CN117457101B - Method, medium and system for predicting moisture content of cured tobacco leaves - Google Patents
Method, medium and system for predicting moisture content of cured tobacco leaves Download PDFInfo
- Publication number
- CN117457101B CN117457101B CN202311773785.1A CN202311773785A CN117457101B CN 117457101 B CN117457101 B CN 117457101B CN 202311773785 A CN202311773785 A CN 202311773785A CN 117457101 B CN117457101 B CN 117457101B
- Authority
- CN
- China
- Prior art keywords
- branch
- module
- prediction
- moisture content
- network
- 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
Links
- 241000208125 Nicotiana Species 0.000 title claims abstract description 104
- 235000002637 Nicotiana tabacum Nutrition 0.000 title claims abstract description 104
- 238000000034 method Methods 0.000 title claims abstract description 52
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 claims abstract description 33
- 230000008569 process Effects 0.000 claims abstract description 15
- 230000010354 integration Effects 0.000 claims description 19
- 239000013598 vector Substances 0.000 claims description 6
- 238000000605 extraction Methods 0.000 claims description 5
- 238000010438 heat treatment Methods 0.000 claims description 3
- 230000007480 spreading Effects 0.000 claims description 3
- 230000000694 effects Effects 0.000 description 9
- 230000008859 change Effects 0.000 description 7
- 238000004422 calculation algorithm Methods 0.000 description 6
- 238000013528 artificial neural network Methods 0.000 description 5
- 230000009286 beneficial effect Effects 0.000 description 2
- 239000000284 extract Substances 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 206010063385 Intellectualisation Diseases 0.000 description 1
- 230000006978 adaptation Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000009792 diffusion process Methods 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 230000003313 weakening effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16C—COMPUTATIONAL CHEMISTRY; CHEMOINFORMATICS; COMPUTATIONAL MATERIALS SCIENCE
- G16C20/00—Chemoinformatics, i.e. ICT specially adapted for the handling of physicochemical or structural data of chemical particles, elements, compounds or mixtures
- G16C20/20—Identification of molecular entities, parts thereof or of chemical compositions
-
- 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
-
- 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
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0004—Industrial image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10024—Color image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30108—Industrial image inspection
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Computing Systems (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- General Engineering & Computer Science (AREA)
- Molecular Biology (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Spectroscopy & Molecular Physics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Chemical & Material Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Quality & Reliability (AREA)
- Crystallography & Structural Chemistry (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention provides a method, a medium and a system for predicting the moisture content of cured tobacco, belonging to the technical field of prediction of the moisture content of cured tobacco, comprising the following steps: acquiring a color tobacco leaf image acquired by a camera in the baking process; inputting the tobacco leaf image into a scattering-like network integrated structure formed by a complete sharing module, a partial sharing module and an independent module; extracting initial features of the tobacco leaf image by using a complete sharing module; extracting the characteristics of a plurality of branch structures by using a part sharing module; extracting different features for each branch structure independently using an independent module; each branch structure outputs a predicted tobacco leaf moisture content; and calculating the predicted water content output by each branch structure and outputting the final water content of the tobacco leaves. The problems that in the prior art, the prediction of the moisture content of the cured tobacco leaves is low in efficiency and poor in generalization capability, and the cured tobacco leaves in different curing states are difficult to adapt to; the prediction result is unstable, and the technical problem of large errors exists in different characteristics or model judgment.
Description
Technical Field
The invention belongs to the technical field of prediction of moisture content of cured tobacco leaves, and particularly relates to a method, medium and system for predicting moisture content of cured tobacco leaves.
Background
At present, the main reasons for influencing unmanned intelligent baking of tobacco leaves are the problems of inconsistent color change and water content of the tobacco leaves in the baking process caused by factors such as intrinsic quality differences of the maturity of the tobacco leaves and the like, so that the temperature and humidity control is influenced, and the baking quality of the tobacco leaves is influenced. The color change can be accurately judged by directly observing the appearance change of the tobacco leaves through human eyes, the water content is taken as the change of the internal components, the water content is difficult to visually observe through the appearance, and the water content has stronger subjectivity when judged by people, so that different people have different judging standards. The traditional method relies on manual experience to bake, and individual differences exist in judging the color and the water content of tobacco leaves, so that automatic baking cannot be realized. In the prior art, the water content prediction method mainly comprises three methods: firstly, subjective observation by people is relied on; secondly, manually extracting characteristics to establish a linear prediction model; thirdly, a nonlinear relation model of the tobacco leaf image and the water content is established by using a single neural network. The main problems of the method are as follows: the efficiency is low, the generalization capability is poor, and the tobacco leaves in different baking states are difficult to adapt; the predicted result is unstable.
