is calculated as vision
Step, often prematurely loses useful information, and study is to character representation relevant to task directly from image, than setting by hand
It is more efficient to count feature.
It is disclosed application No. is 201710102806.5 Chinese invention patent application and a kind of network is stacked based on depth
Weed images recognition methods.This includes: to be collected to training image based on the weed images recognition methods that depth stacks network
And pretreatment；It constructs and depth is trained to stack network model；Depth of the test sample input after training is stacked into network mould
Type carries out the automatic identification of weed images.But depth stacks the image data that network architecture is directed to large sample, has instruction
Practice the disadvantages of time is long and robustness is not strong.
Summary of the invention
The first purpose of this invention is the defect in order to solve the above-mentioned prior art, is provided a kind of based on convolutional Neural
The rice seedling of network and Weeds at seedling recognition methods, this method can be from rice seedling and the color catalog images of Weeds at seedling
Middle study and the feature for extracting to obtain strong robustness, and obtain preferable recognition effect.
Second object of the present invention is to provide a kind of rice seedling based on convolutional neural networks and Weeds at seedling is known
Other system.
Third object of the present invention is to provide a kind of computer equipment.
Fourth object of the present invention is to provide a kind of storage medium.
The first purpose of this invention can be reached by adopting the following technical scheme that:
A kind of rice seedling and Weeds at seedling recognition methods based on convolutional neural networks, which comprises
Obtain the color catalog image of rice seedling and Weeds at seedling；Wherein, the coloured silk of the rice seedling and Weeds at seedling
Colo(u)r atlas image has corresponding type label；
The color catalog image of rice seedling and Weeds at seedling is expanded, training set and test set are formed；
Rice seedling and Weeds at seedling identification model based on convolutional neural networks are constructed, three kinds of heterogeneous networks depth are designed
Convolutional neural networks model；
The network parameter for adjusting the convolutional neural networks model of three kinds of network depths, by training set to three kinds of network depths
The convolutional neural networks model of heterogeneous networks parameter is trained, and selects the highest convolutional neural networks model of recognition accuracy.
Further, the rice seedling and Weeds at seedling identification model of the building based on convolutional neural networks, design three
The convolutional neural networks model of kind heterogeneous networks depth, specifically includes:
Construct rice seedling and Weeds at seedling identification model based on convolutional neural networks；Wherein, the rice seedling and
Weeds at seedling identification model includes importation, feature learning part and classified part, and the feature learning part includes four layers
Convolution unit, preceding two layers of convolution unit include convolutional layer, Normalization layers of Batch, ReLU function and pond layer, and rear two
Layer convolution unit includes convolutional layer, Normalization layers of Batch and ReLU function, and the classified part includes full connection
Layer, softmax classification function and output layer composition；
Design convolutional layer in the convolutional neural networks model, every layer of convolution unit that convolutional layer is three layers in every layer of convolution unit
The convolutional neural networks model for being five layers for convolutional layer in four layers of convolutional neural networks models and every layer of convolution unit.
Further, the network parameter of the convolutional neural networks model for adjusting three kinds of network depths, passes through training set
The convolutional neural networks model of three kinds of network depth heterogeneous networks parameters is trained, the highest convolution of recognition accuracy is selected
Neural network model specifically includes:
For the convolutional neural networks model of every kind of network depth, taper off rule in given section by convolution kernel size
Rule, convolution kernel network parameter is adjusted in progressive law to convolution nuclear volume at double；
The convolutional neural networks model of every kind of network depth is combined according to different convolution kernel size and number
To different convolution kernel network parameters, to construct the convolutional neural networks model of heterogeneous networks parameter；
It is trained by convolutional neural networks model of the training set to three kinds of network depth heterogeneous networks parameters, selects knowledge
The other highest convolutional neural networks model of accuracy rate.
Further, the convolution kernel size is selected in [11,9,7,5,3,1], and the convolution nuclear volume exists
It is selected in [16,32,64,128,256,512].
Second object of the present invention can be reached by adopting the following technical scheme that:
A kind of rice seedling and Weeds at seedling identifying system based on convolutional neural networks, the system comprises:
Module is obtained, for obtaining the color catalog image of rice seedling and Weeds at seedling；Wherein, the rice seedling and
The color catalog image of Weeds at seedling has corresponding type label；
Expand module, expanded for the color catalog image to rice seedling and Weeds at seedling, formed training set and
Test set；
Module is constructed, for constructing rice seedling and Weeds at seedling identification model based on convolutional neural networks, design three
The convolutional neural networks model of kind heterogeneous networks depth；
Training module, the network parameter of the convolutional neural networks model for adjusting three kinds of network depths, passes through training set
The convolutional neural networks model of three kinds of network depth heterogeneous networks parameters is trained, the highest convolution of recognition accuracy is selected
Neural network model.
