CN109886155A - Man power single stem rice detection localization method, system, equipment and medium based on deep learning - Google Patents
Man power single stem rice detection localization method, system, equipment and medium based on deep learning Download PDFInfo
- Publication number
- CN109886155A CN109886155A CN201910089817.3A CN201910089817A CN109886155A CN 109886155 A CN109886155 A CN 109886155A CN 201910089817 A CN201910089817 A CN 201910089817A CN 109886155 A CN109886155 A CN 109886155A
- Authority
- CN
- China
- Prior art keywords
- rice
- candidate region
- image data
- network
- man power
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Abstract
The invention discloses a kind of, and the man power single stem rice based on deep learning detects localization method, system, equipment and medium, which comprises obtains field rice sample image data；Field rice sample image data is pre-processed and marked, preprocessing image data is obtained；Establish depth convolutional neural networks detection model；Training is optimized to depth convolutional neural networks detection model using preprocessing image data；Detection positioning is carried out using man power single stem rice of the depth convolutional neural networks detection model after training to field rice testing image.The present invention is using the detection method based on deep learning in computer vision, the defect of the above-mentioned prior art can be substantially improved, pass through projected depth convolutional neural networks model, extract the high-dimensional Spatial Semantics feature of plant, still there are good positioning accuracy and robustness under complex environment, can be widely applied for the automation of agricultural, in intelligent production management.
Description
Technical field
The present invention relates to a kind of detection localization method, especially a kind of man power single stem rice based on deep learning detects positioning side
Method, system, equipment and medium belong to the object detection field of computer vision.
Background technique
Rice is one of main cereal crops in the world, and 20 world's last phase circle Monitoring of Paddy Rice Plant Area are up to 1.54 hundred million hm2.I
State's yield accounts for the 40% of total output of grain, and Rice Production is responsible for the weighty responsibility for ensuring China's grain security.Investment is reduced to produce
Compare out, the economic benefit for increasing Rice Cropping is particularly important.Weeds are used as one of the main reason for leading to the rice underproduction, with
Crops compete the resources such as moisture, nutrient, sunlight, so that the rice underproduction is serious.On the other hand, to rice carry out Precision Irrigation,
Fertilising or the do mechanization operations such as spraying pesticide, can be improved means of agricultural production utilization rate and effective percentage, reduce the means of production waste and
Therefore bring environmental pollution, protecting ecology water resource, air and soil.
Therefore, realize that Agricultural Intelligent System Operation control and management are particularly important, wherein accurate by intelligent rice
Location technology realizes that mechanical intelligent weeding, intelligence finely spray agriculture medicine fertilizer etc., and then realizes and improve rice yield, mentions
Agricultural production fining and automatization level are risen, China's agricultural sustainable development is pushed.
In recent years, precise positioning technology is widely studied and applied in agriculture field, mainly there is intelligent machine
Weeding, mechanical independent navigation, insecticidal fertilizer standard are sprayed and are picked automatically with agricultural product.Foreign study status, 2011, Xuewen
Wu etc. proposes a kind of rice detection method based on position and edge feature, will using the color difference of green plants and Soil Background
Plant splits, and crop center is determined using pixel histogram, and crop edge is that terminal fills crop area realization detection,
=this method is difficult to use in the environment of paddy field.2015, Kazmi etc. carried out fusion using color and edge feature and proposes that one kind is new
Local feature carry out weeds segmentation, combination supporting vector machine SVM classifier completes detection to rice, but this method is by nature
Complex illumination influence factor is larger in environment.Domestic research at present, 2015, Lian Ning carried out gray scale by shooting image to field
The pretreatment operations such as change, binaryzation, denoising analyze the method for being most suitable for distinguishing weeds and crop, in Image Edge-Detection
On the basis of propose based on enhanced fuzzy crop image contours extract algorithm and position.2017, Jiang Yu etc. proposed basal part of stem
Subregion divides the method for edge fitting to carry out plant positioning, solve the rice weeding phase because rice canopy in succession caused by positioning
Inaccurate problem.But this method robustness is poor, and the rice for coming in every shape and reflective ponding soil locating effect are poor.
In conclusion the method for expert's research both at home and abroad is mostly based on the low dimensionals feature such as color, shape, texture of crop and detects
Positioning also has a small number of features in conjunction with engineer to carry out detection positioning.These methods are generally susceptible to complex illumination back
The factors such as scape, plant block mutually, indenting canopy influence and cannot reach the positioning accuracy of requirement, or even mistake occurs
Positioning.
Summary of the invention
In consideration of it, the man power single stem rice detection localization method that the present invention provides a kind of based on deep learning, system, equipment and
Medium uses the detection method in computer vision based on deep learning, can substantially improve the defect of the above-mentioned prior art,
By projected depth convolutional neural networks model, the high-dimensional Spatial Semantics feature of plant is extracted, is still had under complex environment
Good positioning accuracy and robustness can be widely applied for the automation of agricultural, in intelligent production management.
The first purpose of this invention is to provide a kind of man power single stem rice detection localization method based on deep learning.
Second object of the present invention is to provide a kind of man power single stem rice detection positioning system based on deep learning.