Disclosure of Invention
In view of the above, the invention provides a method, medium and system for predicting the moisture content of cured tobacco, which can solve the problems of low existing efficiency, poor generalization capability and difficult adaptation to tobacco in different curing states in the prior art for predicting the moisture content of cured tobacco; and the predicted result is unstable.
The invention is realized in the following way:
the first aspect of the invention provides a method for predicting the moisture content of cured tobacco leaves, which comprises the following steps:
s10, acquiring a color tobacco leaf image acquired by a camera in the baking process;
s20, inputting the tobacco leaf image into a scattering-like network integrated structure formed by a complete sharing module, a partial sharing module and an independent module;
s30, extracting initial characteristics of the tobacco leaf images by using a complete sharing module;
s40, extracting the initial features by using a part sharing module to obtain features of a plurality of branch structures;
s50, independently extracting different features for each branch structure by using an independent module;
s60, outputting a predicted tobacco leaf water content by each branch structure;
s70, calculating the predicted water content output by each branch structure by adopting an arithmetic average strategy;
s80, outputting the calculated final tobacco leaf water content prediction result.
On the basis of the technical scheme, the method for predicting the moisture content of the cured tobacco leaves can be further improved as follows:
the step of acquiring the color tobacco leaf image acquired by the camera in the baking process specifically comprises the following steps of: setting baking equipment to obtain tobacco leaves to be baked; setting a color image acquisition device to enable a shooting range of the color image acquisition device to cover a baking device; heating the baking equipment and sending the tobacco leaves into the baking equipment for baking; controlling an image acquisition device to shoot tobacco leaves in a baking process regularly or continuously to obtain a plurality of images; and calibrating and correcting the image.
The step of inputting the tobacco leaf image into a scattering network integrated structure formed by a complete sharing module, a partial sharing module and an independent module specifically comprises the following steps: the scattering-like network integrated structure is constructed and comprises a complete sharing convolution module, a partial sharing convolution module and an independent convolution module; determining the number of branch networks in the independent module, wherein each branch network corresponds to a subsequent prediction output; initializing network parameters; each frame of tobacco leaf image is input to the input end of the network integrated structure.
The step of extracting the initial characteristics of the tobacco leaf image by using the complete sharing module specifically comprises the following steps: defining a network structure of a complete sharing module, wherein the network structure comprises an input layer, a plurality of convolution layers or a complete connection layer; determining operation parameters of each layer in the complete sharing module, including convolution kernel size, step length and the like; forward spreading the image input in the step S20 through a complete sharing module; and storing the feature vector of the output end of the complete sharing module as an initial feature.
The step of extracting the initial features by using a partial sharing module to obtain features of a plurality of branch structures specifically includes: confirming the number n of required branch structures, and defining the network structure of part of the sharing modules; copying the initial characteristics output by the complete sharing module into n parts, and inputting one branch into each part; setting operation parameters of a convolution layer in each branch; forward operation is carried out on the characteristics of each branch in each network; and saving the characteristics of each branch output end as the characteristics of each branch structure.
The step of extracting different features for each branch structure by using an independent module specifically comprises the following steps: defining a network structure of independent modules in each branch structure; inputting each branch characteristic output in the step S40 into a corresponding independent module; setting network parameters of independent modules in each branch; performing forward extraction operation of features in each branch independent module; and storing the characteristic vectors of the output ends of the independent branch modules.
The step of outputting a predicted tobacco leaf moisture content by each branch structure specifically comprises the following steps: setting a prediction layer of each branch structure, wherein the prediction layer comprises an output node; connecting the characteristics output by each branch independent module to a corresponding prediction layer; forward operation is carried out in the prediction layer of each branch to obtain the prediction output of each branch; and storing the prediction results output by each branch structure.
The step of calculating the predicted water content output by each branch structure by adopting an arithmetic average strategy specifically comprises the following steps: reading each branch prediction output saved in the step S60; calculating an arithmetic mean of the branch prediction outputs; the arithmetic mean is output as the final prediction.
A second aspect of the present invention provides a computer readable storage medium having stored therein program instructions that, when executed, are configured to perform a cured tobacco moisture content prediction method as described above.
A third aspect of the present invention provides a cured tobacco moisture content prediction system comprising the computer readable storage medium described above.