Further, the building module, specifically includes:
Construction unit, for constructing rice seedling and Weeds at seedling identification model based on convolutional neural networks；Wherein, institute
It states rice seedling and Weeds at seedling identification model includes importation, feature learning part and classified part, the feature learning
Part includes four layers of convolution unit, and preceding two layers of convolution unit includes convolutional layer, Normalization layers of Batch, ReLU function
With pond layer, rear two layers of convolution unit includes convolutional layer, Normalization layers of Batch and ReLU function, the division
Divide including full articulamentum, softmax classification function and output layer composition；
Design cell, for designing the volume of convolutional neural networks model, every layer that convolutional layer is three layers in every layer of convolution unit
The convolution mind that convolutional layer is five layers in convolutional layer is four layers in product unit convolutional neural networks model and every layer of convolution unit
Through network model.
Further, the training module, specifically includes:
Unit is adjusted, for the convolutional neural networks model for every kind of network depth, presses convolution kernel in given section
Size tapers off, and regular, convolution kernel network parameter is adjusted in progressive law to convolution nuclear volume at double；
Assembled unit, for the convolutional neural networks model for every kind of network depth, according to different convolution kernel sizes
And quantity, combination obtains different convolution kernel network parameters, to construct the convolutional neural networks model of heterogeneous networks parameter；
Training unit, for by training set to the convolutional neural networks models of three kinds of network depth heterogeneous networks parameters into
Row training, selects the highest convolutional neural networks model of recognition accuracy.
Further, the convolution kernel size is selected in [11,9,7,5,3,1], and the convolution nuclear volume exists
It is selected in [16,32,64,128,256,512].
Third object of the present invention can be reached by adopting the following technical scheme that:
A kind of computer equipment, including processor and for the memory of storage processor executable program, the place
When managing the program of device execution memory storage, above-mentioned rice seedling and Weeds at seedling recognition methods are realized.
Fourth object of the present invention can be reached by adopting the following technical scheme that:
A kind of storage medium is stored with program, when described program is executed by processor, realizes above-mentioned rice seedling and seedling
Phase weed identification method.
The present invention have compared with the existing technology it is following the utility model has the advantages that
The present invention can reduce people to avoid needing to be split background before manual feature extraction and artificial selection feature
Work selection and the uncertainty for extracting feature, model strong robustness can be realized and carry out accurately to rice seedling and Weeds at seedling
Identification；The recognition methods of convolutional neural networks is introduced into rice seedling and Weeds at seedling identifies among this particular problem, is had
Effect improves the discrimination of rice seedling and Weeds at seedling, realizes the automatic identification to rice seedling and Weeds at seedling, is water
The automatic identification of rice sprouts and Weeds at seedling is laid a good foundation.
Detailed description of the invention
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this
Some embodiments of invention for those of ordinary skill in the art without creative efforts, can be with
The structure shown according to these attached drawings obtains other attached drawings.
Fig. 1 is the rice seedling based on convolutional neural networks of the embodiment of the present invention 1 and the stream of Weeds at seedling recognition methods
Cheng Tu.
Fig. 2 a is the rice seedling image schematic diagram of the embodiment of the present invention 1.
Fig. 2 b is the alternanthera philoxeroides image schematic diagram of the rice seedling of the embodiment of the present invention 1.
Fig. 2 c is the Eclipta prostrata image schematic diagram of the rice seedling of the embodiment of the present invention 1.
Fig. 2 d is the false loosestrife image schematic diagram of the rice seedling of the embodiment of the present invention 1.
Fig. 2 e is the angustifolia arrowhead herb image schematic diagram of the rice seedling of the embodiment of the present invention 1.
Fig. 2 f is the barnyard grass image schematic diagram of the rice seedling of the embodiment of the present invention 1.
Fig. 2 g is a thousand pieces of gold subgraph schematic diagram of the rice seedling of the embodiment of the present invention 1.
Fig. 3 is the convolutional neural networks prototype network structure chart of the embodiment of the present invention 1.
Fig. 4 is the convolutional neural networks model training schematic diagram of the embodiment of the present invention 1.
Fig. 5 is the rice seedling based on convolutional neural networks of the embodiment of the present invention 2 and the knot of Weeds at seedling identifying system
Structure block diagram.
Fig. 6 is the structural block diagram of the building module of the embodiment of the present invention 2.
Fig. 7 is the structural block diagram of the training module of the embodiment of the present invention 2.
Fig. 8 is the structural block diagram of the computer equipment of the embodiment of the present invention 3.
Specific embodiment
In order to make the object, technical scheme and advantages of the embodiment of the invention clearer, below in conjunction with the embodiment of the present invention
In attached drawing, technical scheme in the embodiment of the invention is clearly and completely described, it is clear that described embodiment is
A part of the embodiments of the present invention, instead of all the embodiments, based on the embodiments of the present invention, ordinary skill people
Member's every other embodiment obtained without creative efforts, shall fall within the protection scope of the present invention.
Embodiment 1:
As shown in Figure 1, present embodiments providing a kind of rice seedling based on convolutional neural networks and Weeds at seedling identification
Method, method includes the following steps:
S1, the color catalog image for obtaining rice seedling and Weeds at seedling.
