A method and system for predicting the process stage of flue-cured tobacco based on model fusion

Abstract

The present application belongs to the technical field of tobacco curing, and discloses a flue-cured tobacco process stage prediction method and system based on model fusion, comprising: constructing and training a flue-cured tobacco prediction model fusing a LightGBM model, a GRU model and a CNN model, and acquiring image, temperature and humidity, and water loss data in the tobacco curing process; and using the trained flue-cured tobacco prediction model fusing the LightGBM model, the GRU model and the CNN model to predict the flue-cured tobacco process stage based on the preprocessed image, temperature and humidity, and water loss data in the tobacco curing process. The model fusion technology of the present application trains a fusion model, and the trained fusion model is used for flue-cured tobacco process stage prediction, which can realize real-time prediction of tobacco, improve the prediction accuracy and efficiency, and reduce the working intensity of the curing personnel and save labor costs.

Description

Technical Field

[0001] This invention belongs to the field of tobacco curing technology, and particularly relates to a method and system for predicting stages of tobacco curing process based on model fusion. Background Technology

[0002] Currently, tobacco is one of my country's major economic crops with high economic benefits. However, the prediction and adjustment of flue-cured tobacco processing technology in my country is still mainly done manually, which has problems such as high labor intensity, low work efficiency and high labor costs.

[0003] With the development of computer technology, it has become a trend to use computer technology to replace manual judgment of tobacco leaf condition. Theoretically, the changes in tobacco leaf area, color, texture, water loss, and the content of some chemical substances during the tobacco curing process are significant and can be used as input features for the model. However, due to the complex environment of intensive curing barns, it is difficult to obtain the tobacco leaf area and chemical substance content in real time. Only features such as color, texture, and water loss are relatively easy to extract, which reduces the number of usable features for the model.

[0004] The existing technology has the following problems and shortcomings:

[0005] Existing technologies mainly rely on manual inspection of tobacco leaves during curing, which is highly subjective, prone to errors, labor-intensive, and inaccurate. In contrast, the accuracy of model recognition is low.

[0006] The difficulty in solving the defects mentioned above is:

[0007] How to reduce manual intervention in the prediction of all stages of tobacco curing process and reduce labor intensity; how to accurately acquire and analyze the stages of tobacco curing process by utilizing multiple information sources; and how to further mine and collect data to improve curing quality in an intelligent manner.

[0008] The significance of solving the above problems is as follows:

[0009] By studying and analyzing the collected data, it is possible to achieve automated and intelligent curing of tobacco, with self-controlled environmental parameters, thereby reducing the labor intensity of tobacco curing, breaking through the technical bottlenecks of tobacco curing, and making great significance for making up for the shortcomings of tobacco curing technology and raw materials, and promoting the high-quality development of the tobacco industry. Summary of the Invention

[0010] To address the problems existing in the prior art, this invention provides a method and system for predicting the stages of flue-cured tobacco processing based on model fusion.

[0011] This invention is implemented as follows: a method for predicting the stages of flue-cured tobacco processing based on model fusion, the method comprising:

[0012] A flue-cured tobacco prediction model integrating LightGBM, GRU, and CNN models was constructed and trained. Images, temperature, humidity, and water loss data during the tobacco curing process were acquired and preprocessed. The trained flue-cured tobacco prediction model was then used to predict the stages of the flue-curing process based on the preprocessed images, temperature, humidity, and water loss data during the tobacco curing process.

[0013] Furthermore, the model fusion-based method for predicting the stages of flue-cured tobacco processing includes the following steps:

[0014] Step 1: Collect relevant data on tobacco leaves and curing barns during the curing process according to a pre-set collection cycle; and construct unstructured and structured datasets based on the collected data on tobacco leaves and curing barns.

[0015] Step 2: Extract tobacco leaf state based on tobacco leaf images in unstructured dataset, and digitally encode the extracted tobacco leaf state to obtain tobacco leaf state code.

[0016] Step 3: Construct a flue-cured tobacco prediction model that integrates the LightGBM model, GRU model, and CNN model; and train the flue-cured tobacco prediction model that integrates the LightGBM model, GRU model, and CNN model using the unstructured dataset, structured dataset, and tobacco leaf state code.

[0017] Step four: Collect relevant data on tobacco leaves and curing barns in real time during the curing process, preprocess the collected data, and use the trained tobacco curing prediction model that integrates LightGBM, GRU, and CNN models to predict the stages of the tobacco curing process based on the preprocessed data.