Compared with the prior art, the method, the medium and the system for predicting the moisture content of the cured tobacco leaf provided by the invention have the beneficial effects that:
1. the network integrated structure is adopted to describe the change of the water content of the tobacco leaves in the baking process, so that the influence of the difference between the tobacco leaves on the prediction result is reduced, and the accuracy of the water content prediction is enhanced;
2. the integration strategy gathers a plurality of branch structures to obtain results, and the final water content prediction is carried out by considering most judgment, so that the deviation caused by single network prediction errors is reduced, and the stability of water content prediction is greatly enhanced;
3. the method has the advantages that manual judgment is replaced by an automatic acquisition and analysis mode, the problem of low efficiency of the traditional method is solved, and the intellectualization of a water content prediction task is realized;
4. the multiple branch networks respectively learn different characteristics, extract multi-aspect information of the complex change process of tobacco leaves, strengthen the generalization capability of the model and can adapt to different baking states.
Therefore, the technical scheme of the invention can solve the problems that in the prior art, the prediction of the moisture content of the cured tobacco leaves has low efficiency and poor generalization capability, and is difficult to adapt to the tobacco leaves in different curing states; and the predicted result is unstable.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings that are needed in the description of the embodiments of the present invention will be briefly described below, it being obvious that the drawings in the following description are only some embodiments of the present invention, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a flow chart of a method provided by the present invention;
FIG. 2 is a flow chart for intelligent prediction of the moisture content of cured tobacco leaves;
FIG. 3 is a network architecture diagram of a diffusion network integration algorithm;
fig. 4 is a flowchart of a training algorithm.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention more apparent, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.
As shown in fig. 1, a flowchart of a method for predicting the moisture content of cured tobacco leaves according to a first aspect of the present invention is provided, the method comprising the steps of:
s10, acquiring a color tobacco leaf image acquired by a camera in the baking process, wherein the method specifically comprises the following steps of: setting baking equipment to obtain tobacco leaves to be baked; setting a color image acquisition device to enable a shooting range of the color image acquisition device to cover a baking device; heating the baking equipment and sending the tobacco leaves into the baking equipment for baking; controlling an image acquisition device to shoot tobacco leaves in a baking process at regular time or continuously to obtain a series of images; and calibrating and correcting the image.
The effect of this step is: and acquiring tobacco leaf images acquired by a camera in the whole baking process for subsequent processing. This is the key input raw data of the present invention.
S20, inputting the tobacco leaf image into a scattering-like network integrated structure formed by a complete sharing module, a partial sharing module and an independent module, wherein the method specifically comprises the following steps of: constructing the scattering-like network integration structure, comprising: the system comprises a complete sharing convolution module, a partial sharing convolution module and an independent convolution module; determining the number of branch networks in the independent module, wherein each branch network corresponds to a subsequent prediction output; initializing network parameters; each frame of tobacco leaf image is input to the input end of the network integrated structure.
The effect of this step is: and constructing a scattering network integrated prediction model, inputting the picture into the model, and preparing data for subsequent feature extraction and prediction.
S30, extracting initial characteristics of the tobacco leaf image by using a complete sharing module, wherein the method specifically comprises the following steps: defining a network structure of a complete sharing module, wherein the network structure comprises an input layer, a plurality of convolution layers or a complete connection layer; determining operation parameters of each layer in the complete sharing module, including convolution kernel size, step length and the like; forward spreading the image input in the step S20 through a complete sharing module; and storing the feature vector of the output end of the complete sharing module as an initial feature.
The effect of this step is: and uniformly extracting initial characteristics of all images by using a parameter sharing module, and providing the same initial characteristics for each branch structure so as to reduce the number of parameters.
S40, extracting the initial characteristics by using a part sharing module to obtain characteristics of a plurality of branch structures, wherein the method specifically comprises the following steps: confirming the number n of required branch structures, and defining the network structure of part of the sharing modules; copying the initial characteristics output by the complete sharing module into n parts, and inputting one branch into each part; setting operation parameters of a convolution layer in each branch; forward operation is carried out on the characteristics of each branch in each network; and saving the characteristics of each branch output end as the characteristics of each branch structure.
The effect of this step is: and under the condition of keeping certain sharing, different branch networks learn the characteristics of partial independence so as to keep the generalization capability of the network integrated structure.
S50, independently extracting different features for each branch structure by using an independent module, wherein the method specifically comprises the following steps: defining a network structure of independent modules in each branch structure; inputting each branch characteristic output in the step S40 into a corresponding independent module; setting network parameters of independent modules in each branch; performing forward extraction operation of features in each branch independent module; and storing the characteristic vectors of the output ends of the independent branch modules.