The present embodiment first obtains initial rice seedling and Weeds at seedling color catalog image (i.e. RGB sample image), can
To obtain by acquisition, such as the color catalog image of rice seedling and Weeds at seedling is shot by camera, it can also be from number
It is obtained according to library lookup, such as in advance in databases water storage rice sprouts and the color catalog image of Weeds at seedling, from database
The color catalog image of middle search rice seedling and Weeds at seedling can be obtained.
After obtaining initial rice seedling and Weeds at seedling color catalog image, it can be made by way of handmarking
Initial rice seedling and Weeds at seedling color catalog image enclose corresponding type label.
Rice seedling image is shown in Fig. 2 a, Fig. 2 b~Fig. 2 g shows six kinds of weed images of rice seedling, respectively
It is alternanthera philoxeroides, Eclipta prostrata, false loosestrife, angustifolia arrowhead herb, barnyard grass and a thousand pieces of gold subgraph, while it is desirable to which the image of acquisition is cromogram
Picture, but the disclosure of patent application for convenience, are black white images shown in Fig. 2 a~Fig. 2 g.
S2, the color catalog image of rice seedling and Weeds at seedling is expanded, forms training set and test set.
S3, rice seedling and Weeds at seedling identification model of the building based on convolutional neural networks, design three kinds of heterogeneous networks
The convolutional neural networks model of depth.
Specifically, step S3 includes:
S301, building are based on the rice of convolutional neural networks (Convolutional Neural Network, abbreviation CNN)
Rice shoot and Weeds at seedling identification model.
In the present embodiment, rice seedling and Weeds at seedling identification model based on convolutional neural networks, as convolutional Neural
Network model, structure is as shown in figure 3, include importation (input layer), feature learning part and classified part, feature learning
Part includes four layers of convolution unit, and preceding two layers of convolution unit includes convolutional layer, Batch Normalization (batch standardization)
Layer, ReLU (Rectified linear unit corrects linear unit) function and pond layer, rear two layers of convolution unit include volume
Lamination, Normalization layers of Batch and ReLU function, classified part include full articulamentum, softmax classification function and defeated
Layer forms out.
It is special that the convolution kernel that the convolutional layer mainly passes through different number and size carries out convolution operation extraction to input picture
Sign, shown in convolution process such as following formula (1):
In formula,Indicate that the ith feature figure to l layers of convolutional layer, f () indicate activation primitive, MiIndicate convolutional layer input
The quantity of characteristic pattern,Indicate convolution kernel,For biasing.
The pond layer is not under the premise of changing characteristic pattern quantity, by not being overlapped convolution to characteristic pattern, in turn
Achieve the purpose that reduce aspect of model parameter, shown in pond process such as following formula (2):
In formula, s indicates the convolution mask size of down-sampling,Indicate the biasing of convolution mask.
Described Batch Normalization layers use Batch Normalization algorithm, to solve in convolution mind
When through network model using the training of mini-batch stochastic gradient descent algorithm, convolutional neural networks model need to constantly adjust network
Parameter adapts to the problems such as slowing down and be fitted hardly possible under the changes in distribution bring gradient because of each convolution unit input data, specific mistake
Journey is as follows:
The mean μ and variances sigma for calculating n sample in each batch first, such as following formula (3) and (4):
Then data are normalized, as shown in following formula (5):
In formula, ε is constant, is to avoid causing formula invalid when variances sigma is 0；For retain primitive character distribution,
It is realized by restructuring transformation, detailed process is as follows:
yi=γixi+βi (6)
βi=E [xi] (8)
In formula, γiAnd βiIt is obtained by model training, Var is variance function, and E is mean value.
Activation primitive of the ReLU function as neuron, specifically:
S302, the convolutional neural networks model for designing three kinds of heterogeneous networks depth.
In the present embodiment, the convolutional neural networks model main distinction of three kinds of heterogeneous networks depth is in feature learning part
The number of plies of the convolutional layer of every layer of convolution unit is different, specifically, designs the convolutional neural networks model, volume that convolutional layer is three layers
The convolutional neural networks model that the convolutional neural networks model and convolutional layer that lamination is four layers are five layers；As shown in table 1 below,
Middle C indicates convolution kernel, the digital representation convolution kernel size before C, convolution nucleus number is indicated after C, as 11C16 indicates that convolutional layer is
The convolution kernel of 16 11*11；Wherein MP indicates pond layer, as MP2 indicates that layer template in pond is 2*2.
Rice seedling and Weeds at seedling the training accuracy rate of 1 different depth network structure of table
S4, adjust three kinds of network depths convolutional neural networks model network parameter, by training set to each network
The convolutional neural networks model of parameter is trained, and selects the highest convolutional neural networks model of recognition accuracy.
Specifically, step S4 includes:
S401, for the convolutional neural networks model of every kind of network depth, presented in given section by convolution kernel size
Subtract rule, convolution kernel network parameter is adjusted in progressive law to convolution nuclear volume at double.