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 man power single stem rice detection localization method based on deep learning, which comprises
Obtain field rice sample image data；
Field rice sample image data is pre-processed and marked, preprocessing image data is obtained；
Establish depth convolutional neural networks detection model；
Training is optimized to depth convolutional neural networks detection model using preprocessing image data；
It is carried out using man power single stem rice of the depth convolutional neural networks detection model after training to field rice testing image
Detection positioning.
Further, described that field rice sample image data is pre-processed and marked, it specifically includes:
Field rice sample image data is cleaned, the figure of no rice example or the visual recognition rice strain difficulty of people is abandoned
As data；
The field rice sample image data after cleaning is labeled according to the standard drafted；Wherein, described to draft
Standard are as follows: using plant stem foot center as the center of circle, define the stem foot border circular areas A that radius is γ1, circle A2It is the twice round A of radius1
Concentric circles, take round A2It is external square be used as the exemplary detection true tag of man power single stem rice.
Further, described to establish depth convolutional neural networks detection model, it specifically includes:
Image feature information sub-network is extracted into pretreatment image input after zoom operations, it is special by extracting image
Reference breath sub-network exports the depth convolution higher dimensional space characteristic information figure of the image；
Sub-network is extracted into the characteristic information figure input candidate region for extracting the output of image feature information sub-network, passes through time
Favored area extracts candidate region of the possibility comprising rice that sub-network exports R high quality；
The characteristic information figure for extracting the output of image feature information sub-network and candidate region are extracted into sub-network output
Candidate region input detection positioning sub-network, is exported in characteristic information figure where detection positioning rice by detection positioning sub-network
Position.
Further, the zoom operations include: to carry out long short side equal proportion scaling to pretreatment image, so that pretreatment
The short side of image is less than or equal to the first presetted pixel value, and long side is less than or equal to the second presetted pixel value.
Further, it is a convolutional layer that the top of sub-network is extracted in the candidate region, and two are connected after convolutional layer
Convolutional layer branch, Liang Ge convolutional layer branch are respectively used to the classification and recurrence of candidate region；
It is described that candidate region of the possibility comprising rice that sub-network exports R high quality is extracted by candidate region, specifically
Are as follows: through convolutional layer on the characteristic information figure for extracting the output of image feature information sub-network, carry out sliding convolution point by point, and
K candidate region is generated at each sliding center, K candidate region is sent into Liang Ge convolutional layer branch, classifies for candidate region
Convolutional layer branch output channel number be K × 2, indicate to candidate region be classified as foreground and background two classification scores, be used for
The convolutional layer branch output channel number that candidate region returns is K × 4, indicates four correction amounts to K candidate region bounding box；
Non- maximum is carried out to the higher preceding T candidate region of prospect class score is divided in the convolutional layer branch classified for candidate region into
Inhibit removal redundancy candidate region, the possibility for exporting less R (R < T) a high quality includes the candidate region of rice；
Wherein, K candidate region is generated, comprising: according to the width W and height H of characteristic information figure, passing through zoom operations
One anchor point is set at interval of D pixel in the row, column of input picture afterwards, in total W × H anchor point, centered on each anchor point
K candidate region is generated, K candidate region is divided into U group, every group of each candidate region size phase by size difference
Together, all anchor points collectively generate W × H × K candidate region.
Further, the top of the detection positioning sub-network is the pond of an output fixed size characteristic information figure
Layer, sequentially stacks multiple convolutional layers/full articulamentum after the layer of pond, for further extracting R candidate region characteristic information,
Two convolutional layers/full articulamentum branch, two convolutional layers/full articulamentum branch are connected after multiple convolutional layers/full articulamentum
It is respectively used to the classification and recurrence of candidate region；
It is described that detection positioning rice position in characteristic information figure is exported by detection positioning sub-network, specifically: into
R candidate region characteristic information of onestep extraction, convolutional layer/full articulamentum branch output channel number for candidate region classification are
2, indicate the score that rice and background are classified as to candidate region, the convolutional layer returned for candidate region/full articulamentum branch
Output channel number is 4 × 2, is indicated to rice and the respective four recurrence correction amount of two class candidate region bounding box of background；To category
After all candidate regions of rice category score carry out non-maxima suppression removal redundancy, final detection positioning rice is exported
Candidate region.
Further, described that instruction is optimized to depth convolutional neural networks detection model using preprocessing image data
Practice, specifically:
From preprocessing image data training set, each iteration randomly selects multiple preprocessing image datas and constitutes one batch
Secondary progress entire depth convolutional neural networks detection model parameter updates, and is carried out using backpropagation and stochastic gradient descent algorithm
Optimization training；Wherein, the optimization training process iteration carries out E times, and initial learning rate is set as lr, every step repetitive exercise
Learning rate is reduced to original 1/10th afterwards, and entire training process iteration is until loss function tends towards stability when not declining, i.e.,
Training convergence terminates.
Second object of the present invention can be reached by adopting the following technical scheme that:
A kind of man power single stem rice detection positioning system based on deep learning, the system comprises:
Module is obtained, for obtaining field rice sample image data；
Preprocessing module obtains pretreatment image for field rice sample image data to be pre-processed and marked
Data；
Module is established, for establishing depth convolutional neural networks detection model；
Training module, for optimizing instruction to depth convolutional neural networks detection model using preprocessing image data
Practice；
Detection module, for utilizing the depth convolutional neural networks detection model after training to field rice testing image
Man power single stem rice carries out detection positioning.
