Cigar tobacco leaf process stage prediction method, system and medium based on data fusion

Through the combination of data fusion and machine learning models, the subjectivity and labor intensity problems of manual judgment during the drying process of cigar tobacco leaves are solved, and the accurate judgment and intelligent management of the drying status of cigar tobacco leaves are achieved, and the drying technology is broken. The bottleneck has promoted the high-quality development of the cigar tobacco industry.

CN114266419BActive Publication Date: 2024-11-08HUAZHONG UNIV OF SCI & TECH
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202210032426.X
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-01-12
Publication Date
2024-11-08
Estimated Expiration
2042-01-12

AI Technical Summary

Technical Problem

The prior art has strong subjectivity, large errors, high labor intensity in the process of drying cigar tobacco leaves, and low intelligence. It is impossible to accurately obtain and analyze the drying status, resulting in a bottleneck in the drying technology.

Method used

Using a data fusion-based method, data and images of the drying process are collected through preset acquisition cycles, unstructured data preprocessing and data fusion are performed, and decision tree model and support vector machine model are combined with weighted voting method for prediction. Determine the drying status in real time.

Benefits of technology

It realizes accurate judgment of the drying status of cigar tobacco leaves, reduces the subjectivity and labor intensity of manual participation, improves the intelligence of the drying process, improves the accuracy and efficiency of the judgment, and saves labor costs.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114266419B_ABST
    Figure CN114266419B_ABST
Patent Text Reader

Abstract

The present invention belongs to the technical field of cigar drying, and discloses a method, system and medium for predicting the process stage of cigar tobacco leaves based on data fusion. The prediction method includes: collecting raw data and raw images of cigar tobacco leaves in the drying process in a preset collection cycle; processing the raw data and raw images respectively, and training a decision tree model and a support vector machine model based on the processed images and data, and using a weighted voting method to fuse the prediction results; collecting raw data and raw images of cigar tobacco leaves in the drying process in real time, and using them as model input items after processing to discriminate the drying stage of cigar tobacco leaves. The present invention provides a method for predicting the process stage of cigar tobacco leaves based on data fusion, which can accurately discriminate the drying state of cigar tobacco leaves, facilitate humidity management in drying rooms, improve the drying process, and save labor costs at the same time.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention belongs to the technical field of cigar drying, and in particular relates to a method, system and medium for predicting the process stage of cigar tobacco leaves based on data fusion. Background Art

[0002] At present, the air-drying period is a critical period for the formation of tobacco leaf quality, especially appearance quality. At present, there is little research on the air-drying technology of cigar tobacco leaves in China. Many traditional tobacco areas are still using the sun-drying method (or half-drying and half-drying) to adjust cigar tobacco leaves. The modulation technical parameters and methods mainly remain on the summary of traditional experience and habits, or are kept secret as important technologies. There is a lack of mature air-drying technology and supporting air-drying facilities. During the air-drying process of cigar tobacco leaves, the state of tobacco leaves is mainly judged based on human eyes and contact. The information source is single and subjective. It is impossible to observe the air-drying state of tobacco leaves at all times during the air-drying cycle, and the labor intensity is high.

[0003] With the increasing maturity of machine vision and the continuous improvement of sensor performance, information collection equipment with different functions can be used to obtain all-round data on the cigar drying process.

[0004] Through the above analysis, the problems and defects of the existing technology are as follows: the existing technology mainly relies on manual judgment of cigar tobacco leaves during drying, which is highly subjective, has large errors and high labor intensity, is inaccurate, and has a low degree of intelligence.

[0005] The difficulty of solving the above problems and defects is:

[0006] How to reduce manual participation in the whole process of cigar tobacco leaf drying status judgment and reduce labor intensity; how to use multiple information sources to accurately obtain and analyze the cigar tobacco leaf drying status; how to further mine the collected data and improve the bottleneck of cigar drying technology in an intelligent way.

[0007] The significance of solving the above problems and defects is: through research and analysis of collected data, intelligent drying of cigars with automatic control of process and environmental parameters can be realized, the labor intensity of cigar drying can be reduced, and the bottleneck of drying technology can be broken through. It is of great significance to fill the gaps in cigar technology and raw materials and promote the high-quality development of the cigar industry. Summary of the invention

[0008] In view of the problems existing in the prior art, the present invention provides a method, system and medium for predicting the process stage of cigar tobacco leaves based on data fusion.

