CN113435309B - Rice seedling row identification method based on row vector grid classification - Google Patents

Rice seedling row identification method based on row vector grid classification Download PDF

Info

Publication number
CN113435309B
CN113435309B CN202110704097.4A CN202110704097A CN113435309B CN 113435309 B CN113435309 B CN 113435309B CN 202110704097 A CN202110704097 A CN 202110704097A CN 113435309 B CN113435309 B CN 113435309B
Authority
CN
China
Prior art keywords
row
layer
seedling
output
vector
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.)
Active
Application number
CN202110704097.4A
Other languages
Chinese (zh)
Other versions
CN113435309A (en
Inventor
王姗姗
余山山
张文毅
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Nanjing Research Institute for Agricultural Mechanization Ministry of Agriculture
Original Assignee
Nanjing Research Institute for Agricultural Mechanization Ministry of Agriculture
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Nanjing Research Institute for Agricultural Mechanization Ministry of Agriculture filed Critical Nanjing Research Institute for Agricultural Mechanization Ministry of Agriculture
Priority to CN202110704097.4A priority Critical patent/CN113435309B/en
Publication of CN113435309A publication Critical patent/CN113435309A/en
Application granted granted Critical
Publication of CN113435309B publication Critical patent/CN113435309B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
    • 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)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Software Systems (AREA)
  • Mathematical Physics (AREA)
  • Computing Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Probability & Statistics with Applications (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Image Analysis (AREA)

Abstract

The invention relates to the field of rice seedling identification, in particular to a rice seedling row identification method based on row vector grid classification, which comprises the following steps: firstly, extracting features of an input image; classifying based on the row vector grids; and thirdly, outputting the horizontal position of the seedling row. The invention provides an end-to-end CNN network model based on row vector grid classification, which can directly identify rice seedling rows under different illumination intensities and weeds, fuzziness, duckweed and other scenes, and solves the influence of factors such as illumination change, duckweed weed noise, fuzziness caused by machine vibration and the like on the rice seedling row identification. The invention converts seedling row identification into a selection problem based on row vectors by using global characteristics, and can remarkably reduce the calculation cost through the selection based on the row vectors.