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 man power single stem rice detection localization method is 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 man power single stem rice detection
Localization method.
The present invention have compared with the existing technology it is following the utility model has the advantages that
1, the present invention, can be with the list in precise positioning image using the detection method based on deep learning in computer vision
Strain rice has deep directive significance the farm works such as man power single stem rice effectively being irrigated, being applied fertilizer, not only can be improved
Resource utilization, effective percentage reduce the pollution of the waste and medicine fertilizer of the means of production to environment, can with protecting ecology water resource,
Air and soil have universality and versatility, and have a wide range of applications scene.
2, automation, intelligence of the present invention by the Disciplinary Frontiers knowledge deep learning in computer vision, applied to agricultural
Change in production management, by projected depth convolutional neural networks model, extracts the high-dimensional Spatial Semantics feature of rice plant, for
Under the environment such as rice detection is located in complex illumination background, plant is blocked mutually, indenting canopy, there still have to be fixed well
Position precision and robustness is agricultural automation instantly with development potential, intelligent emerging technology.
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 that the man power single stem rice based on deep learning of the embodiment of the present invention 1 detects the flow chart of localization method.
Fig. 2 is that the man power single stem rice detection localization method based on deep learning of the embodiment of the present invention 1 is applied in field rice
The structure chart of testing image detection positioning.
Fig. 3 is the structure chart of the extraction image feature information sub-network of the embodiment of the present invention 1.
Fig. 4 is that the structure chart of sub-network is extracted in the candidate region of the embodiment of the present invention 1.
Fig. 5 is the structure chart of the detection positioning sub-network of the embodiment of the present invention 1.
Fig. 6 a~Fig. 6 d is that the man power single stem rice detection localization method based on deep learning of the embodiment of the present invention 1 is examined outdoors
Measure the result figure of position rice.
Fig. 7 is that the man power single stem rice based on deep learning of the embodiment of the present invention 2 detects the structural block diagram of positioning system.
Fig. 8 is the structural block diagram of the preprocessing module of the embodiment of the present invention 2.
Fig. 9 is the structural block diagram for establishing module of the embodiment of the present invention 2.
Figure 10 is the structural block diagram of the detection module of the embodiment of the present invention 2.
Figure 11 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 depicted in figs. 1 and 2, a kind of man power single stem rice detection localization method based on deep learning is present embodiments provided,
Method includes the following steps:
S1, field rice sample image data is obtained.
The field rice sample image data of the present embodiment can be acquired by image capturing system and be obtained, the Image Acquisition
System includes color camera, camera lens, computer, shade, mounting platform etc., and the image is installed on agriculture weeding machine and is adopted
Collecting system sets machine travel speed as 2 metre per second (m/s)s, and image capturing system frame per second is that 10 frames are per second, the fields such as simulation weeding, fertilising
Between operation process, use image capturing system acquisition different shape field rice testing image, sufficiently increase image data base
Diversity, to enhance the generalization ability and positioning accuracy of depth convolutional neural networks model.
S2, field rice sample image data is pre-processed and is marked, obtain preprocessing image data.
Step S2 is specifically included:
S201, data cleansing: cleaning field rice sample image data, the vision for abandoning no rice example or people are distinguished
Know the difficult image data of rice strain.
S202, labeled data: the field rice sample image data after cleaning is labeled according to the standard drafted；Its
In, the standard drafted are as follows: using plant stem foot center as the center of circle, define the stem foot border circular areas A that radius is γ1, circle A2It is radius two
Times circle A1Concentric circles, take round A2It is external square be used as the exemplary detection true tag of man power single stem rice.This drafts standard ginseng
It has examined paddy rice field weeding or has precisely sprayed the professional standard of medicine fertilizer.
S3, depth convolutional neural networks detection model is established.
The depth convolutional neural networks detection model of the present embodiment includes extracting image feature information sub-network, candidate region
Sub-network and detection positioning sub-network are extracted, three sub-networks constitute an entirety, make depth convolutional neural networks detection model
Detection process is completed end-to-endly.
Step S3 is specifically included:
S301, image feature information sub-network is extracted into the preprocessing image data input after zoom operations, passes through
The depth convolution higher dimensional space characteristic information figure that image feature information sub-network exports the image is extracted, size is W × H.
The extraction image feature information sub-network of the present embodiment carries out mentioning for image information using depth convolutional neural networks
It takes, wherein depth convolutional neural networks generally use classical efficient structural model, such as Oxford University's visual geometric group
The network structure models series such as VGG13, VGG16, VGG19 of (Visual Geometry Group) invention, Facebook are artificial
Intelligent study institute (FAIR) scientist He Kaiming et al. invention depth residual error network series (ResNet18, ResNet50,
ResNet101) etc..As shown in figure 3, the extraction image feature information sub-network of the present embodiment uses VGG16 model.Scaling behaviour
Make specifically: the image for typically entering neural network model is smaller, such as 224 pixels × 224 pixels, 480 pixels × 640 pixels
Etc., but in the present embodiment, long short side equal proportion is carried out to preprocessing image data and is scaled, so that short side is less than or equal to the
One presetted pixel value, and long side is less than or equal to the second presetted pixel value, and the first presetted pixel value is 600 pixels, and second is pre-
If pixel value is 1000 pixels, the benefit done so can allow the picture of input larger, finally to the effect meeting of small target deteection
More preferably.Specifically, if original image (pretreatment image) is M1×N1Pixel might as well set M1It is shorter a line, N1It is longer one
Side.According to original image M1:N1Scaling so that short side M1To 600 pixel of designated length, at this time another after scaling
Side is that length is set as N2, present image size is 600 × N2Pixel.