The convolution kernel size of the present embodiment can be selected in [11,9,7,5,3,1], and convolution nuclear volume can be
It is selected in [16,32,64,128,256,512].
S402, for the convolutional neural networks model of every kind of network depth, according to different convolution kernel size and number, group
Conjunction obtains different convolution kernel network parameters, to construct the convolutional neural networks model of heterogeneous networks parameter.
The present embodiment is illustrated so that convolutional layer is four layers of convolutional neural networks model as an example, [16,32,64,128,
256,512] [16-32-64-128], [32-64-128-256] number different with three kinds of [64-128-256-512] are selected in region
The network parameter combination for measuring convolution kernel, selects [11-9-7-5], [9-7-5-3] and [7-5- in [11,9,7,5,3,1] region
3-1] 3 kinds of convolution kernel size network parameter combinations, in the case where convolutional layer is four layers of convolutional neural networks model, according to the above network
Parameter combination has constructed nine kinds of convolutional neural networks models, the network parameter of nine kinds of convolutional neural networks models such as the following table 2 institute
Show.
The rice seedling and Weeds at seedling training accuracy rate of the different convolution nuclear parameters of table 2
S403, it is trained by convolutional neural networks model of the training set to three kinds of network depth heterogeneous networks parameters,
Select the highest convolutional neural networks model of recognition accuracy.
By taking convolutional layer is four layers of convolutional neural networks model as an example, other each layer network ginsengs in convolutional neural networks model
In the case of number is kept fixed, the convolutional neural networks model of heterogeneous networks parameter is trained one by one, the convolution of the present embodiment
Neural network model training is as shown in Figure 4.
It can be seen that, the training accuracy rate highest of the first model, i.e. recognition accuracy highest are first selected from upper table 2
The model；To convolutional layer be three layers convolutional neural networks model and convolutional layer be five layers convolutional neural networks model carry out
It is same to handle, the highest model of recognition accuracy is selected in the convolutional neural networks model of each leisure heterogeneous networks parameter, it will
Three is compared, and finally selects the highest convolutional neural networks model of recognition accuracy, can see in Cong Shangbiao 1, convolution
Layer is the model recognition accuracy highest selected in five layers of convolutional neural networks model, which is used for later rice seedling
In seedling and Weeds at seedling identification.
It will be understood by those skilled in the art that journey can be passed through by implementing the method for the above embodiments
Sequence is completed to instruct relevant hardware, and corresponding program can store in computer readable storage medium.
It should be noted that this is not although describing the method operation of above-described embodiment in the accompanying drawings with particular order
It is required that hint must execute these operations in this particular order, could be real or have to carry out shown in whole operation
Existing desired result.On the contrary, the step of describing can change and execute sequence.Additionally or alternatively, it is convenient to omit certain steps,
Multiple steps are merged into a step to execute, and/or a step is decomposed into execution of multiple steps.
Embodiment 2:
As shown in figure 5, present embodiments providing a kind of rice seedling based on convolutional neural networks and Weeds at seedling identification
System, the system include that acquisition module, amplification module, building module and training module, the concrete function of modules are as follows:
The acquisition module, for obtaining the color catalog image of rice seedling and Weeds at seedling；Wherein, the rice seedling
The color catalog image of seedling and Weeds at seedling has corresponding type label.
The amplification module is expanded for the color catalog image to rice seedling and Weeds at seedling, forms training
Collection and test set.
The building module, for constructing rice seedling and Weeds at seedling identification model based on convolutional neural networks, if
Count the convolutional neural networks model of three kinds of heterogeneous networks depth；The building module is as shown in fig. 6, specifically include:
Construction unit, for constructing rice seedling and Weeds at seedling identification model based on convolutional neural networks；Wherein, institute
It states rice seedling and Weeds at seedling identification model includes importation, feature learning part and classified part, the feature learning
Part includes four layers of convolution unit, and preceding two layers of convolution unit includes convolutional layer, Normalization layers of Batch, ReLU function
With pond layer, rear two layers of convolution unit includes convolutional layer, Normalization layers of Batch and ReLU function, the division
Divide including full articulamentum, softmax classification function and output layer composition.
Design cell, for designing the volume of convolutional neural networks model, every layer that convolutional layer is three layers in every layer of convolution unit
The convolution mind that convolutional layer is five layers in convolutional layer is four layers in product unit convolutional neural networks model and every layer of convolution unit
Through network model.
The training module, the network parameter of the convolutional neural networks model for adjusting three kinds of network depths, passes through instruction
Practice collection to be trained the convolutional neural networks model of three kinds of network depth heterogeneous networks parameters, it is highest to select recognition accuracy
Convolutional neural networks model；The training module is as shown in fig. 7, specifically include:
Unit is adjusted, for the convolutional neural networks model for every kind of network depth, presses convolution kernel in given section
Size tapers off, and regular, convolution kernel network parameter is adjusted in progressive law to convolution nuclear volume at double；Wherein, the convolution kernel
Size is selected in [11,9,7,5,3,1], and the convolution nuclear volume is selected in [16,32,64,128,256,512]
It selects.