[0018] Furthermore, the relevant data for the tobacco leaves and curing barn include: tobacco leaf images, temperature and humidity data, and water loss data.

[0019] Furthermore, in step one, constructing unstructured and structured datasets based on the collected data from tobacco leaves and curing barns includes:

[0020] The median filtering nonlinear smoothing algorithm and the dark channel dehazing algorithm are used to perform noise reduction and color correction preprocessing on the tobacco leaf images in the relevant data of tobacco leaves and curing barns. The preprocessed tobacco leaf images are saved in a unified format to the specified folder and converted into CSV files to obtain an unstructured dataset.

[0021] Simultaneously, feature sets are obtained by extracting features from the temperature, humidity, and water loss data of the collected tobacco leaves and curing barns. Then, missing value processing, outlier processing, and data standardization are performed on the feature data in the feature set to obtain a structured dataset.

[0022] Furthermore, in step three, the flue-cured tobacco prediction model that integrates the LightGBM model, GRU model, and CNN model includes:

[0023] The first base classifier consists of a LightGBM network, a GRU network, and a CNN network; it is used for preliminary identification of the tobacco curing process.

[0024] The second-layer meta-classifier, composed of LightGBM, is used for secondary identification during the tobacco curing process.

[0025] The CNN network is an SE-ResNet50 with a residual learning module; the CNN network uses an improved cross-entropy loss function and selects the Adam algorithm as the adaptive learning rate optimization algorithm.

[0026] Furthermore, in step three, training the flue-cured tobacco prediction model, which integrates the LightGBM model, GRU model, and CNN model, using unstructured datasets, structured datasets, and tobacco leaf state codes includes:

[0027] (1) Based on the two-layer Stacking model fusion method, the structured dataset and the unstructured dataset are used as the input data of the first-layer base classifier, and the tobacco leaf state code is used as the output result of the first-layer base classifier. The LightGBM network, GRU network and CNN network of the first-layer base classifier are trained to obtain the trained first-layer base classifier:

[0028] (1.1) The structured dataset is used as the input data of the LightGBM network of the first base classifier, the tobacco leaf state code is used as the recognition result of the LightGBM network, the LightGBM network is trained, and the model parameters of the LightGBM network are automatically adjusted according to the importance using the grid search method. The state code of the first new training set obtained after the LightGBM network is trained is denoted as A1.

[0029] (1.2) The structured dataset is used as the input data of the GRU network of the first base classifier, and the tobacco leaf state code is used as the output result of the GRU network. The GRU network is trained, and the optimal time step, number of hidden layers and number of neurons in each layer are determined by trial and error. The state code of the second new training set obtained after training the GRU network is denoted as A2.

[0030] (1.3) The unstructured dataset is used as the input data of the CNN network of the first base classifier, and the tobacco leaf state code is used as the output result of the CNN network to train the CNN network; the SoftMax classifier is applied to classify the images, and the state code of the third new training set obtained after training the CNN network is denoted as A3;

[0031] (2) The recognition result output by the first layer base classifier is used as the input data of the second layer meta classifier, and the tobacco leaf state code is used as the output recognition result of the second layer meta classifier. The LightGBM network of the second layer meta classifier is trained, and the model parameters are automatically adjusted by the grid search method to obtain the trained flue-cured tobacco prediction model that integrates the LightGBM model, GRU model and CNN model.

[0032] Another objective of this invention is to provide a model fusion-based flue-cured tobacco process stage prediction system that implements the aforementioned model fusion-based flue-cured tobacco process stage prediction method, the model fusion-based flue-cured tobacco process stage prediction system comprising:

[0033] The dataset construction module is used to collect relevant data on tobacco leaves and curing barns during the curing process according to a pre-set collection cycle; and to construct unstructured and structured datasets based on the collected data on tobacco leaves and curing barns; at the same time, it is used to extract tobacco leaf status based on tobacco leaf images in the unstructured dataset, and to digitally encode the extracted tobacco leaf status to obtain tobacco leaf status codes.

[0034] The fusion model building module is used to build a flue-cured tobacco prediction model that integrates the LightGBM model, GRU model, and CNN model.

[0035] The fusion model module is used to train the flue-cured tobacco prediction model, which integrates the LightGBM model, GRU model, and CNN model, using the unstructured dataset, the structured dataset, and the tobacco leaf state code.