[0009] The present invention is implemented as follows: a method for predicting the process stage of cigar tobacco leaves based on data fusion, the method for predicting the process stage of cigar tobacco leaves based on data fusion comprising:

[0010] The original data and original images of cigar tobacco leaves during the drying process are collected in a preset collection cycle; the original data and original images are processed respectively, and the decision tree model and support vector machine model are trained based on the processed images and data, and the prediction results are fused using the weighted voting method; the original data and original images of cigar tobacco leaves during the drying process are collected in real time, and the processed data are used as model input items to distinguish the drying stage of cigar tobacco leaves.

[0011] Furthermore, the cigar tobacco leaf process stage prediction method based on data fusion includes the following steps:

[0012] Step 1: Classify the cigars to be dried into different types according to their attributes and varieties; when the cigars are dried, collect the drying room environment data and the cigar tobacco leaf data during the drying process at a certain collection cycle to obtain the original data and original images;

[0013] Step 2: performing unstructured data preprocessing on the collected original images of cigar tobacco leaves to obtain a primary processed image and a secondary processed image;

[0014] Step 3: crop the obtained secondary processed image to obtain a tertiary processed image; fuse the obtained tertiary processed image with the original data to obtain primary processed data;

[0015] Step 4, processing the first-processed data to obtain second-processed data, third-processed data, fourth-processed data, and fifth-processed data in sequence;

[0016] Step 5: Import the five processed data into the decision tree model and the support vector machine model, and fuse the prediction results using the weighted voting method based on the performance of the model;

[0017] Step six, collecting raw data and raw images of cigar tobacco leaves during the drying process in real time, and processing them as input items of the decision tree model and the support vector machine model to identify the drying stage of the cigar tobacco leaves.

[0018] Furthermore, the environmental data includes air-drying room humidity; and the cigar tobacco leaf data includes tobacco leaf air-drying images and water loss data.

[0019] Further, the method of collecting the air-drying room environment data and the cigar tobacco leaf data during the air-drying process in a certain collection cycle to obtain the original data and original image includes:

[0020] The camera is used to collect the drying images of cigar tobacco leaves as original images at a preset collection period; and each sensor is used to collect the drying water loss of cigar tobacco leaves and the humidity data of the drying room as original data at a preset collection period.

[0021] Furthermore, the method of collecting the air-drying room environment data and the cigar tobacco leaf data during the air-drying process in a certain collection cycle to obtain the original data and original image also includes:

[0022] The format of the collected original data is unified; the format of the collected original images is unified.

[0023] Furthermore, the collected original image of the cigar tobacco leaf is subjected to unstructured data preprocessing to obtain a primary processed image and a secondary processed image, including:

[0024] The non-local mean image denoising algorithm is used to determine pixels with similar neighborhood structures in the original image in units of pixels; then the weighted average of the determined areas is calculated to obtain a processed image;

[0025] The image color correction algorithm is used to perform color matching correction on the primary processed image with the help of the color histogram of the standard color correction card to obtain the secondary processed image.

[0026] Furthermore, the obtained secondary processed image is cropped to obtain a tertiary processed image; and the obtained tertiary processed image is fused with the original data to obtain primary processed data, including:

[0027] A fixed area in the center of the secondary processed image is cropped to obtain a tertiary processed image;

[0028] The OpenCV computer vision library is used to calculate the L*, a*, and b* color values ​​of each pixel in the three processed images, and the average is taken as the L*, a*, and b* color mean of the three processed images; where L* represents lightness, a* represents the component from green to red, and b* represents the component from blue to yellow;

[0029] The L*, a*, and b* features extracted from the three processed images are fused with the original data in a time-corresponding relationship to obtain the first processed data.

