Fusion method for identifying tobacco leaf curing stage
By building a tobacco leaf baking stage identification model on a cloud server, combining tobacco leaf weight changes and environmental information, and combining feature sub-models for identification, the calculation efficiency problem of the existing technology on the cloud server is solved, and rapid and accurate tobacco leaf baking is achieved Phase identification.
Patent Information
- Application Number
- CN202411382136.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-30
- Publication Date
- 2024-12-06
- Estimated Expiration
- 2044-09-30
AI Technical Summary
The existing tobacco leaf baking stage identification technology is difficult to perform fast and accurate calculations on cloud servers, fails to effectively simplify the model to adapt to the characteristics of cloud servers, and does not comprehensively consider image data processing and model running time.
A fusion method is adopted to obtain the weight change data of tobacco leaves and the environment information of the baking room, and to identify the tobacco leaves in a baking phase, and to reduce unnecessary feature extraction and modeling. Enter to adapt to the computing resource limitations of cloud servers.
It realizes the efficient identification of tobacco leaves on cloud servers, ensures the recognition accuracy, and reduces the consumption of computing resources, which is suitable for real-time data processing.
Smart Images

Figure CN119073641A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to a fusion method for identifying tobacco leaf baking stages, and belongs to the technical field of tobacco leaf curing. Background Art
[0002] In intelligent tobacco baking, sensors collect data from the baking room every 5 minutes. The collected data is processed and input into the prediction model to obtain the current stage of tobacco baking. When running in real time, the server is usually deployed in the cloud; if the scale is large, the cloud server may have to receive data from dozens of kangs within five minutes, and calculate all the data within five minutes, so as not to affect the next data collection. In addition, the cloud server cannot be too expensive, and renting a cloud server with a CPU is a suitable choice.
[0003] However, the current tobacco leaf baking stage recognition technology is either based on laboratory conditions without analyzing the actual situation; or it is based on the accuracy of tobacco leaf recognition without simplifying the model based on the characteristics of deployment on cloud servers; or it does not comprehensively consider the time for image data processing and model running time. This makes it difficult to deploy tobacco leaf baking stage recognition technology on cloud servers for fast and accurate calculations. Summary of the invention
[0004] Based on the above, the present invention provides a fusion method for identifying tobacco leaf baking stages suitable for deployment on a cloud server to overcome the shortcomings of the prior art.
[0005] The technical solution of the present invention is: a fusion method for identifying tobacco leaf baking stages, comprising: S1, obtaining first information of cured tobacco leaves, wherein the first information includes tobacco leaf weight change data and curing room environment information; S2, inputting the first information into a baking stage recognition model to preliminarily determine the current baking stage of the tobacco leaves, wherein the baking stage recognition model is obtained by integrating a tobacco leaf weight feature sub-model and a tobacco leaf environment feature sub-model; S3, selecting a tobacco leaf stage state recognition model according to the baking stage, wherein the tobacco leaf stage state recognition model is obtained by fusing a baking stage change feature sub-model with a tobacco leaf weight feature sub-model and a tobacco leaf environment feature sub-model, wherein the baking stage change feature sub-model is trained by feature data that changes significantly in the current baking stage; S4, obtaining second information of the cured tobacco leaves, where the second information is change characteristic data in a change characteristic sub-model of the curing stage; S5, inputting the second information and the first information into the tobacco leaf stage state recognition model to obtain the tobacco leaf baking stage.
[0006] Preferably, the method for constructing the baking stage identification model is: Obtain tobacco leaf weight change data corresponding to the baking stage and baking room environment information, wherein the baking room environment information includes temperature and humidity information inside the baking room and current time information; Two initialized LightGBM models were trained using tobacco leaf weight change data and flue-curing room environment information to obtain tobacco leaf weight feature sub-model and tobacco leaf environment feature sub-model. The tobacco leaf weight feature sub-model and the tobacco leaf environment feature sub-model are fused to obtain the baking stage recognition model.