The effect of this step is: the diversity among the branch structures is improved to the maximum extent, the characteristic representation of different angles is provided for the subsequent prediction, and the generalization capability of the network is further enhanced.
S60, outputting a predicted tobacco leaf water content by each branch structure, wherein the method specifically comprises the following steps: setting a prediction layer of each branch structure, wherein the prediction layer comprises an output node; connecting the characteristics output by each branch independent module to a corresponding prediction layer; forward operation is carried out in the prediction layer of each branch to obtain the prediction output of each branch; and storing the prediction results output by each branch structure.
The effect of this step is: and respectively obtaining a prediction output by utilizing the different characteristics extracted from different branch networks, and preparing for subsequent prediction result fusion.
S70, calculating the predicted water content output by each branch structure by adopting an arithmetic average strategy, wherein the method specifically comprises the following steps: reading each branch prediction output saved in the step S60; calculating an arithmetic mean of the branch prediction outputs; the arithmetic mean is output as the final prediction.
The effect of this step is: by integrating the prediction results of all branches, the judgment of different structures is synthesized, so that the accuracy and stability of prediction are improved.
S80, outputting a calculated final tobacco leaf water content prediction result, wherein the method specifically comprises the following steps of: the prediction average value obtained in the step S70 is used as a final prediction result to be output; and simultaneously, optionally outputting the prediction result of each branch structure to perform contrast analysis.
The effect of this step is: providing a final result of the tobacco leaf moisture content prediction. The user can judge the prediction credibility according to each branch result.
Specifically, the principle of the invention is as follows: the invention provides a network integration model for predicting the water content of tobacco leaves. The traditional single model is difficult to extract comprehensive characteristics when facing to the individual complex variation difference of tobacco leaves, and has poor generalization capability for different tobacco leaves. The scheme constructs a scattering structure, namely a tree topology structure consisting of a sharing layer, a part of sharing layer and an independent layer, wherein the sharing layer extracts public information, and the independent layer learns different characteristics. Thus, through the characteristic difference among different branches, the subtle change among tobacco leaves can be fully represented, and the generalization capability is enhanced. And simultaneously, the branches respectively output predicted values, and then the integrated strategy is adopted for comprehensive judgment. Compared with single judgment, the individual prediction deviation can be reduced, and the stability is improved. These two points are the problems and needs faced by current water cut predictions.
The following provides a specific embodiment of the scattering network integration structure according to the present invention, where the symbols involved in this embodiment are shown in the following table:
as shown in fig. 2, the intelligent prediction of the moisture content of cured tobacco leaves in this embodiment includes three processes: constructing a prediction frame of the moisture content of the cured tobacco leaves; calculating the number of branch structures of a prediction network for the moisture content of the cured tobacco leaves; and (5) the baked tobacco moisture content prediction network integration framework is trained. Firstly, designing a baked tobacco moisture content prediction frame, and establishing a nonlinear relation between a baked tobacco image and the tobacco moisture content; then calculating the optimal quantity of branch structures forming a network integration framework, and balancing the generalization capability and the operation time of the network structure, so that the network achieves the optimal effect in the aspects of prediction accuracy and the operation time; then, parameter training is carried out on a complete sharing module, a partial sharing module and an independent module of the network integration framework, characteristics describing the water content difference between tobacco leaves are obtained, and a water content prediction result of a branch structure is obtained; and finally integrating the branch result water content prediction result to obtain a final tobacco leaf water content prediction result.
1. Network frame algorithm for predicting moisture content of cured tobacco leaves
As shown in fig. 3, the flue-cured tobacco moisture content prediction integrated network is composed of a complete sharing module, a partial sharing module and an independent module, wherein the modules are internally provided with a plurality of unequal standard convolutions. The purpose of the shared (fully shared and partially shared) modules is to reduce parameters and connect inputs to the various branch structures including the fully shared and partially shared modules, each branch of the independent modules being parallel to each other and trained independently, which is beneficial to enhance diversity of the branch structures and maintain independence between the branch structures. The fully shared module is a common module shared by all the branch structures, directly connected to the input. As a junction hub it provides the same input for all branch structures. The partial sharing module is a module shared by partial branch structures, and aims to reduce parameters among branch structures and reduce correlation among the branch structures. The independent module is typically located at the end of the encoder, with many parallel branches that play an important role in maintaining independence between the branch structures. First, the tobacco leaf image is input to the complete sharing module, and initial features are extracted. The fully shared module contains only one branch, including several consecutive layers, called fully shared layers. The plurality of branch structures share an initial feature extraction module to reduce model complexity, computation time, and parameters. The initial characteristics are then input to the partial sharing module. The partially shared module consists of several parallel branches, which are connected to the branches of the fully connected module. Next, the features extracted by the partially shared modules are input to the individual modules. The independent module consists of a number of branches, the number of which is the same as the number of branch structures, usually at the end of the encoder. The main functions of the method are to extract different characteristics for each branch structure and ensure that each branch structure is independent of each other so as to maintain generalization capability, and each branch structure outputs a tobacco leaf moisture content prediction result. And finally, taking an arithmetic average value as an integration strategy, and carrying out integration treatment on the water content prediction results to obtain a final tobacco water content prediction result.