[0030] Further, the processing of the primary processed data to obtain secondary processed data, tertiary processed data, quaternary processed data, and quintuple processed data in sequence includes:

[0031] According to the corresponding relationship between the drying time and the drying stage in the drying process of cigar tobacco leaves, the corresponding drying stage is marked for each piece of primary processing data, which is divided into withering stage, yellowing stage, browning stage and dry tendon stage, and the secondary processing data is obtained;

[0032] When there are a small number of missing eigenvalues ​​in the secondary processing data, the mean of the adjacent eigenvalues ​​before and after the small number of missing eigenvalues ​​is used for interpolation and filling to obtain the tertiary processing data;

[0033] The standard deviation detection method is used to identify outliers in the three-times processed data, and the outliers are corrected using the functional relationship of the data before and after the outliers to obtain the four-times processed data;

[0034] The Z-Score standardization method was used to standardize the four-time processing data to obtain the five-time processing data.

[0035] Furthermore, the five processed data are imported into the decision tree model and the support vector machine model, and the prediction results are integrated using the weighted voting method according to the performance of the model, including:

[0036] Importing five processed data into the decision tree model for training and prediction includes:

[0037] The water loss of tobacco leaves during air-drying, air-drying room humidity, L*, a* and b* are used as input features, and the air-drying stage is used as the input label; the training process uses the information gain ratio to determine the input features and construct a decision tree; the obtained classification probability and classification result are taken as the first result, recorded as R1;

[0038] Importing five processed data into the support vector machine model for training and prediction includes:

[0039] The water loss of tobacco leaves during air-drying, air-drying room humidity, L*, a* and b* are used as input features, and the air-drying stage is used as the input label; the grid search method is used to automatically adjust the model parameters during the training process; the obtained classification probability and classification result are used as the second result, recorded as R2;

[0040] The models are weighted according to the accuracy ranking, the prediction accuracy of multiple models after fusion under different weights is tested, and the most appropriate weight ratio is selected.

[0041] Another object of the present invention is to provide a cigar tobacco process stage prediction system based on data fusion, comprising:

[0042] The original data and original image acquisition module is used to classify the cigars to be dried into different types according to their attributes and varieties; when the cigars are dried, the drying room environment data and the cigar tobacco leaf data during the drying process are collected at a certain collection cycle to obtain the original data and original images;

[0043] The processing image acquisition module is used to perform unstructured data preprocessing on the collected original images of cigar tobacco leaves to obtain a primary processed image and a secondary processed image;

[0044] The row data fusion module is used to crop the obtained secondary processed image to obtain a tertiary processed image; and fuse the obtained tertiary processed image with the original data to obtain primary processed data;

[0045] The multiple processing data acquisition module is used to process the primary processing data to obtain the secondary processing data, the tertiary processing data, the fourth processing data, and the fifth processing data in sequence;

[0046] The fusion prediction result module is used to import the five processed data into the decision tree model and the support vector machine model, and fuse the prediction results using the weighted voting method according to the performance of the model;

[0047] The discrimination module is used to collect raw data and raw images of cigar tobacco leaves in the drying process in real time, and after processing, use them as input items of the decision tree model and the support vector machine model to discriminate the drying stage of the cigar tobacco leaves.

[0048] Another object of the present invention is to provide a computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, the processor executes the method for predicting the process stage of cigar tobacco leaves based on data fusion.

[0049] Combining all the above technical solutions, the advantages and positive effects of the present invention are as follows:

[0050] The present invention provides a method for predicting the process stage of cigar tobacco leaves based on data fusion. The output result of the fusion prediction model can accurately determine the drying state of cigar tobacco leaves, thereby avoiding the subjectivity, error and high labor intensity of manual determination of the drying stage of cigar tobacco leaves, facilitating the humidity management of the drying room and improving the drying process.

[0051] The present invention extracts L*, a*, and b* color values ​​from the cigar tobacco leaf drying image and converts them into structured data, and fuses the structured data with the original data in a time-corresponding relationship to improve the training efficiency of the prediction model;

[0052] The present invention is based on data fusion and utilizes weight value combination to obtain fusion model prediction results, which can distinguish the state of cigar tobacco leaves in the drying stage in real time and throughout the whole process, thereby improving the accuracy and efficiency of the distinction, while reducing the workload of the drying personnel and saving labor costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 It is a schematic diagram of a method for predicting the process stages of cigar tobacco leaves based on data fusion provided in an embodiment of the present invention.