[0007] Preferably, the baking stage change characteristic submodel includes a color characteristic submodel, a texture characteristic submodel and a length characteristic submodel; in stages 1, 2, 3 and 4 of tobacco leaf baking, the color characteristic submodel is selected as the baking stage change characteristic submodel; in stages 5, 6 and 7 of tobacco leaf baking, the texture characteristic submodel is selected as the baking stage change characteristic submodel; in stages 8 and 9 of tobacco leaf baking, the length characteristic submodel is selected as the baking stage change characteristic submodel; in stage 10 of tobacco leaf baking, the color characteristic submodel, the texture characteristic submodel and the length characteristic submodel are selected as the baking stage change characteristic submodel.
[0008] Preferably, the method for constructing the color feature sub-model is: Obtain tobacco leaf images corresponding to tobacco leaf baking stages; Preprocessing the tobacco leaf image; Separating the tobacco leaf image from the background; Extracting color features from the tobacco leaf image, the color features including RGB, HSV, HIS, Lab, YCbCr, and third-order color moments, a total of 21 dimensions of features; The above tobacco leaf baking stage and color feature data are used to train the initialized LightGBM model to obtain the color feature sub-model.
[0009] Preferably, the method for constructing the texture feature sub-model is: Obtain tobacco leaf images corresponding to tobacco leaf baking stages; Preprocessing the tobacco leaf image; Separating the tobacco leaf image from the background; Extracting texture features from the tobacco leaf image, extracting energy (ENE) features, contrast (CON) features, correlation (CORR) features, and inverse moment (HOM) features with pixel distances of 3 and 6 and calculation angles of 0°, 45°, 90°, and 135°, for a total of 48-dimensional features; The above tobacco leaf baking stage and texture feature data are used to train the initialized LightGBM model to obtain the texture feature sub-model.
[0010] Preferably, the length characteristic sub-model is constructed by: Obtain tobacco leaf images corresponding to tobacco leaf baking stages; Preprocessing the tobacco leaf image; Separating the tobacco leaf image from the background; Extracting the length information in the tobacco leaf image to extract the average length of a pack of tobacco leaves; The above tobacco leaf baking stage and length information data are used to train the initialized LightGBM model to obtain the length feature sub-model.
[0011] Beneficial effects of the present invention: The present invention first preliminarily determines the tobacco leaf baking stage by real-time collection and analysis of tobacco leaf weight change data and baking room environment information, then selects the feature sub-model corresponding to the baking stage and fuses it with the baking stage recognition model, and obtains the feature information corresponding to the corresponding stage and the initial weight change data and baking room environment information as input, thereby obtaining an accurate tobacco leaf baking stage judgment result. On the premise of ensuring the recognition accuracy, the present invention reduces the extraction of features with unclear stage feature changes, reduces the input of the model, and reduces the participation of the model, thereby speeding up the recognition of the current stage of the tobacco leaf. It is particularly adapted to the characteristics of cloud servers and can perform efficient data processing and analysis with limited computing resources. BRIEF DESCRIPTION OF THE DRAWINGS
[0012] Figure 1 It is a schematic diagram of the fusion method of the present invention; In the figure: sub-model 1 is the tobacco leaf environment feature sub-model, sub-model 2 is the tobacco leaf weight feature sub-model, sub-model 3 is the color feature sub-model, sub-model 4 is the texture feature sub-model, and sub-model 5 is the length feature sub-model. DETAILED DESCRIPTION
[0013] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below in conjunction with the accompanying drawings. In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without violating the connotation of the present invention, so the present invention is not limited by the specific implementation disclosed below.
[0014] See also Figure 1 In this embodiment, a fusion method for identifying tobacco leaf baking stages comprises the following steps: S1, obtaining first information of cured tobacco leaves, wherein the first information includes tobacco leaf weight change data and curing room environment information; Through the flue-cured tobacco data collection system, the weight of a clip of tobacco leaves can be obtained through the weight sensor; the flue-curing room environmental information includes the temperature and humidity information inside the flue-curing room and the current time information. The temperature and humidity conditions inside the flue-curing room can be obtained through the dry-bulb and wet-bulb sensors, and the real-time baking time of the tobacco leaves can be obtained through calculation.