[0036] The data preprocessing module is used to collect relevant data on tobacco leaves and curing barns in real time during the curing process, and to preprocess the collected data on tobacco leaves and curing barns.

[0037] The prediction module is used to predict the stages of the tobacco curing process based on pre-processed tobacco leaves and relevant data from the curing barn using a pre-trained flue-cured tobacco prediction model that integrates LightGBM, GRU, and CNN models.

[0038] Another object of the present invention is to provide a computer device, the computer device including a memory and a processor, the memory storing a computer program, which, when executed by the processor, causes the processor to perform the model fusion-based method for predicting the stage of flue-cured tobacco processing as follows:

[0039] Step 1: Collect relevant data on tobacco leaves and curing barns during the curing process according to a pre-set collection cycle; and construct unstructured and structured datasets based on the collected data on tobacco leaves and curing barns.

[0040] Step 2: Extract tobacco leaf state based on tobacco leaf images in unstructured dataset, and digitally encode the extracted tobacco leaf state to obtain tobacco leaf state code.

[0041] Step 3: Construct a flue-cured tobacco prediction model that integrates the LightGBM model, GRU model, and CNN model; and train the flue-cured tobacco prediction model that integrates the LightGBM model, GRU model, and CNN model using the unstructured dataset, structured dataset, and tobacco leaf state code.

[0042] Step four: Collect relevant data on tobacco leaves and curing barns in real time during the curing process, preprocess the collected data, and use the trained tobacco curing prediction model that integrates LightGBM, GRU, and CNN models to predict the stages of the tobacco curing process based on the preprocessed data.

[0043] Another object of the present invention is to provide a computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the model fusion-based method for predicting the stage of flue-cured tobacco processing as follows:

[0044] Step 1: Collect relevant data on tobacco leaves and curing barns during the curing process according to a pre-set collection cycle; and construct unstructured and structured datasets based on the collected data on tobacco leaves and curing barns.

[0045] Step 2: Extract tobacco leaf state based on tobacco leaf images in unstructured dataset, and digitally encode the extracted tobacco leaf state to obtain tobacco leaf state code.

[0046] Step 3: Construct a flue-cured tobacco prediction model that integrates the LightGBM model, GRU model, and CNN model; and train the flue-cured tobacco prediction model that integrates the LightGBM model, GRU model, and CNN model using the unstructured dataset, structured dataset, and tobacco leaf state code.

[0047] Step four: Collect relevant data on tobacco leaves and curing barns in real time during the curing process, preprocess the collected data, and use the trained tobacco curing prediction model that integrates LightGBM, GRU, and CNN models to predict the stages of the tobacco curing process based on the preprocessed data.

[0048] Another objective of this invention is to provide an information data processing terminal for implementing the model fusion-based tobacco curing process stage prediction system.

[0049] Based on the above technical solutions and the technical problems solved, please analyze the advantages and positive effects of the technical solution to be protected by this invention from the following aspects:

[0050] The model fusion technology of this invention trains a fusion model, and the trained fusion model can be used to predict the tobacco curing process in real time, which improves the prediction accuracy and efficiency, while reducing the workload of curing personnel and saving labor costs.

[0051] This invention preprocesses the tobacco leaf images to create a dataset suitable for prediction by a convolutional neural network (CNN) model. The CNN used is SE-ResNet50 with a residual learning module. An improved cross-entropy loss function is employed, and the Adam algorithm is selected as the adaptive learning rate optimization algorithm. Simultaneously, data such as temperature, humidity, and water loss rate collected by sensors in the curing barn are used to extract features, creating a structured dataset usable by the model. The simultaneous use of structured and unstructured data makes the prediction results more convincing and practical.

[0052] This invention utilizes a stacking fusion model method, using structured datasets to train a LightGBM model and a GRU model, and using unstructured data to train a CNN model to obtain the first layer model. Then, using the first feature set and the output of the first layer model, the corresponding LightGBM model for the second layer is trained to obtain the fusion model.

[0053] This invention provides a model fusion-based method for predicting the stages of flue-cured tobacco processing, enabling more accurate predictions. It solves the technical challenge of high cost and manpower required for process stage prediction that has plagued tobacco farmers. No similar model fusion method for predicting flue-cured tobacco processing stages has been found in literature and patent search engines, filling a technological gap both domestically and internationally. Attached Figure Description

[0054] Figure 1 This is a schematic diagram of the tobacco curing process stage prediction method based on model fusion provided in this embodiment of the invention.