[0054] Figure 2 It is a flow chart of a method for predicting the process stages of cigar tobacco leaves based on data fusion provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0055] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0056] In view of the problems existing in the prior art, the present invention provides a method for predicting the process stage of cigar tobacco leaves based on data fusion. The present invention is described in detail below in conjunction with the accompanying drawings.

[0057] The method for predicting the process stage of cigar tobacco leaves based on data fusion provided in the embodiment of the present invention includes:

[0058] The original data and original images of cigar tobacco leaves during the drying process are collected in a preset collection cycle; the original data and original images are processed respectively, and the decision tree model and support vector machine model are trained based on the processed images and data, and the prediction results are fused using the weighted voting method; the original data and original images of cigar tobacco leaves during the drying process are collected in real time, and the processed data are used as model input items to distinguish the drying stage of cigar tobacco leaves.

[0059] like Figure 1 As shown, the method for predicting the process stage of cigar tobacco leaves based on data fusion provided in the embodiment of the present invention comprises the following steps:

[0060] S101, classifying the cigars to be aired into different types according to their attributes and varieties; when the cigars are aired, collecting airing room environment data and cigar tobacco leaf data during the airing process at a certain collection cycle to obtain original data and original images;

[0061] S102, performing unstructured data preprocessing on the collected original image of the cigar tobacco leaf to obtain a primary processed image and a secondary processed image; cropping the obtained secondary processed image to obtain a tertiary processed image; and fusing the obtained tertiary processed image with the original data to obtain primary processed data;

[0062] S103, processing the primary processed data to obtain secondary processed data, tertiary processed data, quaternary processed data, and quintuple processed data in sequence; importing the quintuple processed data into a decision tree model and a support vector machine model, and fusing the prediction results using a weighted voting method based on the performance of the model;

[0063] S104, collecting original data and original images of cigar tobacco leaves during the drying process in real time, and processing them as input items of a decision tree model and a support vector machine model to identify the drying stage of the cigar tobacco leaves.

[0064] The environmental data provided by the embodiment of the present invention includes the humidity of the air-drying room; the cigar tobacco leaf data includes tobacco leaf air-drying images and water loss data.

[0065] The embodiment of the present invention provides a method of collecting air-drying room environment data and cigar tobacco leaf data during the air-drying process in a certain collection cycle to obtain original data and original images, including:

[0066] The camera is used to collect the drying images of cigar tobacco leaves as original images at a preset collection period; and each sensor is used to collect the drying water loss of cigar tobacco leaves and the humidity data of the drying room as original data at a preset collection period.

[0067] The method provided in the embodiment of the present invention for collecting air-drying room environment data and cigar tobacco leaf data during the air-drying process in a certain collection cycle to obtain original data and original images also includes:

[0068] The format of the collected original data is unified; the format of the collected original images is unified.

[0069] The method provided in the embodiment of the present invention performs unstructured data preprocessing on the collected original image of cigar tobacco leaves to obtain a primary processed image and a secondary processed image, including:

[0070] The non-local mean image denoising algorithm is used to determine pixels with similar neighborhood structures in the original image in units of pixels; then the weighted average of the determined areas is calculated to obtain a processed image;

[0071] The image color correction algorithm is used to perform color matching correction on the primary processed image with the help of the color histogram of the standard color correction card to obtain the secondary processed image.

[0072] The embodiment of the present invention provides that the obtained secondary processed image is cropped to obtain a tertiary processed image; and the obtained tertiary processed image is data-fused with the original data to obtain primary processed data, including:

[0073] A fixed area in the center of the secondary processed image is cropped to obtain a tertiary processed image;

[0074] The OpenCV computer vision library is used to calculate the L*, a*, and b* color values ​​of each pixel in the three processed images, and the average is taken as the L*, a*, and b* color mean of the three processed images; where L* represents lightness, a* represents the component from green to red, and b* represents the component from blue to yellow;

[0075] The L*, a*, and b* features extracted from the three processed images are fused with the original data in a time-corresponding relationship to obtain the first processed data.