Description

Rice seedling row identification method based on row vector grid classification
Technical Field
The invention relates to the field of rice seedling identification, in particular to a rice seedling row identification method based on row vector grid classification.
Background
The rice seedling row identification is one of key links of rice mechanical weeding, and has important guiding effects on improving row weeding precision of weeding machinery, improving weed removal rate and reducing seedling injury rate.
The rice seedling row identification is easily affected by factors such as duckweed weed noise, fuzzy factors caused by illumination change of farmland environment and machine vibration, and the like, so that a better rice seedling row identification method needs to be researched urgently.
Disclosure of Invention
In order to solve the above mentioned disadvantages in the background art, the present invention provides a method for identifying rice seedling rows based on row vector grid classification.
The purpose of the invention can be realized by the following technical scheme:
a rice seedling row identification method based on row vector grid classification comprises the following steps:
firstly, extracting features of an input image;
classifying based on the row vector grids;
and thirdly, outputting the horizontal position of the seedling row.
Further, the specific steps of the first step are as follows:
a1, inputting a rice seedling image H W;
a2, extracting features of the original input rice seedling image by using a convolution kernel of 64 × 7 × 7, wherein the step size stride is 2, and obtaining an output feature map of the 0 th layer
Figure BDA0003131490670000021
A3, performing feature extraction on the 0 th layer output feature diagram obtained by A2 by using two basic blocks 1 to obtain the 1 st layer output feature diagram
Figure BDA0003131490670000022
A4, performing feature extraction on the layer 1 output feature diagram obtained by A3 by using two basic blocks 2 to obtain the layer 2 output feature diagram
Figure BDA0003131490670000023
A5, performing feature extraction on the layer 2 output feature diagram obtained by A4 by using two basic blocks 3 to obtain the layer 3 output feature diagram
Figure BDA0003131490670000024
A6, performing feature extraction on the layer 3 output feature diagram obtained by A5 by using two basic blocks 4 to obtain the layer 4 output feature diagram
Figure BDA0003131490670000025
A7, performing feature extraction on the 4 th layer output feature map obtained by A6 by using convolution and full connection to obtain the 5 th layer output vector 1O5。
Further, the specific steps of the second step are as follows:
b1, output vector 1 x O to layer 55Performing full connection operation to obtain output vector 1O of layer 66In which O is6Is the product of M x (N +1) x C, where M is the number of predefined row vectors, N +1 represents the number of grid divisions N plus the absence of seedling rows, and C is the total number of seedling rows;
and B2, performing convolution operation on the output vector of the layer 6 to obtain an output characteristic diagram M x (N +1) x C of the layer 7.
Further, the third step comprises the following specific steps:
for the output characteristic map M x (N +1) x C of the 7 th layer, prediction P of seedling rowi,j＝Fij(X),i∈[1,C],j∈[1,M]Selecting the grid unit with the highest probability from the 1 st row vector to the Mth row vector for the ith seedling row, wherein the grid units at all the row vectors form the seedling row;
wherein the value of i is from 1 to C, the value of j is from 1 to M, and the prediction P of seedling rowi,jIs a (N +1) -dimensional vector representing the probability that the ith seedling row selects (N +1) grid cells at the jth row vector.
The invention has the beneficial effects that:
the invention provides an end-to-end CNN network model based on row vector grid classification, which can directly identify rice seedling rows under different illumination intensities and weeds, fuzziness, duckweed and other scenes, and solves the influence of factors such as illumination change, duckweed weed noise, fuzziness caused by machine vibration and the like on rice seedling row identification.
The invention converts seedling row identification into a selection problem based on row vectors by using global characteristics, and can remarkably reduce the calculation cost through the selection based on the row vectors.
Drawings
In order to more clearly illustrate the embodiments or technical solutions in the prior art of the present invention, the drawings used in the description of the embodiments or prior art will be briefly described below, and it is obvious for those skilled in the art to obtain other drawings without creative efforts;
FIG. 1 is a block diagram of the process of the rice seedling row identification method based on row vector grid classification according to the present invention;
FIG. 2 is a schematic flow chart of the rice seedling row identification method based on row vector grid classification according to the present invention;
FIG. 3 is a schematic diagram of horizontal position selection at the row vector of the present invention;
FIG. 4 is a diagram of test results of different scenarios in an embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
A method for identifying rice seedling rows based on row vector grid classification, as shown in fig. 1 and 2, comprises the following steps:
firstly, extracting features of an input image;
classifying based on the row vector grids;
and thirdly, outputting the horizontal position of the seedling row.
The method for extracting the features of the input image comprises the following steps:
a1, inputting a rice seedling image H W;
a2, extracting features of the original input rice seedling image by using a convolution kernel of 64 × 7 × 7, wherein the step size stride is 2, and obtaining an output feature map of the 0 th layer
Figure BDA0003131490670000041
A3, performing feature extraction on the 0 th layer output feature map obtained by A2 by using two basic block1, wherein the structure of basic block1 is shown as Res1 in Table 1, the basic block1 comprises two convolution kernels of 3 x 64, and the obtained 1 st layer output feature map is shown as
Figure BDA0003131490670000042
A4, performing feature extraction on the layer 1 output feature map obtained by A3 by using two basic block2, wherein the structure of basic block2 is shown as Res2 in Table 1, the basic block2 comprises two convolution kernels of 3 x 128, and the obtained layer 2 output feature map is shown as
Figure BDA0003131490670000043
A5, performing feature extraction on the layer 2 output feature map obtained by A4 by using two basic block3, wherein the structure of basic block3 is shown as Res3 in Table 1, the basic block3 comprises two convolution kernels of 3 x 256, and the obtained layer 3 output feature map is shown as
Figure BDA0003131490670000044
A6, performing feature extraction on the output feature map of the layer 3 obtained by A5 by using two basic block4, wherein the structure of basic block4 is shown as Res4 in Table 1, the basic block4 comprises two convolution kernels of 3 x 512, and the output feature map of the layer 4 is obtained
Figure BDA0003131490670000045
A7, performing feature extraction on the 4 th layer output feature map obtained by A6 by using convolution and full connection to obtain the 5 th layer output vector of 1 xO5。
Table 1 network architecture for feature extraction
Figure BDA0003131490670000046
Figure BDA0003131490670000051
Wherein the classification based on the row vector grid comprises the following steps:
b1, output vector 1 x O to layer 55Performing full connection operation to obtain output vector 1O of layer 66In which O is6Is the product of M x (N +1) x C, where M is the number of predefined row vectors, N +1 represents the number of grid divisions N plus the absence of seedling rows, and C is the total number of seedling rows;
and B2, performing convolution operation on the output vector of the layer 6 to obtain an output characteristic diagram M x (N +1) x C of the layer 7.
Wherein, output seedling row horizontal position includes the following steps:
for the output characteristic map M x (N +1) x C of the 7 th layer, prediction P of seedling rowi,j＝Fij(X),i∈[1,C],j∈[1,M]Wherein i takes on a value from 1 to C, and j takes on a value from 1 to M; prediction of seedling rows Pi,jIs a (N +1) -dimensional vector representing the probability that the ith seedling row selects (N +1) grid cells at the jth row vector.
For the ith seedling row, the grid cells with the highest probability at the row vectors from 1 st to Mth are selected, and the grid cells at all the row vectors form the seedling row, as shown in FIG. 3.
Example (b):
and respectively carrying out image acquisition on 15 days and 30 days of rice transplanting, wherein the height of the rice seedling after 15 days of rice transplanting is about 20mm, and the height of the rice seedling after 30 days of rice transplanting is about 30 mm. The row spacing of the seedlings is 30mm, and the number of the seedlings per mu is about 70000.
TABLE 2 comparison of experimental results for different scenarios in the test set
Figure BDA0003131490670000052
In table 2, Scenario (scene) has three scenes of duckweed, much weeds and vibration blur, Accuracy is precision, Recall is Recall, F1 score is F1 value, and Runtime is average running time.
As can be seen from table 2, the precision, recall and F1 values in the duckweed scenario were the lowest in the three test scenarios, 84.49%, 86.97% and 85.71%, respectively, and were all lower than the average prediction precision, recall and F1 values of the entire dataset, whereas the prediction precision in the two scenarios, weed-rich and vibration-blurred, were very close, 88.55% and 88.59%, respectively, and were all higher than the average prediction precision 88.52% of the entire dataset. The average running time of scenes with duckweeds, much weeds and vibration blurring is 6.01ms, 5.98ms and 5.92ms respectively, the running time difference of the three scenes is very small, and the fps is close to 170 fps.
Fig. 4 shows the test effect of three typical scenarios, wherein (a), (b), and (c) in fig. 4 are test effect graphs in the scenarios of duckweed, weed abundance, and vibration blur, respectively, and the prediction results at the predefined row vectors are represented by dot sets.
As shown by the oval circles in the graph (a), the seedling line identification result in the duckweed scene has a smaller deviation from the actual seedling line, and the seedling line identification effect in the weed multi-scene and the vibration fuzzy scene is relatively better, compared with the case of more weeds (flaky weeds).
As shown in the figure (b), there is substantially no case where weeds are erroneously recognized as seedlings, and the method of the present invention provides a better distinction between seedlings and weeds.
For example, in the vibration fuzzy scene in the graph (c), the seedlings in the image are unclear at the moment, and only the general outline can be seen, so that the difficulty in accurately identifying the seedling rows in the fuzzy scene is high.
In the above, the rice seedling row identification method based on row vector grid classification of the invention provides an end-to-end CNN network model, which can directly identify rice seedling rows under different illumination intensities and weeds, fuzziness, duckweed and other scenes, and solves the influence of factors such as illumination change, duckweed weed noise, fuzziness caused by machine vibration and the like on rice seedling row identification.
The invention converts seedling row identification into a selection problem based on row vectors by using global characteristics, and can remarkably reduce the calculation cost through the selection based on the row vectors.
The foregoing shows and describes the general principles, essential features, and advantages of the invention. It will be understood by those skilled in the art that the present invention is not limited to the embodiments described above, which are described in the specification and illustrated only to illustrate the principle of the present invention, but that various changes and modifications may be made therein without departing from the spirit and scope of the present invention, which fall within the scope of the invention as claimed.