[0055] Figure 2 This is a flowchart of the tobacco curing process stage prediction method based on model fusion provided in this embodiment of the invention;

[0056] Figure 3 This is a diagram of the CNN network structure provided in an embodiment of the present invention;

[0057] Figure 4 This is a schematic diagram of the structure of the tobacco curing process stage prediction system based on model fusion provided in an embodiment of the present invention;

[0058] In the diagram: 1. Dataset construction module; 2. Fusion model construction module; 3. Fusion model module; 4. Data preprocessing module; 5. Prediction module. Detailed Implementation

[0059] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0060] I. Explanatory and Illustrative Embodiments. To enable those skilled in the art to fully understand how the present invention is specifically implemented, this section provides an explanatory and illustrative description of the embodiments described in the claims.

[0061] like Figure 1 As shown, the model fusion-based method for predicting the stages of flue-cured tobacco processing provided in this embodiment of the invention includes:

[0062] A flue-cured tobacco prediction model integrating LightGBM, GRU, and CNN models was constructed and trained. Images, temperature, humidity, and water loss data of tobacco leaves during the curing process were acquired. The trained flue-cured tobacco prediction model integrating LightGBM, GRU, and CNN models was used to predict the stages of the flue-curing process based on the preprocessed images, temperature, humidity, and water loss data of tobacco leaves during the curing process.

[0063] like Figure 2 As shown, the model fusion-based method for predicting the stages of flue-cured tobacco processing provided in this embodiment of the invention includes the following steps:

[0064] S101, Collect relevant data on tobacco leaves and curing barns during the curing process according to a pre-set collection cycle; and construct unstructured and structured datasets based on the collected relevant data on tobacco leaves and curing barns.

[0065] S102, tobacco leaf state is extracted based on tobacco leaf images in unstructured dataset, and the extracted tobacco leaf state is digitally encoded to obtain tobacco leaf state code;

[0066] S103, Construct a flue-cured tobacco prediction model that integrates the LightGBM model, GRU model, and CNN model; and train the flue-cured tobacco prediction model that integrates the LightGBM model, GRU model, and CNN model using the unstructured dataset, structured dataset, and tobacco leaf state code.

[0067] S104 collects relevant data on tobacco leaves and curing barns in real time during the curing process, preprocesses the collected data, and uses a trained tobacco curing prediction model that integrates LightGBM, GRU, and CNN models to predict the stages of the tobacco curing process based on the preprocessed data.

[0068] The relevant data on tobacco leaves and curing barns provided in this embodiment of the invention include: tobacco leaf images, temperature and humidity data, and water loss data.

[0069] In step S101, the construction of unstructured and structured datasets based on collected data from tobacco leaves and curing barns, as provided in this embodiment of the invention, includes:

[0070] The median filtering nonlinear smoothing algorithm and the dark channel dehazing algorithm are used to perform noise reduction and color correction preprocessing on the tobacco leaf images in the relevant data of tobacco leaves and curing barns. The preprocessed tobacco leaf images are saved in a unified format to the specified folder and converted into CSV files to obtain an unstructured dataset.

[0071] Simultaneously, feature sets are obtained by extracting features from the temperature, humidity, and water loss data of the collected tobacco leaves and curing barns. Then, missing value processing, outlier processing, and data standardization are performed on the feature data in the feature set to obtain a structured dataset.

[0072] In step S103, the flue-cured tobacco prediction model merging the LightGBM model, GRU model, and CNN model provided in this embodiment of the invention includes:

[0073] The first base classifier consists of a LightGBM network, a GRU network, and a CNN network; it is used to make preliminary predictions for the tobacco curing process.

[0074] The second-level meta-classifier, composed of LightGBM, is used for secondary prediction during the flue-cured tobacco processing stage.

[0075] The CNN network provided in this embodiment of the invention is SE-ResNet50 with a residual learning module; the CNN network adopts an improved cross-entropy loss function and selects the Adam algorithm as the adaptive learning rate optimization algorithm.