[0015] S2, inputting the first information into a baking stage recognition model to preliminarily determine the current baking stage of the tobacco leaves, wherein the baking stage recognition model is obtained by integrating a tobacco leaf weight feature sub-model and a tobacco leaf environment feature sub-model; Specifically, the construction method of the baking stage recognition model is as follows: S21, obtaining tobacco leaf weight change data corresponding to the baking stage and curing room environment information, wherein the curing room environment information includes temperature and humidity information inside the curing room and current time information; S22, using tobacco leaf weight change data and flue-curing room environment information to train two initialized LightGBM models respectively, to obtain tobacco leaf weight feature sub-model and tobacco leaf environment feature sub-model; S23, fusing the tobacco leaf weight feature sub-model and the tobacco leaf environment feature sub-model to obtain a baking stage recognition model.
[0016] By inputting tobacco leaf weight change data and curing room environment information (temperature, humidity and time information) into the curing stage identification model, the curing stage can be preliminarily obtained.
[0017] In the present invention, the tobacco leaf baking stage is divided into 10 stages, with reference to a baking process of ten key stable temperature points for honey sweet flavor tobacco leaves (authorization announcement number CN109998144B). The first stage is the pre-yellowing key point (35 / 34°C), the second stage is the main yellowing key point (38 / 36-37°C), the third stage is the advance dehumidification key point (40 / 36-37°C), the fourth stage is the full yellowing key point (42 / 36-37°C), the fifth stage is the softening and withering key point (45 / 36-37°C), the sixth stage is the semi-dry tobacco leaf key point (48 / 37-38°C), the seventh stage is the main vein whitening key point (51 / 38°C), the eighth stage is the aroma formation key point (54 / 39°C), the ninth stage is the color balance key point (60 / 40°C), and the tenth stage is the tobacco leaf drying key point (68 / 41°C).
[0018] S3, selecting a tobacco leaf stage state recognition model according to the baking stage, wherein the tobacco leaf stage state recognition model is obtained by fusing a baking stage change feature sub-model with a tobacco leaf weight feature sub-model and a tobacco leaf environment feature sub-model, wherein the baking stage change feature sub-model is trained by feature data that changes significantly in the current baking stage; Specifically, according to the baking stage, the feature model with the most obvious changes in the current stage is selected for fusion. The baking stage change feature submodel includes color feature submodel, texture feature submodel and length feature submodel; in the 1st, 2nd, 3rd and 4th stages of tobacco leaf baking, the color feature submodel is selected as the baking stage change feature submodel; in the 5th, 6th and 7th stages of tobacco leaf baking, the texture feature submodel is selected as the baking stage change feature submodel; in the 8th and 9th stages of tobacco leaf baking, the length feature submodel is selected as the baking stage change feature submodel; in the 10th stage of tobacco leaf baking, the color feature submodel, texture feature submodel and length feature submodel are selected as the baking stage change feature submodel.
[0019] In the study, after collecting tobacco leaf images of the entire tobacco leaf baking process, the tobacco leaf images were first preprocessed, and then the color, texture and length information features of the tobacco leaves were obtained by separating the tobacco leaf images from the background and extracting features. The feature dimension reduction was then achieved through feature engineering, and the mutual information method was used to calculate the size of the correlation between the features and the labels, and the correlation was compared to select some features with a greater correlation with the labels. At the same time, the Pearson correlation coefficient was used to calculate the correlation between the features, and the features with less correlation between the features were selected, so as to reduce the extraction of features and ultimately reduce the input of the model to reduce the running time of the model.
[0020] After extracting the structured data of tobacco leaves, we obtained the color, texture and length information of the tobacco leaves. By comparing these information, we found that the characteristic changes of tobacco leaf color were more obvious in stages 1, 2, 3, and 4, the characteristic changes of tobacco leaf texture were more obvious in stages 5, 6, and 7, and the length information of tobacco leaves changed more obviously in stages 8 and 9.
[0021] The construction method of the color feature sub-model is: Obtain tobacco leaf images corresponding to tobacco leaf baking stages (1, 2, 3, 4); Preprocessing the tobacco leaf image includes denoising the image using a bilateral filter to remove the pepper and salt noise in the dark area of the original tobacco image, which is caused by the interference of random signals during the acquisition and transmission process of the image due to the high temperature and high humidity environment of the tobacco leaf curing room; Separating the tobacco leaf from the background of the tobacco leaf image by using a Grab-cut algorithm; Extracting color features from the tobacco leaf image, the color features including RGB, HSV, HIS, Lab, YCbCr, and third-order color moments, a total of 21 dimensions of features; The above tobacco leaf baking stages (1, 2, 3, 4) and color feature data are used to train the initialized LightGBM model to obtain the color feature sub-model.