Assembled unit, for the convolutional neural networks model for every kind of network depth, according to different convolution kernel sizes
And quantity, combination obtains different convolution kernel network parameters, to construct the convolutional neural networks model of heterogeneous networks parameter.
Training unit, for by training set to the convolutional neural networks models of three kinds of network depth heterogeneous networks parameters into
Row training, selects the highest convolutional neural networks model of recognition accuracy.
It should be noted that system provided by the above embodiment is only illustrated with the division of above-mentioned each functional module
Illustrate, in practical applications, can according to need and be completed by different functional modules above-mentioned function distribution, i.e., by internal junction
Structure is divided into different functional modules, to complete all or part of the functions described above.
Embodiment 3:
A kind of computer equipment is present embodiments provided, which can be computer, as shown in figure 8, including
Processor, memory, input unit, display and the network interface connected by system bus.Wherein, processor is for providing
It calculates and control ability, memory includes non-volatile memory medium and built-in storage, which is stored with
Operating system, computer program and database, the built-in storage are operating system and computer in non-volatile memory medium
The operation of program provides environment, when computer program is executed by processor, realizes that the rice seedling of above-described embodiment 1 and seedling stage are miscellaneous
Careless identification side, as follows:
Obtain the color catalog image of rice seedling and Weeds at seedling；Wherein, the coloured silk of the rice seedling and Weeds at seedling
Colo(u)r atlas image has corresponding type label；
The color catalog image of rice seedling and Weeds at seedling is expanded, training set and test set are formed；
Rice seedling and Weeds at seedling identification model based on convolutional neural networks are constructed, three kinds of heterogeneous networks depth are designed
Convolutional neural networks model；
The network parameter for adjusting the convolutional neural networks model of three kinds of network depths, by training set to three kinds of network depths
The convolutional neural networks model of heterogeneous networks parameter is trained, and selects the highest convolutional neural networks model of recognition accuracy.
It is appreciated that the computer equipment of the present embodiment can also be server, mobile terminal etc..
Embodiment 4:
A kind of storage medium is present embodiments provided, which is stored with one or more programs, described program quilt
When processor executes, the rice seedling and Weeds at seedling recognition methods of above-described embodiment 1 are realized, as follows:
Obtain the color catalog image of rice seedling and Weeds at seedling；Wherein, the coloured silk of the rice seedling and Weeds at seedling
Colo(u)r atlas image has corresponding type label；
The color catalog image of rice seedling and Weeds at seedling is expanded, training set and test set are formed；
Rice seedling and Weeds at seedling identification model based on convolutional neural networks are constructed, three kinds of heterogeneous networks depth are designed
Convolutional neural networks model；
The network parameter for adjusting the convolutional neural networks model of three kinds of network depths, by training set to three kinds of network depths
The convolutional neural networks model of heterogeneous networks parameter is trained, and selects the highest convolutional neural networks model of recognition accuracy.
The storage medium of the present embodiment can be disk, CD, computer storage, read-only memory (ROM, Read-
Only Memory), random access memory (RAM, Random Access Memory), USB flash disk, the media such as mobile hard disk.
In conclusion the present invention can be to avoid needing to be split background and artificial selection is special before manual feature extraction
Sign reduces artificial selection and extracts the uncertainty of feature, and model strong robustness can be realized miscellaneous to rice seedling and seedling stage
Grass is accurately identified；The recognition methods of convolutional neural networks is introduced into rice seedling and Weeds at seedling identifies that this is specifically asked
Among topic, the discrimination of rice seedling and Weeds at seedling is effectively increased, is realized to the automatic of rice seedling and Weeds at seedling
Identification, the automatic identification for rice seedling and Weeds at seedling are laid a good foundation.
The above, only the invention patent preferred embodiment, but the scope of protection of the patent of the present invention is not limited to
This, anyone skilled in the art is in the range disclosed in the invention patent, according to the present invention the skill of patent
Art scheme and its inventive concept are subject to equivalent substitution or change, belong to the scope of protection of the patent of the present invention.
Claims (10)
1. a kind of rice seedling and Weeds at seedling recognition methods based on convolutional neural networks, which is characterized in that the method packet
It includes:
Obtain the color catalog image of rice seedling and Weeds at seedling；Wherein, the colored sample of the rice seedling and Weeds at seedling
This image has corresponding type label；
The color catalog image of rice seedling and Weeds at seedling is expanded, training set and test set are formed；
Rice seedling and Weeds at seedling identification model based on convolutional neural networks are constructed, the volume of three kinds of heterogeneous networks depth is designed
Product neural network model；
The network parameter for adjusting the convolutional neural networks model of three kinds of network depths, by training set to three kinds of network depth differences
The convolutional neural networks model of network parameter is trained, and selects the highest convolutional neural networks model of recognition accuracy.