[0076] The embodiment of the present invention provides processing the primary processed data to obtain the secondary processed data, the tertiary processed data, the quadratic processed data, and the quintuple processed data in sequence, including:

[0077] According to the corresponding relationship between the drying time and the drying stage in the drying process of cigar tobacco leaves, the corresponding drying stage is marked for each piece of primary processing data, which is divided into withering stage, yellowing stage, browning stage and dry tendon stage, and the secondary processing data is obtained;

[0078] When there are a small number of missing eigenvalues ​​in the secondary processing data, the mean of the adjacent eigenvalues ​​before and after the small number of missing eigenvalues ​​is used for interpolation and filling to obtain the tertiary processing data;

[0079] The standard deviation detection method is used to identify outliers in the three-times processed data, and the outliers are corrected using the functional relationship of the data before and after the outliers to obtain the four-times processed data;

[0080] The Z-Score standardization method was used to standardize the four-time processing data to obtain the five-time processing data.

[0081] The embodiment of the present invention provides the method of importing five processed data into a decision tree model and a support vector machine model, and fusing the prediction results using a weighted voting method according to the performance of the model, including:

[0082] Importing five processed data into the decision tree model for training and prediction includes:

[0083] The water loss of tobacco leaves during air-drying, air-drying room humidity, L*, a* and b* are used as input features, and the air-drying stage is used as the input label; the training process uses the information gain ratio to determine the input features and construct a decision tree; the obtained classification probability and classification result are taken as the first result, recorded as R1;

[0084] Importing five processed data into the support vector machine model for training and prediction includes:

[0085] The water loss of tobacco leaves during air-drying, air-drying room humidity, L*, a* and b* are used as input features, and the air-drying stage is used as the input label; the grid search method is used to automatically adjust the model parameters during the training process; the obtained classification probability and classification result are used as the second result, recorded as R2;

[0086] The models are weighted according to the accuracy ranking, the prediction accuracy of multiple models after fusion under different weights is tested, and the most appropriate weight ratio is selected.

[0087] The technical solution of the present invention is further described below in conjunction with specific embodiments.

[0088] Embodiment 1:

[0089] A cigar tobacco leaf process stage prediction method based on data fusion is mainly applied to the cigar tobacco leaf air-drying process. The specific implementation plan is as follows:

[0090] S1: Before cigars are air-dried, they are divided into different types according to their properties and varieties. During the cigar drying process, the drying room environment data and the cigar tobacco leaf data during the drying process are collected at a certain collection cycle to form original data and original images. The environmental data includes the drying room humidity; the cigar tobacco leaf data includes the tobacco leaf drying image and water loss data.

[0091] Specifically, during the drying process of cigars, the camera collects drying images of cigar leaves at a preset collection cycle; each sensor collects the drying water loss of cigar leaves and the humidity data of the drying room at the same collection cycle; and then uploads them to the cloud platform database and the corresponding image folder through the network to form the original data and original images for model training. Among them, the preset collection cycle is such as 30 minutes, 40 minutes, 50 minutes, 60 minutes or 90 minutes, etc. The values ​​listed here are only for example and are not specifically limited.

[0092] S2: Perform unstructured data preprocessing on the collected original images of cigar tobacco leaves to obtain secondary processed images.

[0093] Specifically, the original image of the cigar tobacco leaves collected in step S1 is obtained, and the non-local mean image denoising algorithm is used to search for pixels with similar neighborhood structures in the original image in units of pixels, and then the weighted average of these areas is calculated to remove the Gaussian noise in the original image to a large extent, and obtain a processed image; the image color correction algorithm is used to perform color matching on the first processed image with the help of the color histogram of the standard color correction card to perform color correction, reduce the color cast problem of the first processed image caused by the environment, and obtain a secondary processed image. For example, in the non-local mean image denoising algorithm, the search area can be 21×21, and the similarity comparison block can be selected as 7×7. The larger the standard deviation of the Gaussian noise, the larger the search area and the comparison block area. The image color correction algorithm first loads the reference image, that is, the color correction card, and the image to be corrected, and obtains the color histogram of the color matching card in the reference image through the program, and applies it to the image to be corrected.

[0094] S3: Fusing the secondary processed image with the original data to obtain primary processed data.