2. Calculation algorithm for branch structure number of prediction network for moisture content of cured tobacco leaves
The baked tobacco leaf water content prediction network integrated structure is composed of a plurality of branch structures, training samples of each branch structure are selected through a batch self-help sampling method, the training data size of the whole training set is assumed to be N, and the data size of a training subset of each branch structure is assumed to be M. Then there is the following relationship between N and M:and the relation parameter between the training data quantity of the whole training set and the data quantity of the training subset is represented as a positive integer. When a batch of samples is selected from the training dataset as training subset of the branch structure, the probability that each sample is not selected is +.>. When q branch structures are used, q batches of data need to be selected as training subsets of the branch structures, and then the probability that each sample is not selected is:
（1）
at this time, whenWhen the limits of the above formula can be expressed as:
（2）
weakening ofLimitation of (i.e. when->Sufficiently large, the limits calculated by the above formula can be expressed as:
（3）
in the process of training the branch structure, the data of each training set is contained in one or more training subsets as far as possible, namely, the probability that each sample in the training set is not selected after q times of selection is a small probability event. Is arbitrarily smallSo that the limit expression of the above formula is smaller than +.>The establishment is expressed as:
（4）
through inequality transformation, the number q of branch structures can be obtained as follows:
（5）
because the number of branch structures can only be an integer greater than 0, addIs a rounding operation. Based on the above, as the number of branch structures increasesThe generalization error of the integrated framework tends to be limited to an upper bound, so that the optimal number of tobacco leaf moisture content prediction branch structures is calculated.
3. Integrated framework training algorithm for baked tobacco leaf moisture content prediction network
To maintain inconsistencies between branch structures, all branch structures need to be trained in different sub-training data sets to obtain the ability to represent different features. The tobacco leaf moisture content prediction network integration framework has only one input connected with a plurality of branch structures, and if the network integration framework is regarded as a whole for training, similar characteristics are learned among the branch structures, so that the integration capability is reduced or lost. Training of neural network parameters includes forward propagation and backward propagation, typically a combination of forward propagation and backward propagation completes one parameter update. Unlike conventional neural network integration structures, the neural network integration structure presented herein fuses multiple component neural networks into one network, and all/part of the branching structure has some shared layers, resulting in that these shared parameters cannot be updated only by the sub-training data sets. Furthermore, to enhance generalization capability, the branch structure is trained in different training subsets. Thus, forward propagation propagates only in the respective branching structure, and the direction of each forward propagation is controlled by the connections between the fully shared layer, the partially shared layer, and the independent layers. By forward propagation we can get the predicted outcome of the branch structure:
（6）
wherein the method comprises the steps ofIs the predicted outcome of the j-th branch structure, < >>Representing the nonlinear relation between the input and output of the branch structure, < >>Is a training collected by self-help methodSubset, D, is the entire training set. The loss of each branch structure may be represented by the following formula:
（7）
wherein,is a loss function, +.>Is the corresponding tag. Since all branch structures have the same structure and task, the average loss of the corresponding branch structure is taken as the final loss of back propagation.
（8）
The flue-cured tobacco moisture content prediction network integration framework mixes the fully shared layer, the partially shared layer and the independent layer, and is not suitable for training the encoder as a whole. The encoder is divided into three stages to train the full shared layer, the partial shared layer and the independent layer step by step, and the training algorithm flow chart is shown in fig. 4.
For parameter training of the flue-cured tobacco moisture content prediction network integration framework, equation (6) is expressed by the following formula:
（9）
wherein,，/>and->Nonlinear relationships between inputs and outputs with respect to the fully shared module, the partially shared module, and the independent modules, respectively.