S302, sub-network is extracted into the characteristic information figure input candidate region for extracting the output of image feature information sub-network,
Candidate region of the possibility comprising rice that sub-network exports R=300 high quality is extracted by candidate region, it is subsequent to be supplied to
Detection positioning sub-network further detect positioning.
As shown in figure 4, it is 3 × 3 that the top of the candidate region extraction sub-network of the present embodiment, which is a convolution kernel size,
Convolutional layer, convolutional layer connect the convolutional layer branch that two convolution kernel sizes are 1 × 1 later, and Liang Ge convolutional layer branch is respectively used to
The classification and recurrence of candidate region.
The candidate region that sub-network output 300 may include rice is extracted by candidate region, specifically: pass through convolution
Layer carries out point-by-point sliding convolution on the characteristic information figure for extracting the output of image feature information sub-network, and in each sliding
4 candidate regions are generated at the heart, Liang Ge convolutional layer branch is sent into 4 candidate regions, the convolutional layer point for candidate region classification
Branch output channel number is 4 × 2, indicates the two classification scores for being classified as foreground and background to candidate region, returns for candidate region
The convolutional layer branch output channel number returned is 4 × 4, indicates four correction amounts to 4 candidate region bounding boxes；To for candidate
It divides the higher preceding T=6000 candidate region of prospect class score in the convolutional layer branch of territorial classification into and carries out non-maxima suppression
(Non-maximum suppression, abbreviation NMS) removes redundancy candidate region, exports R=300 less high quality
It may include the candidate region of rice.
Wherein, 4 candidate regions are generated, comprising: according to the width W and height H of characteristic information figure, passing through zoom operations
One anchor point is set at interval of D=16 pixel in the row, column of input picture afterwards, W × H anchor point, each anchor point are in total
4 candidate regions are centrally generated, 4 candidate region sizes of the present embodiment are identical, therefore only divide U=1 group candidate regions
Domain, 4 candidate region Aspect Ratios are 1:1, it is to be understood that, the Aspect Ratio of 4 candidate regions also can permit difference, institute
There is anchor point to collectively generate W × H × 4 candidate region.
S303, the characteristic information figure for extracting the output of image feature information sub-network and candidate region are extracted into sub-network
The candidate region input detection positioning sub-network of output, exports detection positioning water in characteristic information figure by detection positioning sub-network
Rice position, the i.e. desired output of depth convolutional neural networks detection model.
As shown in figure 5, the top of the detection positioning sub-network of the present embodiment is an output fixed size characteristic information figure
Pond layer, two full articulamentums are sequentially stacked after the layer of pond, for further extract R candidate region characteristic information,
Liang Gequan articulamentum branch is connected after two full articulamentums, Liang Gequan articulamentum branch is respectively used to the classification of candidate region
And recurrence；It is appreciated that full articulamentum and full articulamentum branch may be convolutional layer and convolutional layer branch.
Detection positioning rice position in characteristic information figure is exported by detection positioning sub-network, specifically: further
R candidate region characteristic information is extracted, convolutional layer/full articulamentum branch output channel number for candidate region classification is 2, table
Show the score that rice and background are classified as to candidate region, the convolutional layer returned for candidate region/full articulamentum branch output
Port number is 4 × 2, is indicated to rice and the respective four recurrence correction amount of two class candidate region bounding box of background；To belonging to water
All candidate regions of rice category score carry out non-maxima suppression (Non-maximum suppression, abbreviation NMS) removal
After redundancy, final detection positioning rice candidate region is exported.
S4, training is optimized to depth convolutional neural networks detection model using preprocessing image data.
Specifically, from preprocessing image data training set, each iteration randomly selects 64 preprocessing image data structures
It carries out entire depth convolutional neural networks detection model parameter at a batch to update, using backpropagation and stochastic gradient descent
Algorithm optimizes training；Wherein, the optimization training process iteration carries out E=1000 times, and initial learning rate is set as lr=
0.0001, learning rate is reduced to original 1/10th after every step=100 repetitive exercise, and momentum parameter is set as 0.9, power
0.0005 is decayed to again, extracts image feature information sub-network, detection positioning sub-network tip portion is classified using ImageNet
Pre-training model is initialized.Until loss function tends towards stability when not declining, i.e. training restrains entire training process iteration
Terminate.
S5, using the depth convolutional neural networks detection model after training to the man power single stem rice of field rice testing image into
Row detection positioning.
Step S5 is specifically included:
S501, obtain field rice testing image: use and the identical image capturing system of step S1 are equipped in agriculture
Mechanically, similarly with 2 metre per second (m/s) travel speeds, 10 frames frame per second per second acquires field rice testing image, color for industry weeding etc.
The field rice testing image of acquisition is sent in computer by form and aspect machine.