2. rice seedling according to claim 1 and Weeds at seedling recognition methods, which is characterized in that the building is based on volume
The rice seedling and Weeds at seedling identification model of product neural network, design the convolutional neural networks mould of three kinds of heterogeneous networks depth
Type specifically includes:
Construct rice seedling and Weeds at seedling identification model based on convolutional neural networks；Wherein, the rice seedling and seedling stage
Weed identification model includes importation, feature learning part and classified part, and the feature learning part includes four layers of convolution
Unit, preceding two layers of convolution unit include convolutional layer, Normalization layers of Batch, ReLU function and pond layer, and latter two layers volume
Product unit includes convolutional layer, Normalization layers of Batch and ReLU function, the classified part include full articulamentum,
Softmax classification function and output layer composition；
Designing convolutional layer in the convolutional neural networks model, every layer of convolution unit that convolutional layer in every layer of convolution unit is three layers is four
The convolutional neural networks model that convolutional layer is five layers in the convolutional neural networks model of layer and every layer of convolution unit.
3. rice seedling according to claim 1 or 2 and Weeds at seedling recognition methods, which is characterized in that described to adjust three
The network parameter of the convolutional neural networks model of kind of network depth, by training set to three kinds of network depth heterogeneous networks parameters
Convolutional neural networks model is trained, and is selected the highest convolutional neural networks model of recognition accuracy, is specifically included:
For the convolutional neural networks model of every kind of network depth, taper off rule, volume in given section by convolution kernel size
Convolution kernel network parameter is adjusted in progressive law to product nuclear volume at double；
For the convolutional neural networks model of every kind of network depth, according to different convolution kernel size and number, combination is obtained not
Same convolution kernel network parameter, to construct the convolutional neural networks model of heterogeneous networks parameter；
It is trained by convolutional neural networks model of the training set to three kinds of network depth heterogeneous networks parameters, it is quasi- to select identification
The highest convolutional neural networks model of true rate.
4. rice seedling according to claim 3 and Weeds at seedling recognition methods, it is characterised in that: the convolution kernel size
It is selected in [11,9,7,5,3,1], the convolution nuclear volume is selected in [16,32,64,128,256,512].
5. a kind of rice seedling and Weeds at seedling identifying system based on convolutional neural networks, which is characterized in that the system packet
It includes:
Module is obtained, for obtaining the color catalog image of rice seedling and Weeds at seedling；Wherein, the rice seedling and seedling stage
The color catalog image of weeds has corresponding type label；
Module is expanded, is expanded for the color catalog image to rice seedling and Weeds at seedling, training set and test are formed
Collection；
Module is constructed, for constructing rice seedling and Weeds at seedling identification model based on convolutional neural networks, three kinds of design is not
With the convolutional neural networks model of network depth；
Training module, the network parameter of the convolutional neural networks model for adjusting three kinds of network depths, by training set to three
The convolutional neural networks model of kind network depth heterogeneous networks parameter is trained, and selects the highest convolutional Neural of recognition accuracy
Network model.
6. rice seedling according to claim 5 and Weeds at seedling identifying system, which is characterized in that the building module,
It specifically includes:
Construction unit, for constructing rice seedling and Weeds at seedling identification model based on convolutional neural networks；Wherein, the water
Rice sprouts and Weeds at seedling identification model include importation, feature learning part and classified part, the feature learning part
Including four layers of convolution unit, preceding two layers of convolution unit includes convolutional layer, Normalization layers of Batch, ReLU function and pond
Change layer, rear two layers of convolution unit includes convolutional layer, Normalization layers of Batch and ReLU function, the division subpackage
Include full articulamentum, softmax classification function and output layer composition；
Design cell, for designing the convolutional neural networks model, every layer of convolution list that convolutional layer is three layers in every layer of convolution unit
The convolutional Neural net that convolutional layer is five layers in convolutional layer is four layers in member convolutional neural networks model and every layer of convolution unit
Network model.
7. rice seedling according to claim 5 or 6 and Weeds at seedling identifying system, which is characterized in that the trained mould
Block specifically includes:
Unit is adjusted, for the convolutional neural networks model for every kind of network depth, presses convolution kernel size in given section
Tapering off, regular, convolution kernel network parameter is adjusted in progressive law to convolution nuclear volume at double；
Assembled unit, for the convolutional neural networks model for every kind of network depth, according to different convolution kernel size sum numbers
Amount, combination obtains different convolution kernel network parameters, to construct the convolutional neural networks model of heterogeneous networks parameter；
Training unit, for being instructed by convolutional neural networks model of the training set to three kinds of network depth heterogeneous networks parameters
Practice, selects the highest convolutional neural networks model of recognition accuracy.
8. rice seedling according to claim 7 and Weeds at seedling identifying system, it is characterised in that: the convolution kernel size
It is selected in [11,9,7,5,3,1], the convolution nuclear volume is selected in [16,32,64,128,256,512].
9. a kind of computer equipment, including processor and for the memory of storage processor executable program, feature exists
In, when the processor executes the program of memory storage, the described in any item rice seedlings of realization claim 1-4 and seedling stage
Weed identification method.