The first stage is to train the fully shared module, all of the training data is used to train the fully shared layer of the integration framework. The training parameters of the fully shared modules are reserved through the training branch structure, the parameters of the rest modules are abandoned, and then the weights of the fully shared modules in the branch structure are directly transplanted to the fully shared modules in the network integrated structure.
The second stage is to train the partially shared modules, at which stage the weights of the fully shared modules are fixed and the other module weights are randomly initialized. Obtaining a plurality of training subsets by self-service method, equation (9) is represented by the following formula:
（10）
wherein the method comprises the steps ofIs a fixed fully shared module nonlinear relationship, < >>Is part of training set, ++>Is a partially shared module nonlinear relationship requiring training.
The third stage is to train the independent modules, at which the weights of the fully shared modules and the partially shared modules are fixed and the weights of the other modules are initialized randomly. Using self-sampling to divide the training data set into as many training subsets as there are branch structures, equation (9) is represented by the following equation:
（11）
wherein,is a fixed partially shared module nonlinear relationship, < >>Is independent layer nonlinear relation requiring training, < ->Representing as many training subsets as there are branch structures.
Through the above stage, various parameters of the complete sharing module, the partial sharing module and the independent module can be learned to describe the characteristics of each branch structure, and the prediction result of each branch structure on the moisture content of tobacco leaves can be obtained.
The foregoing is merely illustrative of the present invention, and the present invention is not limited thereto, and any person skilled in the art will readily recognize that variations or substitutions are within the scope of the present invention.
Claims (7)
1. The method for predicting the moisture content of the cured tobacco leaves is characterized by comprising the following steps of:
s10, acquiring a color tobacco leaf image acquired by a camera in the baking process;
s20, inputting the tobacco leaf image into a scattering-like network integrated structure formed by a complete sharing module, a partial sharing module and an independent module;
s30, extracting initial characteristics of the tobacco leaf images by using a complete sharing module;
s40, extracting the initial features by using a part sharing module to obtain features of a plurality of branch structures;
s50, independently extracting different features for each branch structure by using an independent module;
s60, outputting a predicted tobacco leaf water content by each branch structure;
s70, calculating the predicted water content output by each branch structure by adopting an arithmetic average strategy;
s80, outputting a final tobacco leaf water content prediction result obtained by calculation;
the step of extracting the initial characteristics of the tobacco leaf image by using the complete sharing module specifically comprises the following steps: defining a network structure of a complete sharing module, wherein the network structure comprises an input layer, a plurality of convolution layers or a complete connection layer; determining operation parameters of each layer in the complete sharing module, including convolution kernel size and step length; forward spreading the image input in the step S20 through a complete sharing module; the feature vector of the output end of the complete sharing module is saved and used as an initial feature;
the step of extracting the initial features by using a partial sharing module to obtain features of a plurality of branch structures specifically includes: confirming the number n of required branch structures, and defining the network structure of part of the sharing modules; copying the initial characteristics output by the complete sharing module into n parts, and inputting one branch into each part; setting operation parameters of a convolution layer in each branch; forward operation is carried out on the characteristics of each branch in each network; storing the characteristics of each branch output end as the characteristics of each branch structure;
the step of extracting different features for each branch structure by using an independent module specifically comprises the following steps: defining a network structure of independent modules in each branch structure; inputting each branch characteristic output in the step S40 into a corresponding independent module; setting network parameters of independent modules in each branch; performing forward extraction operation of features in each branch independent module; and storing the characteristic vectors of the output ends of the independent branch modules.
2. The method for predicting the moisture content of cured tobacco leaves according to claim 1, wherein the step of acquiring the color tobacco leaf image acquired by the camera during the curing process comprises the following steps: setting baking equipment to obtain tobacco leaves to be baked; setting a color image acquisition device to enable a shooting range of the color image acquisition device to cover a baking device; heating the baking equipment and sending the tobacco leaves into the baking equipment for baking; controlling an image acquisition device to shoot tobacco leaves in a baking process regularly or continuously to obtain a plurality of images; and calibrating and correcting the image.
3. The method according to claim 1, wherein the step of inputting the tobacco leaf image into a scattering-like network integration structure composed of a complete sharing module, a partial sharing module and an independent module, comprises: the scattering-like network integrated structure is constructed and comprises a complete sharing convolution module, a partial sharing convolution module and an independent convolution module; determining the number of branch networks in the independent module, wherein each branch network corresponds to a subsequent prediction output; initializing network parameters; each frame of tobacco leaf image is input to the input end of the network integrated structure.