[0022] The construction method of the texture feature sub-model is: Obtain tobacco leaf images corresponding to tobacco leaf baking stages (5, 6, 7); Preprocessing the tobacco leaf image includes denoising the image using a bilateral filter to remove the pepper and salt noise in the dark area of the original tobacco image, which is caused by the interference of random signals during the acquisition and transmission process of the image due to the high temperature and high humidity environment of the tobacco leaf curing room; Separating the tobacco leaf from the background of the tobacco leaf image by using a Grab-cut algorithm; Extracting texture features from the tobacco leaf image, extracting energy (ENE) features, contrast (CON) features, correlation (CORR) features, and inverse moment (HOM) features with pixel distances of 3 and 6 and calculation angles of 0°, 45°, 90°, and 135°, for a total of 48-dimensional features; The above tobacco leaf baking stage and texture feature data are used to train the initialized LightGBM model to obtain the texture feature sub-model.
[0023] The construction method of the length characteristic sub-model is: Obtain tobacco leaf images corresponding to tobacco leaf baking stages (8, 9); Preprocessing the tobacco leaf image includes denoising the image using a bilateral filter to remove the pepper and salt noise in the dark area of the original tobacco image, which is caused by the interference of random signals during the acquisition and transmission process of the image due to the high temperature and high humidity environment of the tobacco leaf curing room; Separating the tobacco leaf from the background of the tobacco leaf image by using a Grab-cut algorithm; Extracting the length information in the tobacco leaf image to extract the average length of a pack of tobacco leaves; The above tobacco leaf baking stage and length information data are used to train the initialized LightGBM model to obtain the length feature sub-model.
[0024] S4, obtaining second information of the cured tobacco leaves, where the second information is change characteristic data in a change characteristic sub-model of the curing stage; Specifically, when in stages 1, 2, 3, and 4, the color characteristics of the tobacco leaves are obtained; when in stages 5, 6, and 7, the texture characteristics of the tobacco leaves are obtained; when in stages 8 and 9, the length characteristics of the tobacco leaves are obtained; when in stage 10, the color characteristics, texture characteristics, and length characteristics are simultaneously obtained as the second information.
[0025] S5, inputting the second information and the first information into the tobacco leaf stage state recognition model to obtain the tobacco leaf baking stage.
[0026] When the prediction results are at stages 1, 2, 3, and 4, the color feature sub-model is added and fused with the weight and environmental feature sub-model to obtain the final prediction result. Similarly, when the prediction results are at stages 5, 6, and 7, the texture feature sub-model is added and fused with the weight and environmental feature sub-model to obtain the final prediction result. When the prediction results are at stages 8 and 9, the length feature sub-model is added and fused with the weight and environmental feature sub-model to obtain the final prediction result. At stage 10, all extracted features have not changed significantly, and all feature sub-models are fused to classify and identify the stage of tobacco baking.
[0027] In the present invention, the basic models are all LightGBM, and the clustering model SVM is selected for fusion. LightGBM is a gradient boosting decision tree algorithm (GBDT) with gradient-based one-sided sampling (GOSS) and exclusive feature bundling (EFB). It can solve the problem of slow speed of traditional GBDT in massive data training. It is particularly suitable for real-time tobacco baking state recognition and can more accurately and quickly identify the baking state of tobacco leaves. Model fusion is decision-level fusion, using stacking to fuse each sub-model. The present invention uses clustering fusion to fuse each sub-model. Compared with the traditional fusion method based on fixed weights, this method has stronger interpretability and higher accuracy.
[0028] The above-mentioned embodiments only express several implementation methods of the present invention, and the description thereof is relatively specific and detailed, but it cannot be understood as limiting the scope of the patent of the present invention. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.