S502, field rice testing image is detected by the depth convolutional neural networks model after training, and defeated
The specific location of rice out, to realize that liquid manure, agriculture are irrigated in agricultural machinery intelligence field weeding operation and intelligentized control method
The resources such as medicine improve resource utilization, effective percentage, reduce the pollution of the waste and medicine fertilizer of the means of production to environment, protecting ecology
Water resource, air and soil.
Fig. 6 a~Fig. 6 d be the present embodiment man power single stem rice detection method outdoors crop field detect man power single stem rice result
Figure, grey black frame is the rice position and size that institute's detection positions in figure, it can be seen that it is basic to detect the rice center positioned
It is overlapped with actual plant center, and the plant of different scale within the vision is all detected to orient and, effect is very smart
Really.
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 fig. 7, present embodiments providing a kind of man power single stem rice detection positioning system based on deep learning, the system
Including obtaining module 701, preprocessing module 702, establishing module 703, training module 704 and detection module 705, modules
Concrete function is as follows:
The acquisition module 701, for obtaining field rice sample image data.
The preprocessing module 702 obtains pre- place for field rice sample image data to be pre-processed and marked
Manage image data.
Further, the preprocessing module 702 is as shown in figure 8, specifically include:
Cleaning unit 7021 abandons no rice example or the view of people for cleaning field rice sample image data
Feel the difficult image data of identification rice strain.
Unit 7022 is marked, for marking according to the standard drafted to the field rice sample image data after cleaning
Note；Wherein, the standard drafted are as follows: using plant stem foot center as the center of circle, define the stem foot border circular areas A that radius is γ1, circle
A2It is the twice round A of radius1Concentric circles, take round A2It is external square be used as the exemplary detection true tag of man power single stem rice.
It is described to establish module 703, for establishing depth convolutional neural networks detection model.
Further, described to establish module 703 as shown in figure 9, specifically including:
First I/O unit 7031 extracts figure for that will input by the preprocessing image data after zoom operations
As characteristic information sub-network, the depth convolution higher dimensional space feature that the image is exported by extracting image feature information sub-network is believed
Breath figure.
Second I/O unit 7032, the characteristic information figure for that will extract the output of image feature information sub-network are defeated
Enter candidate region and extract sub-network, candidate of the possibility comprising rice that sub-network exports R high quality is extracted by candidate region
Region.
Third I/O unit 7033, for the characteristic information figure of image feature information sub-network output will to be extracted, with
And the candidate region input detection positioning sub-network of sub-network output is extracted in candidate region, it is special by detection positioning sub-network output
Levy detection positioning rice position in hum pattern.
The training module 704, for being carried out using preprocessing image data to depth convolutional neural networks detection model
Optimization training.
The detection module 705, for being waited for using the depth convolutional neural networks detection model after training field rice
The man power single stem rice of altimetric image carries out detection positioning.
Further, the detection module 705 as shown in Figure 10, specifically includes:
Acquiring unit 7051, for obtaining field rice testing image.
Output unit 7052 is detected, for waiting for mapping to field rice by the depth convolutional neural networks model after training
As being detected, and export the specific location of rice.
The specific implementation of modules and unit may refer to above-described embodiment 1 in the present embodiment, no longer go to live in the household of one's in-laws on getting married one by one herein
It states.It should be noted that system provided by the above embodiment is only the example of the division of the above functional modules,
In practical applications, it can according to need and be completed by different functional modules above-mentioned function distribution, i.e., draw internal structure
It is divided into different functional modules, to complete all or part of the functions described above.
It is various to be appreciated that term " first ", " second " used in the system of above-described embodiment etc. can be used for describing
Unit, but these units should not be limited by these terms.These terms are only used to distinguish first unit and another unit.It lifts
For example, without departing from the scope of the invention, the first I/O unit can be known as the second input/output
Second I/O unit, can be known as the first I/O unit by unit, and similarly, the first I/O unit and
Second I/O unit both I/O unit, but it is not same I/O unit.
Embodiment 3:
As shown in figure 11, a kind of computer equipment is present embodiments provided, which can be computer, packet
Include processor 1102, memory, input unit 1103, display 1104 and the network interface connected by system bus 1101
1105.Wherein, processor 1102 calculates and control ability, memory include non-volatile memory medium 1106 and interior for providing
Memory 1107, the non-volatile memory medium 1106 are stored with operating system, computer program and database, the built-in storage
1107 provide environment, computer program quilt for the operation of operating system and computer program in non-volatile memory medium 1106
When processor 1102 executes, realize that the man power single stem rice of above-described embodiment 1 detects localization method, as follows:
Obtain field rice sample image data；
Field rice sample image data is pre-processed and marked, preprocessing image data is obtained；
Establish depth convolutional neural networks detection model；
Training is optimized to depth convolutional neural networks detection model using preprocessing image data；
It is carried out using man power single stem rice of the depth convolutional neural networks detection model after training to field rice testing image
Detection positioning.
Computer equipment described in the present embodiment can also be server or other terminal devices with computing function.