10. a kind of storage medium, is stored with program, which is characterized in that when described program is executed by processor, realize claim
The described in any item rice seedlings of 1-4 and Weeds at seedling recognition methods.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811199948.9A CN109522797A (en) | 2018-10-16 | 2018-10-16 | Rice seedling and Weeds at seedling recognition methods and system based on convolutional neural networks |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811199948.9A CN109522797A (en) | 2018-10-16 | 2018-10-16 | Rice seedling and Weeds at seedling recognition methods and system based on convolutional neural networks |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109522797A true CN109522797A (en) | 2019-03-26 |
Family
ID=65770852
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811199948.9A Pending CN109522797A (en) | 2018-10-16 | 2018-10-16 | Rice seedling and Weeds at seedling recognition methods and system based on convolutional neural networks |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109522797A (en) |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110210355A (en) * | 2019-05-24 | 2019-09-06 | 华南农业大学 | Weeds in paddy field category identification method and system, target position detection method and system |
| CN110807425A (en) * | 2019-11-04 | 2020-02-18 | 金陵科技学院 | Intelligent weeding system and weeding method |
| CN110929688A (en) * | 2019-12-10 | 2020-03-27 | 齐齐哈尔大学 | Construction method and acceleration method of rice weed recognition acceleration system |
| CN111126222A (en) * | 2019-12-16 | 2020-05-08 | 山东工商学院 | Plug seedling hole identification method based on neural network and plug seedling supplementing system |
| CN111428798A (en) * | 2020-03-30 | 2020-07-17 | 北京工业大学 | Plant seedling classification method based on convolutional neural network |
| CN111553258A (en) * | 2020-04-26 | 2020-08-18 | 江苏大学 | Tea garden identification weeding method using convolutional neural network |
| CN113486773A (en) * | 2021-07-01 | 2021-10-08 | 山东大学 | Cotton plant growth period identification method, system, storage medium and equipment |
| CN113610035A (en) * | 2021-08-16 | 2021-11-05 | 华南农业大学 | Rice tillering stage weed segmentation and identification method based on improved coding and decoding network |
| CN114140702A (en) * | 2021-11-17 | 2022-03-04 | 安徽荃银高科种业股份有限公司 | Method for removing sundries and improving seed production purity in hybrid rice seed production field |
| CN114596509A (en) * | 2022-03-30 | 2022-06-07 | 华南农业大学 | Machine vision-based rice seedling leaf age period identification method |
| CN114818909A (en) * | 2022-04-22 | 2022-07-29 | 北大荒信息有限公司 | Weed detection method and device based on crop growth characteristics |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107239514A (en) * | 2017-05-19 | 2017-10-10 | 邓昌顺 | A kind of plants identification method and system based on convolutional neural networks |
| CN107316289A (en) * | 2017-06-08 | 2017-11-03 | 华中农业大学 | Crop field spike of rice dividing method based on deep learning and super-pixel segmentation |
| CN108416353A (en) * | 2018-02-03 | 2018-08-17 | 华中农业大学 | Crop field spike of rice fast partition method based on the full convolutional neural networks of depth |
| CN108491848A (en) * | 2018-03-09 | 2018-09-04 | 北京大学深圳研究生院 | Image significance detection method based on depth information and device |
| CN108491765A (en) * | 2018-03-05 | 2018-09-04 | 中国农业大学 | A kind of classifying identification method and system of vegetables image |
-
2018
- 2018-10-16 CN CN201811199948.9A patent/CN109522797A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107239514A (en) * | 2017-05-19 | 2017-10-10 | 邓昌顺 | A kind of plants identification method and system based on convolutional neural networks |
| CN107316289A (en) * | 2017-06-08 | 2017-11-03 | 华中农业大学 | Crop field spike of rice dividing method based on deep learning and super-pixel segmentation |
| CN108416353A (en) * | 2018-02-03 | 2018-08-17 | 华中农业大学 | Crop field spike of rice fast partition method based on the full convolutional neural networks of depth |
| CN108491765A (en) * | 2018-03-05 | 2018-09-04 | 中国农业大学 | A kind of classifying identification method and system of vegetables image |
| CN108491848A (en) * | 2018-03-09 | 2018-09-04 | 北京大学深圳研究生院 | Image significance detection method based on depth information and device |
Non-Patent Citations (3)
| Title |
|---|
| JINGXIAN WANG 等: "CNN Transfer Learning for Automatic Image-Based Classification of Crop Disease", 《IMAGE AND GRAPHICS TECHNOLOGIES AND APPLICATIONS》 * |
| 邓向武 等: "基于多特征融合和深度置信网络的稻田苗期杂草识别", 《农业工程学报》 * |
| 高志强 等: "《深度学习从入门到实践》", 30 June 2018, 中国铁道出版社 * |
Cited By (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110210355B (en) * | 2019-05-24 | 2021-11-26 | 华南农业大学 | Paddy field weed species identification method and system and target position detection method and system |
| CN110210355A (en) * | 2019-05-24 | 2019-09-06 | 华南农业大学 | Weeds in paddy field category identification method and system, target position detection method and system |
| CN110807425A (en) * | 2019-11-04 | 2020-02-18 | 金陵科技学院 | Intelligent weeding system and weeding method |
| CN110807425B (en) * | 2019-11-04 | 2024-02-27 | 金陵科技学院 | Intelligent weeding system and weeding method |
| CN110929688A (en) * | 2019-12-10 | 2020-03-27 | 齐齐哈尔大学 | Construction method and acceleration method of rice weed recognition acceleration system |
| CN111126222A (en) * | 2019-12-16 | 2020-05-08 | 山东工商学院 | Plug seedling hole identification method based on neural network and plug seedling supplementing system |
| CN111428798A (en) * | 2020-03-30 | 2020-07-17 | 北京工业大学 | Plant seedling classification method based on convolutional neural network |
| CN111553258A (en) * | 2020-04-26 | 2020-08-18 | 江苏大学 | Tea garden identification weeding method using convolutional neural network |
| CN113486773A (en) * | 2021-07-01 | 2021-10-08 | 山东大学 | Cotton plant growth period identification method, system, storage medium and equipment |
| CN113486773B (en) * | 2021-07-01 | 2024-03-12 | 山东大学 | Cotton plant growing period identification method, system, storage medium and equipment |
| CN113610035A (en) * | 2021-08-16 | 2021-11-05 | 华南农业大学 | Rice tillering stage weed segmentation and identification method based on improved coding and decoding network |
| CN113610035B (en) * | 2021-08-16 | 2023-10-10 | 华南农业大学 | Rice tillering stage weed segmentation and identification method based on improved coding and decoding network |
| CN114140702A (en) * | 2021-11-17 | 2022-03-04 | 安徽荃银高科种业股份有限公司 | Method for removing sundries and improving seed production purity in hybrid rice seed production field |
| CN114596509A (en) * | 2022-03-30 | 2022-06-07 | 华南农业大学 | Machine vision-based rice seedling leaf age period identification method |
| CN114818909A (en) * | 2022-04-22 | 2022-07-29 | 北大荒信息有限公司 | Weed detection method and device based on crop growth characteristics |
| CN114818909B (en) * | 2022-04-22 | 2023-09-15 | 北大荒信息有限公司 | Weed detection method and device based on crop growth characteristics |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109522797A (en) | Rice seedling and Weeds at seedling recognition methods and system based on convolutional neural networks | |
| CN107067020B (en) | Image identification method and device | |
| CN103823845B (en) | Method for automatically annotating remote sensing images on basis of deep learning | |
| Zhao et al. | ApLeaf: An efficient android-based plant leaf identification system | |
| CN105955962B (en) | The calculation method and device of topic similarity | |
| CN104318562B (en) | A kind of method and apparatus for being used to determine the quality of the Internet images | |
| CN108875934A (en) | A kind of training method of neural network, device, system and storage medium | |
| CN110210355B (en) | Paddy field weed species identification method and system and target position detection method and system | |
| CN106815604A (en) | Method for viewing points detecting based on fusion of multi-layer information | |
| CN103761295B (en) | Automatic picture classification based customized feature extraction method for art pictures | |
| CN106844614A (en) | A kind of floor plan functional area system for rapidly identifying | |
| CN108388905B (en) | A kind of Illuminant estimation method based on convolutional neural networks and neighbourhood context | |
| CN108549658A (en) | A kind of deep learning video answering method and system based on the upper attention mechanism of syntactic analysis tree | |
| Viswanathan | Artist identification with convolutional neural networks | |
| Goëau et al. | Multi-organ plant identification | |
| CN108764192A (en) | A kind of more example multi-tag learning methods towards safe city video monitoring application | |
| CN109410190A (en) | Shaft tower based on High Resolution Remote Sensing Satellites image falls disconnected detection model training method | |
| CN113449776A (en) | Chinese herbal medicine identification method and device based on deep learning and storage medium | |
| CN112861666A (en) | Chicken flock counting method based on deep learning and application | |
| CN107818175B (en) | Legal case problem analysis method and device based on referee document | |
| O’Connor et al. | Black Holes, Black-Scholes, and Prairie Voles: An Essay Review of Simulation and Similarity, by Michael Weisberg-Michael Weisberg, Simulation and Similarity: Using Models to Understand the World. New York: Oxford University Press (2013), 224 pp., $73.00 (cloth). | |
| He et al. | Directed acyclic graphs-based diagnosis approach using small data sets for sustainability | |
| Singh et al. | Performance evaluation of plant leaf disease detection using deep learning models | |
| CN110070120A (en) | Based on the depth measure learning method and system for differentiating sampling policy | |
| CN109523509A (en) | Detection method, device and the electronic equipment of wheat heading stage |
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: 20190326 |
|
| RJ01 | Rejection of invention patent application after publication |