[0095] Specifically, a fixed area in the center of the secondary processed image is cropped to obtain a tertiary processed image; the OpenCV computer vision library is used to calculate the L*, a*, and b* color values ​​of each pixel in the tertiary processed image, and the average is taken as the L*, a*, and b* color average of the tertiary processed image; where L* represents brightness, a* represents the component from green to red, and b* represents the component from blue to yellow; the L*, a*, and b* features extracted from the tertiary processed images are fused with the original data in a time-corresponding relationship to obtain primary processed data.

[0096] For example, a fixed area of ​​1400 pixels × 650 pixels in the center of the secondary image is cropped, and the L*, a*, and b* color values ​​of each pixel in the area are calculated in turn, and their average is calculated as the L*, a*, and b* color value average of the image. In addition, the original image information and the original data information correspond to each other in time, so the extracted L*, a*, and b* features can be fused with the original data in time to form a processed data.

[0097] S4: The five processed data are imported into the decision tree model and the support vector machine model respectively, and the prediction results are integrated using the weighted voting method according to the performance of the model;

[0098] Specifically, the fused data after data preprocessing, namely L*, a*, b* color values, air-drying room humidity and cigar tobacco leaf water loss, are used as input feature parameters, and the digital code of the air-drying stage is used as the input label to import the decision tree model and the support vector machine model for training and tuning. After obtaining the prediction accuracy of the two models, they are compared, and a higher weight is given to the model with a higher accuracy. Finally, the prediction result is obtained based on the prediction probability and weight of the two models.

[0099] In one of the embodiments, S4 specifically includes: S401: according to the correspondence between the drying time and the drying stage in the drying process of cigar tobacco leaves, the corresponding drying stage is marked for each piece of primary processing data, which is divided into withering stage, yellowing stage, browning stage and dry tendon stage, and converted into digital quantities, such as 0, 1, 2, 3, to obtain secondary processing data; S402: when there are a small number of missing eigenvalues ​​in the secondary processing data, the mean of the adjacent eigenvalues ​​before and after the small number of missing eigenvalues ​​is used for interpolation and filling to obtain tertiary processing data; S403: using the standard deviation detection method to identify outliers in the tertiary processing data, and using the functional relationship between the data before and after the outliers to correct the outliers, to obtain the fourth processing data; S404: using the Z-Score standardization method to perform data standardization on the fourth processing data to obtain the fifth processing data; S405: importing the five processing data into the decision tree model for training and prediction. The water loss of tobacco leaves during airing, airing room humidity, L*, a* and b* are used as input features, and the airing stage is used as input label; the training process uses information gain ratio to determine the input features and construct a decision tree; the obtained classification probability and classification result are used as the first result, recorded as R1; S406: import the five processed data into the support vector machine model for training and prediction. The water loss of tobacco leaves during airing, airing room humidity, L*, a* and b* are used as input features, and the airing stage is used as input label; the training process uses grid search method to automatically adjust model parameters; the obtained classification probability and classification result are used as the second result, recorded as R2; S407: divide the model weights according to the accuracy ranking, test the prediction accuracy of multi-model fusion under different weights, and select the most appropriate weight ratio.

[0100] The effects of the present invention are further described below in conjunction with the prediction results.

[0101] Input ten identical new data to the decision tree model, and directly record the prediction results of the model. When the prediction result is greater than 50%, it is considered to be correct. Similarly, input ten identical new data to the support vector machine model, and directly record the prediction results of the model. The assumptions are as follows:

[0102]

[0103] It can be found that the prediction accuracy of the decision tree model is 80%, and the prediction accuracy of the support vector machine model is 80%. The prediction results of the two models are weighted averaged. The weight value of the decision tree model is set to 0.7 and the weight value of the support vector machine model is set to 0.3. The prediction result of the fusion model obtained by weighting is:

[0104]

[0105] Its accuracy is still 80%, but if the weight value of the decision tree model is 0.6 and the weight value of the support vector machine model is 0.4, the prediction result of the weighted fusion model is:

[0106]

[0107] At this time, the accuracy of the fusion model is 90%, which is better than the accuracy of the single model. The weights listed here are only for illustration and are not specifically limited.