Claims
1. A fusion method for identifying tobacco leaf curing stages, characterized in that: include: S1, obtaining first information of cured tobacco leaves, wherein the first information includes tobacco leaf weight change data and curing room environment information; S2, inputting the first information into a baking stage recognition model to preliminarily determine the current baking stage of the tobacco leaves, wherein the baking stage recognition model is obtained by integrating a tobacco leaf weight feature sub-model and a tobacco leaf environment feature sub-model; S3, selecting a tobacco leaf stage state recognition model according to the baking stage, wherein the tobacco leaf stage state recognition model is obtained by fusing a baking stage change feature sub-model with a tobacco leaf weight feature sub-model and a tobacco leaf environment feature sub-model, wherein the baking stage change feature sub-model is trained by feature data that changes significantly in the current baking stage; S4, obtaining second information of the cured tobacco leaves, where the second information is change characteristic data in a change characteristic sub-model of the curing stage; S5, inputting the second information and the first information into the tobacco leaf stage state recognition model to obtain the tobacco leaf baking stage.
2. The fusion method for identifying tobacco leaf curing stages according to claim 1, characterized in that: The method for constructing the baking stage recognition model is as follows: Obtain tobacco leaf weight change data corresponding to the baking stage and baking room environment information, wherein the baking room environment information includes temperature and humidity information inside the baking room and current time information; Two initialized LightGBM models were trained using tobacco leaf weight change data and flue-curing room environment information to obtain tobacco leaf weight feature sub-model and tobacco leaf environment feature sub-model. The tobacco leaf weight feature sub-model and the tobacco leaf environment feature sub-model are fused to obtain the baking stage recognition model.
3. The fusion method for identifying tobacco leaf curing stages according to claim 1, characterized in that: The baking stage change feature submodel includes a color feature submodel, a texture feature submodel and a length feature submodel; in stages 1, 2, 3 and 4 of tobacco leaf baking, the color feature submodel is selected as the baking stage change feature submodel; in stages 5, 6 and 7 of tobacco leaf baking, the texture feature submodel is selected as the baking stage change feature submodel; in stages 8 and 9 of tobacco leaf baking, the length feature submodel is selected as the baking stage change feature submodel; in stage 10 of tobacco leaf baking, the color feature submodel, the texture feature submodel and the length feature submodel are selected as the baking stage change feature submodel.
4. The fusion method for identifying tobacco leaf curing stages according to claim 1, characterized in that: The method for constructing the color feature sub-model is as follows: Obtain tobacco leaf images corresponding to tobacco leaf baking stages; Preprocessing the tobacco leaf image; Separating the tobacco leaf image from the background; Extracting color features from the tobacco leaf image, the color features including RGB, HSV, HIS, Lab, YCbCr, and third-order color moments, a total of 21 dimensions of features; The above tobacco leaf baking stage and color feature data are used to train the initialized LightGBM model to obtain the color feature sub-model.
5. The fusion method for identifying tobacco leaf curing stages according to claim 1, characterized in that: The method for constructing the texture feature sub-model is as follows: Obtain tobacco leaf images corresponding to tobacco leaf baking stages; Preprocessing the tobacco leaf image; Separating the tobacco leaf image from the background; Extracting texture features from the tobacco leaf image, extracting energy (ENE) features, contrast (CON) features, correlation (CORR) features, and inverse moment (HOM) features with pixel distances of 3 and 6 and calculation angles of 0°, 45°, 90°, and 135°, for a total of 48-dimensional features; The above tobacco leaf baking stage and texture feature data are used to train the initialized LightGBM model to obtain the texture feature sub-model.
6. The fusion method for identifying tobacco leaf curing stages according to claim 1, characterized in that: The construction method of the length characteristic sub-model is: Obtain tobacco leaf images corresponding to tobacco leaf baking stages; Preprocessing the tobacco leaf image; Separating the tobacco leaf image from the background; Extracting the length information in the tobacco leaf image to extract the average length of a pack of tobacco leaves; The above tobacco leaf baking stage and length information data are used to train the initialized LightGBM model to obtain the length feature sub-model.
Citation Information
Patent Citations
Tobacco leaf curing state recognition method, device and system based on fusion model
CN112163527A
Flue-cured tobacco process stage prediction method and system based on model fusion
CN114997463A
Baking control method and device, baking equipment and storage medium
CN115211578A
Tobacco leaf data fusion curing characteristic parameter monitoring method based on curing barn observation window
CN117668487A
Tobacco leaf weight correction method in baking process based on transformer
CN118135316A