Claims (2)

1. A rice seedling row identification method based on row vector grid classification is characterized by comprising the following steps:
firstly, extracting features of an input image;
the specific steps of the first step are as follows:
a1, inputting a rice seedling image H W;
a2, extracting features of the original input rice seedling image by using a convolution kernel of 64 × 7 × 7, wherein the step size stride is 2, and obtaining an output feature map of the 0 th layer
Figure FDA0003504550290000011
A3, performing feature extraction on the 0 th layer output feature diagram obtained by A2 by using two basic blocks 1 to obtain the 1 st layer output feature diagram
Figure FDA0003504550290000012
A4, performing feature extraction on the layer 1 output feature diagram obtained by A3 by using two basic blocks 2 to obtain the layer 2 output feature diagram
Figure FDA0003504550290000013
A5, performing feature extraction on the layer 2 output feature diagram obtained by A4 by using two basic blocks 3 to obtain the layer 3 output feature diagram
Figure FDA0003504550290000014
A6, performing feature extraction on the layer 3 output feature diagram obtained by A5 by using two basic blocks 4 to obtain the layer 4 output feature diagram
Figure FDA0003504550290000015
A7, performing feature extraction on the 4 th layer output feature map obtained by A6 by using convolution and full connection to obtain the 5 th layer output vector 1O5；
Classifying based on the row vector grids;
the second step comprises the following specific steps:
b1, output vector 1 x O to layer 55Performing full connection operation to obtain output vector 1O of layer 66In which O is6Is the product of M x (N +1) x C, where M is the number of predefined row vectors, N +1 represents the number of grid divisions N plus the absence of seedling rows, and C is the total number of seedling rows;
b2, performing convolution operation on the output vector of the layer 6 to obtain an output characteristic diagram M x (N +1) x C of the layer 7;
and thirdly, outputting the horizontal position of the seedling row.
2. The method for identifying the rice seedling rows based on the row vector grid classification as claimed in claim 1, wherein the third step comprises the following specific steps:
for the output characteristic map M x (N +1) x C of the 7 th layer, prediction P of seedling rowi,j＝Fij(X),i∈[1，C]，j∈[1，M]Selecting the grid unit with the highest probability from the 1 st row vector to the Mth row vector for the ith seedling row, wherein the grid units at all the row vectors form the seedling row;
wherein i is from 1 to C, j is from 1 to M,prediction of seedling rows Pi,jIs a (N +1) -dimensional vector representing the probability that the ith seedling row selects (N +1) grid cells at the jth row vector.
CN202110704097.4A 2021-06-24 2021-06-24 Rice seedling row identification method based on row vector grid classification Active CN113435309B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110704097.4A CN113435309B (en) 2021-06-24 2021-06-24 Rice seedling row identification method based on row vector grid classification