[0108] In the description of the present invention, unless otherwise specified, "plurality" means two or more than two; the orientations or positional relationships indicated by the terms "upper", "lower", "left", "right", "inner", "outer", "front end", "rear end", "head", "tail", etc. are based on the orientations or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operate in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first", "second", "third", etc. are only used for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0109] It should be noted that the embodiments of the present invention can be implemented by hardware, software, or a combination of software and hardware. The hardware part can be implemented using dedicated logic; the software part can be stored in a memory and executed by an appropriate instruction execution system, such as a microprocessor or dedicated design hardware. It can be understood by a person of ordinary skill in the art that the above-mentioned devices and methods can be implemented using computer executable instructions and / or contained in a processor control code, such as a carrier medium such as a disk, CD or DVD-ROM, a programmable memory such as a read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. Such code is provided on the carrier medium. The device and its modules of the present invention can be implemented by hardware circuits such as very large-scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, etc., or programmable hardware devices such as field programmable gate arrays, programmable logic devices, etc., can also be implemented by software executed by various types of processors, and can also be implemented by a combination of the above-mentioned hardware circuits and software, such as firmware.

[0110] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any modifications, equivalent substitutions and improvements made by any technician familiar with the technical field within the technical scope disclosed by the present invention and within the spirit and principle of the present invention should be covered by the protection scope of the present invention.

Claims

1. A method for predicting the process stage of cigar tobacco leaves based on data fusion, characterized in that: The cigar tobacco leaf process stage prediction method based on data fusion includes: The original data and original images of cigar tobacco leaves during the drying process are collected in a preset collection cycle; the original data and original images are processed respectively, and a decision tree model and a support vector machine model are trained based on the processed images and data, and the prediction results are fused using a weighted voting method; the original data and original images of cigar tobacco leaves during the drying process are collected in real time, and the processed data and original images of cigar tobacco leaves are used as model input items to distinguish the drying stage of cigar tobacco leaves; The cigar tobacco leaf process stage prediction method based on data fusion comprises the following steps: Step 1: Classify the cigars to be dried into different types according to their attributes and varieties; when the cigars are dried, collect the drying room environment data and the cigar tobacco leaf data during the drying process at a certain collection cycle to obtain the original data and original images; Step 2: performing unstructured data preprocessing on the collected original images of cigar tobacco leaves to obtain a primary processed image and a secondary processed image; Step 3: crop the obtained secondary processed image to obtain a tertiary processed image; fuse the obtained tertiary processed image with the original data to obtain primary processed data; Step 4, processing the first-processed data to obtain second-processed data, third-processed data, fourth-processed data, and fifth-processed data in sequence; Step 5: Import the five processed data into the decision tree model and the support vector machine model, and fuse the prediction results using the weighted voting method based on the performance of the model; Step 6: Real-time collection of raw data and raw images of cigar tobacco leaves during the drying process, and processing them as input items of a decision tree model and a support vector machine model to identify the drying stage of the cigar tobacco leaves; The unstructured data preprocessing is performed on the collected original image of the cigar tobacco leaf to obtain the primary processed image and the secondary processed image, including: The non-local mean image denoising algorithm is used to determine pixels with similar neighborhood structures in the original image in units of pixels; then the weighted average of the determined areas is calculated to obtain a processed image; Using the image color correction algorithm, the color matching correction is performed on the primary processed image with the help of the color histogram of the standard color correction card to obtain the secondary processed image; The obtained secondary processed image is cropped to obtain a tertiary processed image; and the obtained tertiary processed image is fused with the original data to obtain primary processed data, including: A fixed area in the center of the secondary processed image is cropped to obtain a tertiary processed image; The OpenCV computer vision library is used to calculate the L*, a*, and b* color values ​​of each pixel in the three processed images, and the average is taken as the L*, a*, and b* color mean of the three processed images; where L* represents lightness, a* represents the component from green to red, and b* represents the component from blue to yellow; The L*, a*, and b* features extracted from the three processed images are fused with the original data in a time-corresponding relationship to obtain the first processed data; The processing of the primary processed data to obtain secondary processed data, tertiary processed data, quaternary processed data, and quintuple processed data in sequence includes: According to the corresponding relationship between the drying time and the drying stage in the drying process of cigar tobacco leaves, the corresponding drying stage is marked for each piece of primary processing data, which is divided into withering stage, yellowing stage, browning stage and dry tendon stage, and the secondary processing data is obtained; When there are a small number of missing eigenvalues ​​in the secondary processing data, the mean of the adjacent eigenvalues ​​before and after the small number of missing eigenvalues ​​is used for interpolation and filling to obtain the tertiary processing data; The standard deviation detection method is used to identify outliers in the three-times processed data, and the outliers are corrected using the functional relationship of the data before and after the outliers to obtain the four-times processed data; The data of the four treatments were standardized using the Z-Score standardization method to obtain the data of the five treatments; The five processed data are imported into the decision tree model and the support vector machine model, and the prediction results are integrated using the weighted voting method according to the performance of the model, including: Importing five processed data into the decision tree model for training and prediction includes: The water loss of tobacco leaves during air-drying, air-drying room humidity, L*, a* and b* are used as input features, and the air-drying stage is used as the input label; the information gain ratio is used in the training process to determine the input features and construct a decision tree; the obtained classification probability and classification result are taken as the first result, recorded as R1; Importing five processed data into the support vector machine model for training and prediction includes: The water loss of tobacco leaves during air-drying, air-drying room humidity, L*, a* and b* are used as input features, and the air-drying stage is used as the input label; the grid search method is used to automatically adjust the model parameters during the training process; the obtained classification probability and classification result are used as the second result, recorded as R2; The models are weighted according to the accuracy ranking, the prediction accuracy of multiple models after fusion under different weights is tested, and the most appropriate weight ratio is selected.