Embodiment 4:
A kind of storage medium is present embodiments provided, which is computer readable storage medium, is stored with meter
Calculation machine program when described program is executed by processor, when processor executes the computer program of memory storage, realizes above-mentioned reality
The man power single stem rice detection localization method of example 1 is applied, as follows:
Obtain field rice sample image data；
Field rice sample image data is pre-processed and marked, preprocessing image data is obtained；
Establish depth convolutional neural networks detection model；
Training is optimized to depth convolutional neural networks detection model using preprocessing image data；
It is carried out using man power single stem rice of the depth convolutional neural networks detection model after training to field rice testing image
Detection positioning.
Storage medium described in 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, mobile hard disk etc. be situated between
Matter.
In conclusion the present invention is using the detection method based on deep learning in computer vision, in precise positioning image
Man power single stem rice, have deep directive significance the farm works such as man power single stem rice effectively being irrigated, being applied fertilizer, not only can be with
Resource utilization, effective percentage are improved, the pollution of the waste and medicine fertilizer of the means of production to environment is reduced, can be provided with protecting ecology water
Source, air and soil have universality and versatility, and have a wide range of applications scene.
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 man power single stem rice based on deep learning detects localization method, which is characterized in that the described method includes:
Obtain field rice sample image data；
Field rice sample image data is pre-processed and marked, preprocessing image data is obtained；
Establish depth convolutional neural networks detection model；
Training is optimized to depth convolutional neural networks detection model using preprocessing image data；
The man power single stem rice of field rice testing image is detected using the depth convolutional neural networks detection model after training
Positioning.
2. man power single stem rice according to claim 1 detects localization method, which is characterized in that described to field rice sample graph
As data are pre-processed and are marked, specifically include:
Field rice sample image data is cleaned, the picture number of no rice example or the visual recognition rice strain difficulty of people is abandoned
According to；
The field rice sample image data after cleaning is labeled according to the standard drafted；Wherein, the standard drafted
Are as follows: using plant stem foot center as the center of circle, define the stem foot border circular areas A that radius is γ1, circle A2It is the twice round A of radius1It is same
Heart circle, takes round A2It is external square be used as the exemplary detection true tag of man power single stem rice.
3. man power single stem rice according to claim 1 detects localization method, which is characterized in that described to establish depth convolutional Neural
Network detection model, specifically includes:
Image feature information sub-network is extracted into pretreatment image input after zoom operations, by extracting characteristics of image letter
Breath sub-network exports the depth convolution higher dimensional space characteristic information figure of the image；
Sub-network is extracted into the characteristic information figure input candidate region for extracting the output of image feature information sub-network, passes through candidate regions
Extract candidate region of the possibility comprising rice that sub-network exports R high quality in domain；
The characteristic information figure for extracting the output of image feature information sub-network and candidate region are extracted to the candidate of sub-network output
It is in place to export detection positioning rice institute in characteristic information figure by detection positioning sub-network for input detection positioning sub-network in region
It sets.
4. man power single stem rice according to claim 3 detects localization method, which is characterized in that the zoom operations include: pair
Pretreatment image carries out long short side equal proportion scaling, so that the short side of pretreatment image is less than or equal to the first presetted pixel value,
And long side is less than or equal to the second presetted pixel value.
5. man power single stem rice according to claim 3 detects localization method, which is characterized in that extract subnet in the candidate region
The top of network is a convolutional layer, connects Liang Ge convolutional layer branch after convolutional layer, Liang Ge convolutional layer branch is respectively used to candidate
The classification and recurrence in region；
It is described that candidate region of the possibility comprising rice that sub-network exports R high quality is extracted by candidate region, specifically: it is logical
Convolutional layer is crossed on the characteristic information figure for extracting the output of image feature information sub-network, carries out point-by-point sliding convolution, and each
K candidate region is generated at sliding center, Liang Ge convolutional layer branch is sent into K candidate region, the volume for candidate region classification
Lamination branch output channel number is K × 2, the two classification scores that foreground and background is classified as to candidate region is indicated, for candidate
The convolutional layer branch output channel number that region returns is K × 4, indicates four correction amounts to K candidate region bounding box；To with
It divides the higher preceding T candidate region of prospect class score into the convolutional layer branch of candidate region classification and carries out non-maxima suppression
Redundancy candidate region is removed, the possibility for exporting R less high quality includes the candidate region of rice；
Wherein, K candidate region is generated, comprising: according to the width W and height H of characteristic information figure, after zoom operations
One anchor point is set at interval of D pixel in the row, column of input picture, in total W × H anchor point, generates K centered on each anchor point
A candidate region, K candidate region are divided into U group by size difference, and every group of each candidate region size is identical, institute
There is anchor point to collectively generate W × H × K candidate region.
6. man power single stem rice according to claim 3 detects localization method, which is characterized in that the detection positioning sub-network
Top is the pond layer of an output fixed size characteristic information figure, and multiple convolutional layer/Quan Lian are sequentially stacked after the layer of pond
Layer is connect, for further extracting R candidate region characteristic information, connects two convolution after multiple convolutional layers/full articulamentum
Layer/full articulamentum branch, two convolutional layers/full articulamentum branch are respectively used to the classification and recurrence of candidate region；
It is described that detection positioning rice position in characteristic information figure is exported by detection positioning sub-network, specifically: further
R candidate region characteristic information is extracted, convolutional layer/full articulamentum branch output channel number for candidate region classification is 2, table
Show the score that rice and background are classified as to candidate region, the convolutional layer returned for candidate region/full articulamentum branch output
Port number is 4 × 2, is indicated to rice and the respective four recurrence correction amount of two class candidate region bounding box of background；To belonging to water
After all candidate regions of rice category score carry out non-maxima suppression removal redundancy, it is candidate to export final detection positioning rice
Region.