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110704097.4A CN113435309B (en) 2021-06-24 2021-06-24 Rice seedling row identification method based on row vector grid classification

Publications (2)

Publication Number Publication Date
CN113435309A CN113435309A (en) 2021-09-24
CN113435309B true CN113435309B (en) 2022-04-26

Family

ID=77753930

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110704097.4A Active CN113435309B (en) 2021-06-24 2021-06-24 Rice seedling row identification method based on row vector grid classification

Country Status (1)

Country Link
CN (1) CN113435309B (en)

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107667802A (en) * 2017-09-27 2018-02-09 舒城县文禾种植专业合作社 A kind of method for culturing seedlings of the high single cropping rice blanket seedling machine transplanting of rice of per mu yield
CN109344843A (en) * 2018-09-07 2019-02-15 华南农业大学 Rice seedling line extracting method, device, computer equipment and storage medium
CN112036437A (en) * 2020-07-28 2020-12-04 农业农村部南京农业机械化研究所 Rice seedling detection model based on improved YOLOV3 network and method thereof

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US11164071B2 (en) * 2017-04-18 2021-11-02 Samsung Electronics Co., Ltd. Method and apparatus for reducing computational complexity of convolutional neural networks

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107667802A (en) * 2017-09-27 2018-02-09 舒城县文禾种植专业合作社 A kind of method for culturing seedlings of the high single cropping rice blanket seedling machine transplanting of rice of per mu yield
CN109344843A (en) * 2018-09-07 2019-02-15 华南农业大学 Rice seedling line extracting method, device, computer equipment and storage medium
CN112036437A (en) * 2020-07-28 2020-12-04 农业农村部南京农业机械化研究所 Rice seedling detection model based on improved YOLOV3 network and method thereof

Also Published As

Publication number Publication date
CN113435309A (en) 2021-09-24

Similar Documents

Publication Publication Date Title
CN110175168B (en) Time sequence data filling method and system based on generation of countermeasure network
CN110909801B (en) Data classification method, system, medium and device based on convolutional neural network
CN109977895B (en) Wild animal video target detection method based on multi-feature map fusion
CN112257738A (en) Training method and device of machine learning model and classification method and device of image
CN106951499A (en) A kind of knowledge mapping method for expressing based on translation model
CN111105160A (en) Steel quality prediction method based on tendency heterogeneous bagging algorithm
CN110674865B (en) Rule learning classifier integration method oriented to software defect class distribution unbalance
CN108846338A (en) Polarization characteristic selection and classification method based on object-oriented random forest
CN108205580A (en) A kind of image search method, device and computer readable storage medium
CN107133626B (en) Medical image classification method based on partial average stochastic optimization model
CN112294341A (en) Sleep electroencephalogram spindle wave identification method and system based on light convolutional neural network
CN110188196B (en) Random forest based text increment dimension reduction method
JP2017041206A (en) Learning device, search device, method, and program
CN108875532A (en) A kind of video actions detection method based on sparse coding and length posterior probability
CN115021679A (en) Photovoltaic equipment fault detection method based on multi-dimensional outlier detection
Jobin et al. Plant identification based on fractal refinement technique (FRT)
CN111046730A (en) Plant data processing method and device, computer equipment and storage medium
CN114596274A (en) Natural background citrus greening disease detection method based on improved Cascade RCNN network
CN113435309B (en) Rice seedling row identification method based on row vector grid classification
CN112115984A (en) Tea garden abnormal data correction method and system based on deep learning and storage medium
CN115546862A (en) Expression recognition method and system based on cross-scale local difference depth subspace characteristics
CN114254705A (en) Abnormal data detection method and device, storage medium and computer equipment
Kishore et al. Paddy Leaf Disease Detection using Deep Learning Methods
CN113591647A (en) Human body action recognition method and device, computer equipment and storage medium
CN118115822B (en) Abnormality detection method and system based on image block feature cascade retrieval model

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