4. The method of claim 1, wherein the step of outputting a predicted moisture content of the tobacco per branch structure comprises: setting a prediction layer of each branch structure, wherein the prediction layer comprises an output node; connecting the characteristics output by each branch independent module to a corresponding prediction layer; forward operation is carried out in the prediction layer of each branch to obtain the prediction output of each branch; and storing the prediction results output by each branch structure.
5. The method for predicting the moisture content of cured tobacco leaves according to claim 1, wherein the step of calculating the predicted moisture content output by each branch structure by adopting an arithmetic average strategy comprises the following steps: reading each branch prediction output saved in the step S60; calculating an arithmetic mean of the branch prediction outputs; the arithmetic mean is output as the final prediction.
6. A computer readable storage medium having stored therein program instructions which, when executed, are adapted to carry out a cured tobacco moisture content prediction method as claimed in any one of claims 1 to 5.
7. A cured tobacco moisture content prediction system comprising the computer readable storage medium of claim 6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311773785.1A CN117457101B (en) | 2023-12-22 | 2023-12-22 | Method, medium and system for predicting moisture content of cured tobacco leaves |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311773785.1A CN117457101B (en) | 2023-12-22 | 2023-12-22 | Method, medium and system for predicting moisture content of cured tobacco leaves |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117457101A CN117457101A (en) | 2024-01-26 |
| CN117457101B true CN117457101B (en) | 2024-03-26 |
Family
ID=89593252
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311773785.1A Active CN117457101B (en) | 2023-12-22 | 2023-12-22 | Method, medium and system for predicting moisture content of cured tobacco leaves |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117457101B (en) |
Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2008018753A1 (en) * | 2006-08-08 | 2008-02-14 | Yong Seok Shim | Tobacco having constant flavor |
| CN110443266A (en) * | 2018-05-04 | 2019-11-12 | 上海商汤智能科技有限公司 | Object prediction method and device, electronic equipment and storage medium |
| CN111950656A (en) * | 2020-08-25 | 2020-11-17 | 深圳思谋信息科技有限公司 | Image recognition model generation method and device, computer equipment and storage medium |
| CN112818827A (en) * | 2021-01-28 | 2021-05-18 | 贵州省烟草科学研究院 | Image recognition-based method for judging stage temperature control point in tobacco leaf baking process |
| CN113331459A (en) * | 2021-07-09 | 2021-09-03 | 中国农业科学院烟草研究所(中国烟草总公司青州烟草研究所) | Intelligent tobacco leaf baking control method |
| CN113569732A (en) * | 2021-07-27 | 2021-10-29 | 厦门理工学院 | Face attribute recognition method and system based on parallel sharing multitask network |
| CN114355857A (en) * | 2022-01-12 | 2022-04-15 | 华中科技大学 | Intelligent control system, method, medium, equipment and terminal for tobacco leaf baking |
| CN114821462A (en) * | 2022-03-29 | 2022-07-29 | 西安电子科技大学 | Target detection method based on multi-branch parallel hybrid hole coding neural network |
| CN114997463A (en) * | 2022-04-25 | 2022-09-02 | 华中科技大学 | Flue-cured tobacco process stage prediction method and system based on model fusion |
| CN116434045A (en) * | 2023-03-07 | 2023-07-14 | 中国农业科学院烟草研究所(中国烟草总公司青州烟草研究所) | Intelligent identification method for tobacco leaf baking stage |
| CN116432070A (en) * | 2021-12-29 | 2023-07-14 | 上海交通大学 | ECG signal classification system and method based on deep learning neural network |
| CN116524224A (en) * | 2022-12-09 | 2023-08-01 | 云南省烟草农业科学研究院 | Machine vision-based method and system for detecting type of cured tobacco leaves |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113988373B (en) * | 2021-09-27 | 2022-09-30 | 河海大学 | Multi-task massive user load prediction method based on multi-channel convolutional neural network |
-
2023
- 2023-12-22 CN CN202311773785.1A patent/CN117457101B/en active Active
Patent Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2008018753A1 (en) * | 2006-08-08 | 2008-02-14 | Yong Seok Shim | Tobacco having constant flavor |
| CN110443266A (en) * | 2018-05-04 | 2019-11-12 | 上海商汤智能科技有限公司 | Object prediction method and device, electronic equipment and storage medium |
| CN111950656A (en) * | 2020-08-25 | 2020-11-17 | 深圳思谋信息科技有限公司 | Image recognition model generation method and device, computer equipment and storage medium |