[0076] In step S103, the training of the flue-cured tobacco prediction model fusion of the LightGBM model, GRU model, and CNN model using unstructured datasets, structured datasets, and tobacco leaf state codes provided in this embodiment of the invention includes:

[0077] (1) According to the two-layer Stacking model fusion method, the structured dataset and the unstructured dataset are used as the input data of the first-layer base classifier, and the tobacco leaf state code is used as the output result of the first-layer base classifier. The LightGBM network, GRU network and CNN network of the first-layer base classifier are trained to obtain the trained first-layer base classifier.

[0078] (2) The recognition result output by the first layer base classifier is used as the input data of the second layer meta classifier, and the tobacco leaf state code is used as the output recognition result of the second layer meta classifier. The LightGBM network of the second layer meta classifier is trained, and the model parameters are automatically adjusted by the grid search method to obtain the trained flue-cured tobacco prediction model that integrates the LightGBM model, GRU model and CNN model.

[0079] Step (1) provided in this embodiment of the invention includes:

[0080] (1.1) The structured dataset is used as the input data of the LightGBM network of the first base classifier, the tobacco leaf state code is used as the recognition result of the LightGBM network, the LightGBM network is trained, and the model parameters of the LightGBM network are automatically adjusted according to the importance using the grid search method. The state code of the first new training set obtained after the LightGBM network is trained is denoted as A1.

[0081] (1.2) The structured dataset is used as the input data of the GRU network of the first base classifier, and the tobacco leaf state code is used as the output result of the GRU network. The GRU network is trained, and the optimal time step, number of hidden layers and number of neurons in each layer are determined by trial and error. The state code of the second new training set obtained after training the GRU network is denoted as A2.

[0082] (1.3) The unstructured dataset is used as the input data of the CNN network of the first base classifier, and the tobacco leaf state code is used as the output result of the CNN network to train the CNN network; the SoftMax classifier is applied to classify the images, and the state code of the third new training set obtained after training the CNN network is denoted as A3.

[0083] like Figure 4 As shown, the model fusion-based tobacco curing process stage prediction system provided in this embodiment of the invention includes:

[0084] Dataset construction module 1 is used to collect relevant data of tobacco leaves and curing barns during curing according to a pre-set collection cycle; and to construct unstructured and structured datasets based on the collected relevant data of tobacco leaves and curing barns; at the same time, it is used to extract tobacco leaf status based on tobacco leaf images in the unstructured dataset, and to digitally encode the extracted tobacco leaf status to obtain tobacco leaf status code.

[0085] Fusion model building module 2 is used to build a flue-cured tobacco prediction model that integrates the LightGBM model, GRU model, and CNN model;

[0086] The fusion model module 3 is used to train the flue-cured tobacco prediction model that integrates the LightGBM model, GRU model, and CNN model using the unstructured dataset, the structured dataset, and the tobacco leaf state code.

[0087] Data preprocessing module 4 is used to collect relevant data of tobacco leaves and curing barn during the curing process in real time, and to preprocess the collected data of tobacco leaves and curing barn.

[0088] Prediction module 5 is used to predict the tobacco curing process stages based on preprocessed tobacco leaves and relevant data from the curing barn using a trained flue-cured tobacco prediction model that integrates the LightGBM model, GRU model, and CNN model.

[0089] The technical solution of the present invention will be further described below with reference to specific embodiments.

[0090] The method for predicting the stages of flue-cured tobacco processing based on model fusion provided in this invention includes:

[0091] (1) During the tobacco curing process, the original dataset of tobacco leaves and curing barn is collected at a preset collection cycle. The original dataset includes: tobacco leaf images, temperature and humidity data and water loss data.

[0092] (2) The processed tobacco leaf images are used as an unstructured dataset, and the temperature and humidity data and water loss data are used as a structured dataset, which together constitute the initial dataset.

[0093] (3) Based on the feature data of the structured data, obtain the feature set, and digitally encode the tobacco leaf state in the tobacco leaf image to obtain the tobacco leaf state code;

[0094] (4) According to the two-layer Stacking model fusion method, the structured feature set and the unstructured dataset are used as the input data of the first layer model, and the tobacco leaf state code is used as the recognition result of the first layer model. The LightGBM model, GRU model and convolutional neural network CNN model are trained respectively to obtain the first layer base classifier.

[0095] (5) The recognition results output by the three base learners of the first layer are used as the input data of the second layer model, and the tobacco leaf state code is used as the recognition result of the second layer model. The LightGBM model, which is used as the second layer meta-classifier, is trained to obtain the fusion model.