2. The method for predicting the process stage of cigar tobacco leaves based on data fusion as claimed in claim 1, characterized in that: The environmental data includes the humidity of the drying room; the cigar tobacco leaf data includes tobacco leaf drying images and water loss data.

3. The method for predicting the process stage of cigar tobacco leaves based on data fusion as claimed in claim 1, characterized in that: The method of collecting the air-drying room environment data and the cigar tobacco leaf data during the air-drying process in a certain collection cycle to obtain the original data and original image includes: The camera is used to collect the drying images of cigar tobacco leaves as original images at a preset collection cycle; and each sensor is used to collect the drying water loss of cigar tobacco leaves and the humidity data of the drying room as original data at a preset collection cycle.

4. The method for predicting the process stage of cigar tobacco leaves based on data fusion as claimed in claim 1, characterized in that: The method of collecting the air-drying room environment data and the cigar tobacco leaf data during the air-drying process in a certain collection cycle to obtain the original data and original image also includes: The format of the collected original data is unified; the format of the collected original images is unified.

5. A cigar tobacco leaf process stage prediction system based on data fusion implementing the cigar tobacco leaf process stage prediction method based on data fusion as claimed in any one of claims 1 to 4, characterized in that: The cigar tobacco leaf process stage prediction system based on data fusion includes: The original data and original image acquisition module is used to classify the cigars to be dried into different types according to their attributes and varieties; when the cigars are dried, the drying room environment data and the cigar tobacco leaf data during the drying process are collected at a certain collection cycle to obtain the original data and original images; The processing image acquisition module is used to perform unstructured data preprocessing on the collected original images of cigar tobacco leaves to obtain a primary processed image and a secondary processed image; The row data fusion module is used to crop the obtained secondary processed image to obtain a tertiary processed image; and fuse the obtained tertiary processed image with the original data to obtain primary processed data; The multiple processing data acquisition module is used to process the primary processing data to obtain the secondary processing data, the tertiary processing data, the fourth processing data, and the fifth processing data in sequence; The fusion prediction result module is used to import the five processed data into the decision tree model and the support vector machine model, and fuse the prediction results using the weighted voting method according to the performance of the model; The discrimination module is used to collect raw data and raw images of cigar tobacco leaves in the drying process in real time, and after processing, use them as input items of the decision tree model and the support vector machine model to discriminate the drying stage of the cigar tobacco leaves.

6. A computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, the processor executes the method for predicting the process stage of cigar tobacco leaves based on data fusion according to any one of claims 1 to 4.