| CN112818827A (en) * | 2021-01-28 | 2021-05-18 | 贵州省烟草科学研究院 | Image recognition-based method for judging stage temperature control point in tobacco leaf baking process |
| CN113331459A (en) * | 2021-07-09 | 2021-09-03 | 中国农业科学院烟草研究所(中国烟草总公司青州烟草研究所) | Intelligent tobacco leaf baking control method |
| CN113569732A (en) * | 2021-07-27 | 2021-10-29 | 厦门理工学院 | Face attribute recognition method and system based on parallel sharing multitask network |
| CN116432070A (en) * | 2021-12-29 | 2023-07-14 | 上海交通大学 | ECG signal classification system and method based on deep learning neural network |
| CN114355857A (en) * | 2022-01-12 | 2022-04-15 | 华中科技大学 | Intelligent control system, method, medium, equipment and terminal for tobacco leaf baking |
| CN114821462A (en) * | 2022-03-29 | 2022-07-29 | 西安电子科技大学 | Target detection method based on multi-branch parallel hybrid hole coding neural network |
| CN114997463A (en) * | 2022-04-25 | 2022-09-02 | 华中科技大学 | Flue-cured tobacco process stage prediction method and system based on model fusion |
| CN116524224A (en) * | 2022-12-09 | 2023-08-01 | 云南省烟草农业科学研究院 | Machine vision-based method and system for detecting type of cured tobacco leaves |
| CN116434045A (en) * | 2023-03-07 | 2023-07-14 | 中国农业科学院烟草研究所(中国烟草总公司青州烟草研究所) | Intelligent identification method for tobacco leaf baking stage |
Non-Patent Citations (3)
| Title |
|---|
| Xu Miao et.al.Synchronous recognition of multiple tobacco leaf parts based on improved Mask R-CNN.《Journal of Hunan Agricultural University》.2023,第49卷(第2期),第170-175页. * |
| 基于BP神经网络的烘烤变黄期烟叶含水量预测;张广普;《天津农业科学》;20211001;第27卷(第10期);第62-67页 * |
| 基于支持向量机回归的鲜烟叶含水量预测模型;詹攀;谢守勇;刘军;黄河;李佩原;;西南大学学报(自然科学版);20160420(04);第170-175页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117457101A (en) | 2024-01-26 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108615071B (en) | Model testing method and device | |
| CN110363290B (en) | Image recognition method, device and equipment based on hybrid neural network model | |
| CN111274860A (en) | Machine vision-based online automatic tobacco leaf grade sorting identification method | |
| CN112784913B (en) | MiRNA-disease association prediction method and device based on fusion of multi-view information of graphic neural network | |
| CN109102885B (en) | Automatic cataract grading method based on combination of convolutional neural network and random forest | |
| CN113012811B (en) | Traditional Chinese medicine syndrome diagnosis and health evaluation method combining deep convolutional network and graph neural network | |
| CN113436167B (en) | No-reference color image quality evaluation method based on deep learning and visual perception | |
| CN106203298A (en) | Biological feather recognition method and device | |
| CN111028203B (en) | CNN blind image quality evaluation method based on significance | |
| CN104636759A (en) | Method for obtaining picture recommending filter information and picture filter information recommending system | |
| CN113011487B (en) | Open set image classification method based on joint learning and knowledge migration | |
| CN110516537A (en) | A kind of face age estimation method based on from step study | |
| CN117457101B (en) | Method, medium and system for predicting moisture content of cured tobacco leaves | |
| CN113658091A (en) | Image evaluation method, storage medium and terminal equipment | |
| CN111931698B (en) | Image deep learning network construction method and device based on small training set | |
| CN109359675A (en) | Image processing method and equipment | |
| CN113159294A (en) | Sample selection algorithm based on companion learning | |
| CN116449713A (en) | Tobacco leaf baking simulation method and system based on high-temperature baking room | |
| CN110866866B (en) | Image color imitation processing method and device, electronic equipment and storage medium | |
| CN112199550A (en) | Short video click rate prediction method based on emotion capsule network | |
| CN108665455B (en) | Method and device for evaluating image significance prediction result | |
| CN116541767A (en) | Multi-element greenhouse environment parameter prediction method and system based on graphic neural network | |
| JP4320438B2 (en) | Character string extraction processing device for printed matter | |
| CN109697511B (en) | Data reasoning method and device and computer equipment | |
| CN112435751B (en) | Peritoneal dialysis mode auxiliary recommendation system based on variation inference and deep learning |
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 |