[0096] (6) Real-time acquisition of raw data and raw images of tobacco leaves during the baking process, which are then processed and used as input for the model to predict the baking process stages of tobacco leaves.

[0097] In one embodiment, step (1) specifically includes:

[0098] During the tobacco curing process, original images, temperature and humidity data, and water loss data of the tobacco leaves are collected at a preset collection cycle; the tobacco leaf images, temperature and humidity data, and water loss data are then transmitted to the cloud platform database to form the original dataset.

[0099] In one embodiment, step (2) specifically includes:

[0100] Median filtering, nonlinear smoothing, and dark channel dehazing algorithms are used to denoise and correct the colors of the original image, resulting in a processed image. This image is then saved in a standardized format as a tobacco leaf image in a designated folder and converted into a CSV file as an unstructured dataset. Missing value processing, outlier processing, and data standardization are then performed on the feature data in the structured data feature set to obtain the structured data.

[0101] In one embodiment, processing the feature data in the structured data feature set by handling missing values, outliers, and standardizing the data to obtain the feature set as structured data specifically includes:

[0102] When the initial feature set contains a small number of missing feature values ​​for the first data, the average of the adjacent feature values ​​before and after the small number of missing feature values ​​is used for imputation and filling; when the initial feature set contains a large number of missing feature values ​​for the second data, the second data is directly deleted.

[0103] Outlier identification is performed on the initial feature set after missing value processing using the box plot method, and the outlier is corrected using the functional relationship between the data before and after the outlier.

[0104] In one embodiment, step (3) specifically includes:

[0105] Feature selection is performed on the structured dataset to obtain an initial feature set; the tobacco leaf states in the tobacco leaf images are digitally encoded, including initial yellowing, yellowing, yellowing and wilting, leaf turning to vegetative state, dry leaf, and dry vegetative state.

[0106] In one embodiment, step (4) specifically includes:

[0107] The structured data feature set is used as the input data of the first-layer model, and the tobacco leaf state code is used as the recognition result of the first-layer model. The LightGBM model is trained, and the model parameters of the LightGBM model are automatically adjusted according to the importance using a grid search method. The state code of the first new training set obtained after training the LightGBM model is denoted as A1.

[0108] The structured data feature set is used as the input data for the first-layer model, and the tobacco leaf state code is used as the recognition result of the first-layer model. The GRU model is trained, and the optimal time step, number of hidden layers, and number of neurons in each layer are determined by trial and error. The state code of the second new training set obtained after training the GRU model is denoted as A2.

[0109] The processed unstructured dataset of tobacco leaf images is used as the input data for the first-layer model, and the tobacco leaf state code is used as the recognition result of the first-layer model to train the convolutional neural network model CNN. The SoftMax classifier is applied to classify the images, and the state code of the third new training set obtained after training the convolutional neural network model CNN is denoted as A3.

[0110] In one embodiment, the CNN model specifically includes:

[0111] The Convolutional Neural Network (CNN) model selected is SE-ResNet50 with a residual learning module. The CNN model uses a modified cross-entropy loss function—focal loss—to address the imbalance in the number of images across different categories in the dataset. The Adam algorithm is chosen as the adaptive learning rate optimization algorithm for the CNN model.

[0112] In one embodiment, step (5) specifically includes:

[0113] The first new training set A1, the second new training set A2, and the third new training set A3 are used as input data for the LightGBM model; the tobacco leaf state code is used as the recognition result of the second layer model and then as the recognition result of the LightGBM model. The model parameters are automatically adjusted using a grid search method to obtain the fusion model.

[0114] An application embodiment of the present invention provides a computer device, the computer device including a memory and a processor, the memory storing a computer program, and when the computer program is executed by the processor, the processor executes the model fusion-based method for predicting the stage of flue-cured tobacco processing.

[0115] An application embodiment of the present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the model fusion-based method for predicting the stages of flue-cured tobacco processing.

[0116] An application embodiment of the present invention provides an information data processing terminal, which is used to implement the model fusion-based tobacco curing process stage prediction system.

[0117] It should be noted that embodiments of the present invention can be implemented in hardware, software, or a combination of both. The hardware portion can be implemented using dedicated logic; the software portion can be stored in memory and executed by a suitable instruction execution system, such as a microprocessor or dedicated-design hardware. Those skilled in the art will understand that the above-described devices and methods can be implemented using computer-executable instructions and / or included in processor control code, for example, such code provided on a carrier medium such as a disk, CD, or DVD-ROM, a programmable memory such as read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. The devices and modules of the present invention can be implemented by hardware circuitry such as very large-scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, or programmable hardware devices such as field-programmable gate arrays, programmable logic devices, etc., or by software executed by various types of processors, or by a combination of the above-described hardware circuitry and software, such as firmware.

[0118] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications, equivalent substitutions, and improvements made by those skilled in the art within the scope of the technology disclosed in the present invention, and within the spirit and principles of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A model fusion-based flue-cured tobacco process stage prediction method, characterized in that, The model fusion-based flue-cured tobacco process stage prediction method comprises the following steps: Step one, collecting the related data of the tobacco and the curing barn in the baking process according to the pre-set collection period; and constructing an unstructured data set and a structured data set based on the collected related data of the tobacco and the curing barn; Step two, extracting the tobacco state based on the tobacco image in the unstructured data set, and obtaining the tobacco state code by digitally encoding the extracted tobacco state; Step three, constructing a flue-cured tobacco prediction model fusing a LightGBM model, a GRU model and a CNN model; and training the flue-cured tobacco prediction model fusing the LightGBM model, the GRU model and the CNN model by using the unstructured data set, the structured data set and the tobacco state code; Step four, collecting the related data of the tobacco and the curing barn in the baking process in real time, preprocessing the collected related data of the tobacco and the curing barn, and predicting the baking process stage of the tobacco based on the preprocessed related data of the tobacco and the curing barn by using the trained flue-cured tobacco prediction model fusing the LightGBM model, the GRU model and the CNN model; In step three, the training of the flue-cured tobacco prediction model fusing the LightGBM model, the GRU model and the CNN model by using the unstructured data set, the structured data set and the tobacco state code comprises: (1) According to a two-layer stacking model fusion method, the structured data set and the unstructured data set are used as the input data of the first layer base classifier, the tobacco state code is used as the output result of the first layer base classifier, the LightGBM network, the GRU network and the CNN network of the first layer base classifier are trained, and a trained first layer base classifier is obtained: (1.1) The structured data set is used as the input data of the LightGBM network of the first layer base classifier, the tobacco state code is used as the recognition result of the LightGBM network, the LightGBM network is trained, the model parameters of the LightGBM network are automatically adjusted by using the grid search method according to the importance, the state code of the first new training set obtained after the training of the LightGBM network is recorded as A1; (1.2) The structured data set is used as the input data of the GRU network of the first layer base classifier, the tobacco state code is used as the output result of the GRU network, the GRU network is trained, and the optimal time step, the number of hidden layers and the number of neurons in each layer are determined by using the trial-and-error method; the state code of the second new training set obtained after the training of the GRU network is recorded as A2; ​ (1.3) taking the unstructured data set as input data of a CNN network of the first layer base classifier, taking the tobacco leaf state code as output result of the CNN network, training the CNN network; applying a SoftMax classifier to classify the image, and taking the state code of a third new training set obtained after training of the CNN network as A3; (2) taking the recognition result output by the first layer base classifier as input data of the second layer meta-classifier, taking the tobacco leaf state code as output recognition result of the second layer meta-classifier, training a LightGBM network of the second layer meta-classifier, and automatically adjusting model parameters by a grid search method to obtain a trained tobacco prediction model of the fusion LightGBM model, GRU model and CNN model.

2. The method according to claim 1, wherein the model fusion based flue-cured tobacco process stage prediction method is characterized by, The related data of the tobacco leaf and the curing barn include: tobacco leaf images, temperature and humidity data and water loss data.

3. The method according to claim 1, wherein the model fusion based flue-cured tobacco process stage prediction method is characterized by, In step one, based on the collected related data of the tobacco leaf and the curing barn, an unstructured data set and a structured data set are constructed. The collected tobacco leaf images in the related data of the tobacco leaf and the curing barn are preprocessed by adopting a median filter nonlinear smoothing algorithm and a dark channel defogging algorithm for noise reduction and color correction, and the preprocessed tobacco leaf images are saved in a specified folder in a unified format as the tobacco leaf images into a csv file to obtain the unstructured data set. Meanwhile, the temperature and humidity data and the water loss data in the collected related data of the tobacco leaf and the curing barn are feature extracted to obtain a feature set, and the feature data in the feature set are processed for missing value, abnormal value and data standardization to obtain the structured data set.

4. The method according to claim 1, wherein the model fusion based flue-cured tobacco process stage prediction method is characterized by, In step three, the tobacco prediction model of the fusion LightGBM model, GRU model and CNN model includes: The first layer base classifier is composed of a LightGBM network, a GRU network and a CNN network, and is used for preliminary identification of the tobacco processing stage; The second layer meta-classifier is composed of a LightGBM, and is used for secondary identification of the tobacco processing stage; The CNN network is an SE-ResNet50 with a residual learning module; the CNN network adopts an improved cross-entropy loss function, and selects an Adam algorithm as an adaptive learning rate optimization algorithm.

5. A model fusion-based cured tobacco process phase prediction system implementing the model fusion-based cured tobacco process phase prediction method according to any one of claims 1 to 4, characterized by The tobacco processing stage prediction system based on model fusion includes: A data set construction module is used for collecting related data of tobacco leaves and curing barns in baking according to a pre-set collection period, constructing an unstructured data set and a structured data set based on the collected related data of the tobacco leaves and the curing barn, and extracting a tobacco leaf state based on a tobacco leaf image in the unstructured data set, and digitally encoding the extracted tobacco leaf state to obtain a tobacco leaf state code; A fusion model construction module is used for constructing a tobacco prediction model of the fusion LightGBM model, GRU model and CNN model; A fusion model module is used for training the tobacco prediction model of the fusion LightGBM model, GRU model and CNN model by using the unstructured data set, structured data set and tobacco leaf state code. The data preprocessing module is used for collecting relevant data of tobacco leaves and curing barns in a curing process in real time and preprocessing the collected relevant data of tobacco leaves and curing barns. The prediction module is used for predicting a tobacco curing process stage based on the preprocessed relevant data of tobacco leaves and curing barns by using the trained tobacco prediction model fused with the LightGBM model, the GRU model and the CNN model.

6. A computer device, comprising: The computer device comprises a memory and a processor, and the memory stores a computer program which, when executed by the processor, causes the processor to execute the model fusion-based tobacco process stage prediction method according to any one of claims 1-4 as follows: Step one: collecting relevant data of tobacco leaves and curing barns in a curing process according to a preset collection period; and constructing an unstructured data set and a structured data set based on the collected relevant data of tobacco leaves and curing barns; Step two: extracting a tobacco state based on tobacco images in the unstructured data set, and digitally encoding the extracted tobacco state to obtain a tobacco state code; Step three: constructing a tobacco prediction model fused with a LightGBM model, a GRU model and a CNN model; and training the tobacco prediction model fused with the LightGBM model, the GRU model and the CNN model by using the unstructured data set, the structured data set and the tobacco state code; Step four: collecting relevant data of tobacco leaves and curing barns in a curing process in real time, preprocessing the collected relevant data of tobacco leaves and curing barns, and predicting a tobacco curing process stage based on the preprocessed relevant data of tobacco leaves and curing barns by using the trained tobacco prediction model fused with the LightGBM model, the GRU model and the CNN model. 7.A computer readable storage medium storing a computer program which, when executed by a processor, causes the processor to execute the model fusion-based tobacco process stage prediction method according to any one of claims 1-4 as follows: Step one: collecting relevant data of tobacco leaves and curing barns in a curing process according to a preset collection period; and constructing an unstructured data set and a structured data set based on the collected relevant data of tobacco leaves and curing barns; Step two: extracting a tobacco state based on tobacco images in the unstructured data set, and digitally encoding the extracted tobacco state to obtain a tobacco state code; Step three: constructing a tobacco prediction model fused with a LightGBM model, a GRU model and a CNN model; and training the tobacco prediction model fused with the LightGBM model, the GRU model and the CNN model by using the unstructured data set, the structured data set and the tobacco state code; Step four: collecting relevant data of tobacco leaves and curing barns in a curing process in real time, preprocessing the collected relevant data of tobacco leaves and curing barns, and predicting a tobacco curing process stage based on the preprocessed relevant data of tobacco leaves and curing barns by using the trained tobacco prediction model fused with the LightGBM model, the GRU model and the CNN model.

8. An information data processing terminal, characterized by The information data processing terminal is used to realize the model fusion based flue-cured tobacco process stage prediction system as claimed in claim 5.

Images