7. man power single stem rice according to claim 1-6 detects localization method, which is characterized in that described to use pre- place
Reason image data optimizes training to depth convolutional neural networks detection model, specifically:
From preprocessing image data training set, each iteration randomly select multiple preprocessing image datas constitute a batches into
Row entire depth convolutional neural networks detection model parameter updates, and is optimized using backpropagation and stochastic gradient descent algorithm
Training；Wherein, the optimization training process iteration carries out E times, and initial learning rate is set as lr, every step repetitive exercise junior scholar
Habit rate is reduced to original 1/10th, entire training process iteration is trained when loss function tends towards stability and do not decline
Convergence terminates.
8. a kind of man power single stem rice based on deep learning detects positioning system, which is characterized in that the system comprises:
Module is obtained, for obtaining field rice sample image data；
Preprocessing module obtains preprocessing image data for field rice sample image data to be pre-processed and marked；
Module is established, for establishing depth convolutional neural networks detection model；
Training module, for optimizing training to depth convolutional neural networks detection model using preprocessing image data；
Detection module, for the single plant using the depth convolutional neural networks detection model after training to field rice testing image
Rice carries out detection positioning.
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 man power single stem rices detections of realization claim 1-7 are fixed
Position 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 man power single stem rices of 1-7 detect localization method.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910089817.3A CN109886155B (en) | 2019-01-30 | 2019-01-30 | Single-plant rice detection and positioning method, system, equipment and medium based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910089817.3A CN109886155B (en) | 2019-01-30 | 2019-01-30 | Single-plant rice detection and positioning method, system, equipment and medium based on deep learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109886155A true CN109886155A (en) | 2019-06-14 |
| CN109886155B CN109886155B (en) | 2021-08-10 |
Family
ID=66927519
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910089817.3A Active CN109886155B (en) | 2019-01-30 | 2019-01-30 | Single-plant rice detection and positioning method, system, equipment and medium based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109886155B (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110853080A (en) * | 2019-09-30 | 2020-02-28 | 广西慧云信息技术有限公司 | Method for measuring size of field fruit |
| CN111062871A (en) * | 2019-12-17 | 2020-04-24 | 腾讯科技（深圳）有限公司 | Image processing method and device, computer equipment and readable storage medium |
| CN111462058A (en) * | 2020-03-24 | 2020-07-28 | 华南农业大学 | Method for quickly detecting effective ears of rice |
| CN111723736A (en) * | 2020-06-19 | 2020-09-29 | 中国农业科学院农业信息研究所 | Fruit tree flowering phase monitoring method and device, computer equipment and storage medium |
| CN113129281A (en) * | 2021-04-13 | 2021-07-16 | 广西大学 | Wheat stem section parameter detection method based on deep learning |
| CN113128576A (en) * | 2021-04-02 | 2021-07-16 | 中国农业大学 | Crop row detection method and device based on deep learning image segmentation |
| CN115619286A (en) * | 2022-11-11 | 2023-01-17 | 中国农业科学院农业资源与农业区划研究所 | Method and system for evaluating sample plot quality of breeding field plot |
| CN116863340A (en) * | 2023-08-16 | 2023-10-10 | 安徽荃银超大种业有限公司 | Rice leaf disease identification method based on deep learning |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107169421A (en) * | 2017-04-20 | 2017-09-15 | 华南理工大学 | A kind of car steering scene objects detection method based on depth 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 |
| CN107576618A (en) * | 2017-07-20 | 2018-01-12 | 华南理工大学 | Pyricularia Oryzae detection method and system based on depth convolutional neural networks |
| CN107871124A (en) * | 2017-11-15 | 2018-04-03 | 陕西师范大学 | A kind of Remote Sensing Target detection method based on deep neural network |
| 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 |
| CN108491854A (en) * | 2018-02-05 | 2018-09-04 | 西安电子科技大学 | Remote sensing image object detection method based on SF-RCNN |
| CN109117876A (en) * | 2018-07-26 | 2019-01-01 | 成都快眼科技有限公司 | A kind of dense small target deteection model building method, model and detection method |
| CN109242844A (en) * | 2018-09-04 | 2019-01-18 | 青岛大学附属医院 | Pancreatic tumour automatic recognition system based on deep learning, computer equipment, storage medium |
| CN109255352A (en) * | 2018-09-07 | 2019-01-22 | 北京旷视科技有限公司 | Object detection method, apparatus and system |
-
2019
- 2019-01-30 CN CN201910089817.3A patent/CN109886155B/en active Active
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107169421A (en) * | 2017-04-20 | 2017-09-15 | 华南理工大学 | A kind of car steering scene objects detection method based on depth 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 |
| CN107576618A (en) * | 2017-07-20 | 2018-01-12 | 华南理工大学 | Pyricularia Oryzae detection method and system based on depth convolutional neural networks |
| CN107871124A (en) * | 2017-11-15 | 2018-04-03 | 陕西师范大学 | A kind of Remote Sensing Target detection method based on deep neural network |
| 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 |
| CN108491854A (en) * | 2018-02-05 | 2018-09-04 | 西安电子科技大学 | Remote sensing image object detection method based on SF-RCNN |
| CN109117876A (en) * | 2018-07-26 | 2019-01-01 | 成都快眼科技有限公司 | A kind of dense small target deteection model building method, model and detection method |
| CN109242844A (en) * | 2018-09-04 | 2019-01-18 | 青岛大学附属医院 | Pancreatic tumour automatic recognition system based on deep learning, computer equipment, storage medium |
| CN109255352A (en) * | 2018-09-07 | 2019-01-22 | 北京旷视科技有限公司 | Object detection method, apparatus and system |
Non-Patent Citations (2)
| Title |
|---|
| 伍伟明: "基于Faster_R_CNN的目标检测算法的研究", 《中国优秀硕士学位论文全文数据库信息科技辑》 * |
| 黄双萍 等: "基于深度卷积神经网络的水稻穗瘟病检测方法", 《农业工程学报》 * |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110853080A (en) * | 2019-09-30 | 2020-02-28 | 广西慧云信息技术有限公司 | Method for measuring size of field fruit |
| CN111062871A (en) * | 2019-12-17 | 2020-04-24 | 腾讯科技（深圳）有限公司 | Image processing method and device, computer equipment and readable storage medium |
| CN111062871B (en) * | 2019-12-17 | 2023-01-24 | 腾讯科技（深圳）有限公司 | Image processing method and device, computer equipment and readable storage medium |
| CN111462058A (en) * | 2020-03-24 | 2020-07-28 | 华南农业大学 | Method for quickly detecting effective ears of rice |
| CN111462058B (en) * | 2020-03-24 | 2024-01-05 | 华南农业大学 | Method for rapidly detecting effective rice ears |
| CN111723736A (en) * | 2020-06-19 | 2020-09-29 | 中国农业科学院农业信息研究所 | Fruit tree flowering phase monitoring method and device, computer equipment and storage medium |
| CN113128576A (en) * | 2021-04-02 | 2021-07-16 | 中国农业大学 | Crop row detection method and device based on deep learning image segmentation |
| CN113129281A (en) * | 2021-04-13 | 2021-07-16 | 广西大学 | Wheat stem section parameter detection method based on deep learning |
| CN113129281B (en) * | 2021-04-13 | 2022-06-21 | 广西大学 | Wheat stem section parameter detection method based on deep learning |
| CN115619286A (en) * | 2022-11-11 | 2023-01-17 | 中国农业科学院农业资源与农业区划研究所 | Method and system for evaluating sample plot quality of breeding field plot |
| CN115619286B (en) * | 2022-11-11 | 2023-10-03 | 中国农业科学院农业资源与农业区划研究所 | Method and system for evaluating quality of sample plot of breeding field district |
| CN116863340A (en) * | 2023-08-16 | 2023-10-10 | 安徽荃银超大种业有限公司 | Rice leaf disease identification method based on deep learning |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109886155B (en) | 2021-08-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109886155A (en) | Man power single stem rice detection localization method, system, equipment and medium based on deep learning | |
| Chen et al. | An AIoT based smart agricultural system for pests detection | |
| Yu et al. | Detection of broadleaf weeds growing in turfgrass with convolutional neural networks | |
| Åstrand et al. | An agricultural mobile robot with vision-based perception for mechanical weed control | |
| CN102564593B (en) | Plant growth condition monitoring system based on compute vision and internet of things | |
| Wu et al. | Detection and counting of banana bunches by integrating deep learning and classic image-processing algorithms | |
| CN107688856B (en) | Indoor robot scene active identification method based on deep reinforcement learning | |
| CN109635875A (en) | A kind of end-to-end network interface detection method based on deep learning | |
| CN107844797A (en) | A kind of method of the milking sow posture automatic identification based on depth image | |
| Tian et al. | Application status and challenges of machine vision in plant factory—A review | |
| Huang et al. | Deep localization model for intra-row crop detection in paddy field | |
| CN103530643A (en) | Pesticide positioned spraying method and system on basis of crop interline automatic identification technology | |
| CN102789579A (en) | Identification method for stressed state of water fertilizer of greenhouse crop on basis of computer vision technology | |
| CN114239756B (en) | Insect pest detection method and system | |
| Miao et al. | Crop weed identification system based on convolutional neural network | |
| Wang et al. | The seedling line extraction of automatic weeding machinery in paddy field | |
| Lu et al. | Citrus green fruit detection via improved feature network extraction | |
| Gao et al. | Maize seedling information extraction from UAV images based on semi-automatic sample generation and Mask R-CNN model | |
| CN110555343B (en) | Method and system for extracting three elements of forest, shrub and grass in typical resource elements | |
| CN116739739A (en) | Loan amount evaluation method and device, electronic equipment and storage medium | |
| Zhang et al. | Application of convolution neural network algorithm based on intelligent sensor network in target recognition of corn weeder at seedling stage | |
| CN113222025B (en) | Feasible region label generation method based on laser radar | |
| Fang et al. | Classification system study of soybean leaf disease based on deep learning | |
| Xu et al. | Cucumber flower detection based on YOLOv5s-SE7 within greenhouse environments | |
| Reddy et al. | Weed detection using alexnet architecture in the farming fields |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |