Method and device for predicting tobacco baking yield based on multi-source data fusion

Abstract

The invention relates to a method and device for predicting the tobacco baking yield based on multi-source data fusion, and the method comprises the steps: collecting the initial multi-source data of to-be-baked tobacco leaves before the tobacco leaves are sent to baking equipment, and the initial multi-source data at least comprise the image data, the moisture content data, the fresh weight data and the regional reference data of the to-be-baked tobacco leaves; performing preprocessing and feature extraction operation on the initial multi-source data to obtain a multi-dimensional feature set for yield prediction; inputting the multi-dimensional feature set into a preset output prediction model to obtain a predicted output rate of the to-be-baked tobacco leaves; and based on the fresh weight data and the predicted output rate, calculating the predicted post-baking yield of the to-be-baked tobacco leaves through a preset single-batch yield prediction formula. Therefore, the problems of large prediction error, delayed prediction time, one-sided prediction model, low precision and the like of the tobacco baking yield are solved, accurate prediction of single-batch and regional total yield is realized before tobacco baking, and prospective support is provided for expected management of tobacco growers and plan adjustment of purchasing units.

Description

Technical Field

[0001] This application relates to the interdisciplinary field of agricultural information technology and intelligent management, specifically to a method and apparatus for predicting tobacco curing yield based on multi-source data fusion. Background Technology

[0002] Tobacco curing is a crucial step in tobacco agricultural production that determines the final yield and economic benefits. After harvesting, tobacco farmers send the fresh tobacco leaves to curing barns for curing. During this process, the tobacco leaves undergo complex physiological and biochemical changes, resulting in significant moisture loss and transformation of dry matter, leading to a substantial reduction in weight. Therefore, accurately predicting the yield of dried tobacco before curing is of paramount importance for tobacco farmers in planning production and assessing expected returns, as well as for tobacco purchasing units in coordinating storage, logistics, and formulating purchasing strategies.

[0003] Currently, production relies heavily on the personal experience of tobacco farmers or technicians, who estimate curing losses by visually observing the color, size, and thickness of tobacco leaves and combining this with past experience. This method is highly dependent on personal experience, cannot be quantified, and different assessors will arrive at vastly different conclusions. The prediction results are extremely unstable and cannot provide a reliable basis for refined agricultural management and business decision-making.

[0004] In related technologies, the prediction of tobacco curing yield is mostly done after the curing process is completed by weighing and counting the dry tobacco leaves to obtain accurate yield data. This "post-event statistics" rather than "pre-event prediction" method cannot provide tobacco farmers with yield expectations before curing begins, thus losing the guiding value of prediction itself for production planning and market expectations. Tobacco farmers are in an "unknown" state about the final output throughout the entire curing cycle, which is not conducive to risk management and resource allocation.

[0005] Furthermore, some technologies rely solely on a single type of data to predict tobacco curing yield. For example, they might use only the initial weight of fresh tobacco leaves and estimate yield using a fixed, empirical yield coefficient; or they might attempt to establish a relationship between a single characteristic (such as color) and yield. Tobacco curing is a complex process influenced by multiple factors, including initial weight, moisture content, maturity (visually represented by color), regional climate, and planting practices, all of which collectively determine the final yield. No single factor can fully reflect the overall condition of the tobacco leaves, resulting in predictive models based on a single data source that are biased, unstable, and lack accuracy, failing to adapt to the actual conditions of different batches and regions of tobacco leaves.

[0006] Therefore, there is currently a lack of a method that can accurately, objectively, and automatically predict tobacco curing yield by comprehensively considering multiple influencing factors before curing, and this problem urgently needs to be solved. Summary of the Invention

[0007] This application provides a method and apparatus for predicting tobacco curing yield based on multi-source data fusion, which solves the problems of large error, delayed prediction timing, one-sided prediction model and low accuracy in the prediction of tobacco curing yield. It enables accurate prediction of single batch and regional total yield before tobacco curing, and provides forward-looking decision support for tobacco farmers' expected management and purchasing unit's plan adjustment.

[0008] The first aspect of this application provides a method for predicting tobacco curing yield based on multi-source data fusion, including the following steps: The initial multi-source data of the tobacco leaves to be cured is collected before they are fed into the curing equipment. The initial multi-source data includes at least: image data, moisture content data, fresh weight data and regional reference data of the tobacco leaves to be cured. The initial multi-source data is preprocessed and feature extracted to obtain a multi-dimensional feature set for yield prediction. The multi-dimensional feature set is input into a preset yield prediction model to obtain the predicted yield of the tobacco leaves to be cured. Based on the fresh weight data and the predicted yield, the predicted yield of the tobacco leaves to be dried is calculated using a preset single-batch yield prediction formula.

[0009] Optionally, in some embodiments, the regional reference data includes at least: historical initial data of tobacco leaves cured by other tobacco farmers in the preset area and their corresponding historical actual output rate, tobacco planting area in the preset area, historical yield per unit area, and historical climate data for the same period.

[0010] Optionally, in some embodiments, after calculating the predicted yield of the tobacco leaves to be cured using a preset single-batch yield prediction formula, the method further includes: Obtain the predicted post-drying yield of all tobacco leaf batches with collected initial multi-source data within the preset area, and sum them to obtain the first yield estimate; Based on the tobacco planting area and the historical yield per unit area, the yield of tobacco leaf batches in the preset area that did not have initial multi-source data collected is estimated by scaling method, and the estimated yield is summed with the first yield estimate to obtain the second yield estimate; Based on the historical climate data for the same period, at least one climate influencing factor is identified, and the second yield estimate is corrected based on the climate influencing factor, and the final regional total post-drying yield forecast is output.

[0011] Optionally, in some embodiments, before inputting the multi-dimensional feature set into a preset yield prediction model, the following steps are included: Obtain a training sample set, which contains initial multi-source data of multiple historical batches of tobacco leaves, as well as the actual post-drying yield of each batch. Feature extraction is performed on the initial multi-source data of each training sample in the training sample set to obtain a multi-dimensional feature set of the training sample set; An initial model is constructed based on the gradient boosting decision tree algorithm, and the initial predicted values ​​of the initial model are initialized based on the central tendency statistics of all actual post-baking yields in the training sample set. The multi-dimensional feature set of the training sample set is input into the initial model, and the actual post-baking output rate is used as the supervision signal for training. During the training process, multiple decision trees are constructed in an iterative manner. Each new decision tree is used to fit the residual between the previous round of overall prediction results and the actual post-baking output rate. When the loss function value of the initial model on the training sample set is less than a preset threshold, or when the number of iterations reaches a preset maximum value, training stops, and the preset output prediction model is obtained.

[0012] Optionally, in some embodiments, after calculating the predicted yield of the tobacco leaves to be cured using a preset single-batch yield prediction formula, the method further includes: Collect the actual weight of the dried tobacco leaves after baking; Based on the actual dry weight of the tobacco leaves after drying and the fresh weight data, the actual yield of the tobacco leaves to be dried is calculated. The initial multi-source data of the tobacco leaves to be cured and the actual post-curing yield of the tobacco leaves to be cured are used to form a new training sample, and the new training sample is added to the training sample set. The preset output prediction model is updated using a preset incremental learning strategy, wherein the preset incremental learning strategy independently configures training parameters, including at least one of learning rate, number of iterations, or loss function threshold.

[0013] Optionally, in some embodiments, the preprocessing and feature extraction operations on the initial multi-source data include: The initial multi-source data is preprocessed, including at least one of data cleaning, format standardization, and outlier handling; Feature extraction is performed on the preprocessed data, including: Visual features are extracted from the image data, including the color and texture features of the tobacco leaves; Physical features are extracted from the moisture content data and the fresh weight data, including indicators for characterizing the uniformity of moisture distribution and indicators for characterizing tobacco leaf density. Reference features are extracted from the regional reference data, including historical yields similar to the current batch of tobacco leaves, determined based on geographical and temporal proximity.

[0014] Optionally, in some embodiments, the preset formula for predicting single-batch yield is: multiplying the fresh weight data by the predicted yield rate to calculate the predicted post-drying yield.

[0015] A second aspect of this application provides an apparatus for predicting tobacco curing yield based on multi-source data fusion, comprising: The acquisition module is used to acquire initial multi-source data of the tobacco leaves to be baked before they are fed into the baking equipment. The initial multi-source data includes at least: image data, moisture content data, fresh weight data and regional reference data of the tobacco leaves to be baked. The preprocessing module is used to preprocess and extract features from the initial multi-source data to obtain a multi-dimensional feature set for yield prediction. The modeling module is used to input the multi-dimensional feature set into a preset yield prediction model to obtain the predicted yield of the tobacco leaves to be cured. The calculation module is used to calculate the predicted yield of the tobacco leaves to be dried based on the fresh weight data and the predicted yield rate, using a preset single-batch yield prediction formula.

[0016] A third aspect of this application provides an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor. The processor executes the program to implement the method for predicting tobacco curing yield based on multi-source data fusion as described in the above embodiments.

[0017] A fourth aspect of this application provides a computer program product having a computer program stored thereon, which is executed by a processor to implement the method for predicting tobacco curing yield based on multi-source data fusion as described in the above embodiments.

[0018] The beneficial effects of the embodiments of this application are as follows: (1) By cross-validating multi-source data such as color features, moisture distribution, and physical parameters, a more accurate yield prediction model was established, which greatly reduced the prediction error rate of tobacco curing yield, effectively improved the prediction accuracy, transformed subjective visual assessment into objective data analysis, and eliminated the uncertainty brought about by human factors. (2) The breakthrough from “post-event statistics” to “pre-event forecasting” enables the yield forecast to be completed before the tobacco leaves are sent to the drying room, bringing forward the forecast time point, providing tobacco farmers with immediate yield expectations, supporting the arrangement of drying batches and the formulation of sales plans, and realizing the transformation from “unknown waiting” to “full-process knowability” in the management model. (3) Provide customized predictions for the unique characteristics (color, moisture, variety, etc.) of each batch of tobacco leaves. Based on geographical and temporal proximity, automatically adapt to the planting characteristics and climate conditions of different regions to achieve personalized and accurate predictions. (4) It can expand from single-batch forecasting to regional total output forecasting, support macro-level decision-making, provide accurate regional output estimates for tobacco purchasing units, optimize warehousing logistics and purchasing plans, and realize full-chain data connectivity and collaborative optimization from farmers' production to enterprise purchasing; (5) By continuously collecting new data, the model in this application embodiment can automatically optimize and adapt to changing production conditions, and realize the systematic accumulation and inheritance of agricultural knowledge.

[0019] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0020] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein: Figure 1 This is a flowchart illustrating a method for predicting tobacco curing yield based on multi-source data fusion, according to an embodiment of this application. Figure 2 A flowchart of a training method for a preset yield prediction model according to a specific embodiment of this application; Figure 3 This is a flowchart illustrating a method for predicting tobacco curing yield based on multi-source data fusion according to a specific embodiment of this application; Figure 4 This is a block diagram of an apparatus for predicting tobacco curing yield based on multi-source data fusion, according to an embodiment of this application. Figure 5 This is a block diagram of an electronic device provided according to an embodiment of this application. Detailed Implementation

[0021] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0022] The following describes a method and apparatus for predicting tobacco curing yield based on multi-source data fusion, according to embodiments of this application, with reference to the accompanying drawings. Addressing the problems mentioned in the background art regarding large prediction errors, delayed prediction timing, and biased and low-accuracy prediction models in tobacco curing yield prediction, this application provides a method for predicting tobacco curing yield based on multi-source data fusion. In this method, initial multi-source data of the tobacco leaves to be cured before being fed into the curing equipment is collected. The initial multi-source data includes at least: image data, moisture content data, fresh weight data, and regional reference data of the tobacco leaves to be cured. Preprocessing and feature extraction operations are performed on the initial multi-source data to obtain a multi-dimensional feature set for yield prediction. The multi-dimensional feature set is input into a preset yield prediction model to obtain the predicted yield rate of the tobacco leaves to be cured. Based on the fresh weight data and the predicted yield rate, the predicted post-curing yield of the tobacco leaves to be cured is calculated using a preset single-batch yield prediction formula. This solves the problems of large prediction errors, delayed prediction timing, and biased and low-accuracy prediction models in tobacco curing yield prediction.

[0023] Specifically, Figure 1 This is a flowchart illustrating the method for predicting tobacco curing yield based on multi-source data fusion provided in an embodiment of this application.

[0024] like Figure 1 As shown, the method for predicting tobacco curing yield based on multi-source data fusion includes the following steps: In step S101, initial multi-source data of the tobacco leaves to be cured are collected before they are fed into the curing equipment. The initial multi-source data includes at least: image data of the tobacco leaves to be cured, moisture content data, fresh weight data and regional reference data.

[0025] In this application embodiment, the initial multi-source data refers to a collection of various types of data about the current batch of tobacco leaves and its surrounding environment, collected through various sensors and information technologies before the tobacco leaves are fed into the curing equipment. Image data of the tobacco leaves to be cured refers to digital information that reflects the visual characteristics of the tobacco leaf surface, acquired through image acquisition devices (such as cameras). Moisture content data represents the percentage of water content in the tobacco leaf tissue relative to the total mass of the tobacco leaf (wet basis). Fresh weight data is the total weight of the current batch of tobacco leaves to be cured in its original moisture content state before curing. Regional reference data does not originate from the current batch itself, but rather from its geographical location and historical reference information related to yield from the same period.

[0026] Optionally, in some embodiments, the regional reference data includes at least: historical initial data of tobacco leaves cured by other tobacco farmers in the preset area and their corresponding historical actual yield, tobacco planting area in the preset area, historical yield per unit area and historical climate data for the same period.

[0027] Historical initial data refers to similar data (such as images, moisture content, and fresh weight) collected from other tobacco farmers in the same region before their tobacco was sent to the curing barn during past curing seasons. Historical actual yield rate refers to the actual yield ratio obtained by calculating the ratio of actual dry tobacco weight to fresh weight after curing for those historical batches. The tobacco planting area of ​​the preset region refers to the total land area actually planted with tobacco during a specific harvest season within a defined geographical area (such as a county or town). Historical yield per unit area refers to the average weight of dry tobacco produced per unit of land area (such as per acre) during past curing seasons in the preset region. Historical climate data for the same period refers to the main meteorological data of the region during the same period last year or in previous years (i.e., the tobacco harvesting and curing season), such as average temperature, rainfall, sunshine hours, and air humidity.

[0028] Specifically, regarding the acquisition of image data of tobacco leaves to be cured, this application embodiment employs technologies such as machine vision, digital image acquisition, and color space models. A stable lighting acquisition box is built at the entrance of the curing barn. When the tobacco leaves are conveyed through, an industrial color camera automatically captures high-definition images, thereby outputting one or more tobacco leaf image files in JPG (Joint Photographic Experts Group) / PNG (Portable Network Graphics) format.

[0029] Regarding the acquisition of moisture content data of tobacco leaves to be cured, this application embodiment adopts near-infrared spectroscopy analysis technology, using a portable or online near-infrared moisture meter. The instrument emits near-infrared light and receives the spectrum reflected back from the tobacco leaves, and quickly calculates the moisture content through a built-in algorithm.

[0030] Regarding the acquisition of fresh weight data of tobacco leaves to be cured, this application embodiment adopts high-precision sensor technology. Before the tobacco leaves are sent into the curing barn, the whole frame or whole stalk of tobacco leaves is weighed by electronic weighbridge or weighing conveyor belt, and the reading is automatically recorded in the system.

[0031] Regarding the historical initial data and corresponding historical actual yield rate in the regional reference data of tobacco leaves to be cured, this application embodiment establishes a centralized cloud database. After each curing barn connected to the cloud database completes the curing of a batch of tobacco leaves, it automatically packages its "initial multi-source data" and the final verified "actual yield rate" into a data package, uploads and stores it. When it is necessary to predict a new batch, this application embodiment can automatically query the cloud database to find historical records that are similar to the current batch in terms of variety, geographical location and harvest time, forming a reference dataset.

[0032] Optionally, the tobacco planting area of ​​the preset area in the regional reference data of this application embodiment can be obtained by connecting with the database interface of an agricultural management department, or by using satellite remote sensing imagery to identify plots and calculate the area.

[0033] Optionally, the historical yield per unit area in the regional reference data of this application embodiment can be obtained from the regional agricultural statistical yearbook database or the tobacco company's acquisition statistics system.

[0034] Optionally, the historical climate data for the same period in the regional reference data of this application embodiment can be obtained by calling the API (Application Programming Interface) provided by the meteorological bureau or commercial meteorological service company (such as China Meteorological Data Network, Weather Underground) to obtain historical meteorological data for a specified area and time period.

[0035] Understandably, image data of tobacco leaves to be cured can reflect the maturity and potential quality of the leaves (e.g., leaves with a yellowish hue may be more mature, and their shrinkage patterns may differ), moisture content data can reflect quality loss during the curing process, fresh weight data provides a benchmark for yield calculation, and regional reference data provides "experience" and localized "calibration" for model learning in subsequent embodiments. Therefore, this application's embodiments, by integrating machine vision, physical sensing, and information technology, transform tobacco curing from a "craft" relying on personal experience into a quantifiable, analyzable, and predictable industrial process, laying a solid data foundation for subsequent intelligent prediction.

[0036] In step S102, the initial multi-source data is preprocessed and feature extracted to obtain a multi-dimensional feature set for yield prediction.

[0037] In this application embodiment, preprocessing refers to a series of cleaning, repairing, and standardizing operations performed on the original data (the initial multi-source data in this application embodiment). Its goal is to resolve inconsistencies, incompleteness, and noise in the original data, ensuring data quality and consistency, and providing a reliable foundation for subsequent analysis. Feature extraction refers to the process of identifying and extracting representative quantitative indicators that have a significant impact on the prediction target (yield) from the preprocessed data through mathematical transformations or statistical methods. In other words, it elevates "raw information" to "effective information." For example, instead of directly using millions of pixel values ​​from a tobacco leaf image, dozens of numbers with clear physical meaning, such as "average greenness" and "texture roughness," are calculated.

[0038] The multi-dimensional feature set in this application embodiment is a structured and digitized collection of all features extracted from data from different sources (images, physical data, regions), which can be represented as a feature vector or feature matrix. The multi-dimensional feature set is the final output of preprocessing and feature extraction operations, and each feature describes the potential state of this batch of tobacco leaves from a unique dimension.

[0039] Therefore, in step S102, the embodiments of this application transform the raw, low-value "data" into high-quality "information" with clear physical and statistical significance. The multi-dimensional feature set comprehensively and quantitatively describes the health status and potential value of the current batch of tobacco leaves, enabling the subsequent model to make accurate predictions.

[0040] Optionally, in some embodiments, preprocessing and feature extraction operations are performed on the initial multi-source data, including: preprocessing the initial multi-source data, including at least one of data cleaning, format standardization, and outlier handling; and extracting features from the preprocessed data, including: extracting visual features from image data, the visual features including color features and texture features of tobacco leaves; extracting physical features from moisture content data and fresh weight data, the physical features including indicators for characterizing the uniformity of moisture distribution and indicators for characterizing tobacco leaf density; and extracting reference features from regional reference data, the reference features including historical yield rates similar to the current batch of tobacco leaves determined based on geographical and temporal proximity.

[0041] "Data cleaning" in this application embodiment refers to processing missing, erroneous, or irrelevant information in the data. For example, if a moisture sensor reading is lost, the control system (which integrates the method for predicting tobacco curing yield based on multi-source data fusion in this application embodiment) fills it in with the average or median of other readings from the same batch. Another example is that if the fresh weight data is recorded as 0.5 kg (significantly too light) due to equipment interference, the control system marks it as invalid and triggers a re-weighing or replaces it with a reasonable value from the previous batch. Yet another example is automatically identifying and cropping non-tobacco leaf areas (such as background or baskets) from an image.

[0042] The "format standardization" in this application refers to converting data from different sources into a unified format and scale to make them comparable. Specifically, it involves standardizing the image data format, that is, converting image data into the same resolution, file format, and color space. For numerical standardization, data with vastly different dimensions, such as "fresh weight" (several hundred kilograms) and "moisture content" (0-100%), are converted to similar orders of magnitude through Z-score normalization or Min-Max scaling, preventing subsequent model calculations from being dominated by features with large numerical values.

[0043] "Outlier handling" in this application refers to identifying and processing data points that significantly deviate from the normal range. Specifically, this application employs statistical methods to identify outliers, such as box plots or the 3σ principle (assuming the data follows a normal distribution, data exceeding three standard deviations from the mean are considered outliers). Thus, outliers are removed or replaced with upper or lower limits to prevent these "noise points" from interfering with model training and prediction.

[0044] Furthermore, in this embodiment, "feature extraction" refers to the quantitative calculation of indicators effective for predicting yield from the cleaned data, which is the core step in transforming data into information. Specifically, the visual features in this embodiment are mathematical indicators calculated from image data that describe the appearance of tobacco leaves. The physical features are indicators derived from moisture content data and fresh weight data that describe the inherent physical properties of tobacco leaves. The reference features are indicators extracted from regional reference data that provide a prediction benchmark for the current batch.

[0045] Specifically, for the color features of tobacco leaves, this application embodiment uses libraries such as OpenCV (Open Source Computer Vision Library) to convert images from RGB (Red, Green, Blue, an additive color model) to HSV (Hue, Saturation, Value, a color model oriented towards visual perception) color space, calculates the average hue (H) and average saturation (S), or counts the percentage of yellow / green pixels to quantify maturity. For the texture features of tobacco leaves, this application embodiment uses the gray-level co-occurrence matrix algorithm to calculate the contrast, entropy, etc., of the image, quantifying the surface structure of the leaves such as wrinkles and thickness.

[0046] The moisture distribution uniformity index in this application refers to the standard deviation or coefficient of variation of the moisture values ​​at multiple sampling points in a batch of tobacco leaves. The larger the value, the more uneven the moisture content of the entire batch of tobacco leaves, and the higher the difficulty of curing and the risk of loss. The tobacco leaf density index is calculated by dividing the fresh weight by the tobacco leaf pile volume (when the tobacco leaf pile volume can be estimated), or by using the leaf area estimated by image analysis to calculate the "weight per unit area" as a proxy index of density.

[0047] Regarding reference features, embodiments of this application can calculate the similarity between the current batch and historical batches in terms of geographical location and harvesting time (such as Euclidean distance). Then, based on the similarity weight, a weighted average of the actual yield rates of the selected similar historical batches is performed to obtain an "expected yield rate of similar tobacco leaves", which serves as the prediction prior for the current batch.

[0048] Understandably, high-quality feature data is a prerequisite for achieving high-precision prediction. Through step S102, this embodiment transforms the qualitative experience of experienced craftsmen—such as "judging maturity by color" and "judging shrinkage by feeling dryness"—into precise numerical features like "average hue = 65" and "moisture variation coefficient = 0.15." This ensures that the data input to the model is refined and highly relevant, thereby improving the model's performance ceiling and training efficiency.

[0049] In step S103, the multi-dimensional feature set is input into the preset yield prediction model to obtain the predicted yield of the tobacco leaves to be cured.

[0050] In other words, in step S103, this embodiment of the application uses a pre-trained intelligent model to calculate the refined "multi-dimensional feature set" and output an accurate and quantitative predicted output rate, thereby scientifically predicting the curing result of the current batch of tobacco leaves.

[0051] The preset yield prediction model in this application embodiment is an algorithm system pre-trained using a large amount of historical data, capable of learning the complex mapping relationship between "initial characteristics of tobacco leaves" and "final yield rate". The predicted yield rate in this application embodiment is the expected curing yield ratio of the current batch of tobacco leaves calculated by the preset yield prediction model based on the input multi-dimensional feature set. The calculation formula is as follows: Predicted yield = Predicted dry tobacco weight / Fresh tobacco weight; The projected output rate is a value between 0 and 1, and is a projected value with direct economic significance. For example, a projected output rate of 0.20 means that 100 kg of fresh tobacco leaves are expected to produce 20 kg of dried tobacco leaves.

[0052] This application provides a modeling method for a preset yield prediction model: by utilizing accumulated historical big data, an initial "blank" algorithm system (initial model) automatically learns and masters the complex and hidden mapping rules between "initial multi-source characteristics of tobacco leaves" and "final baking yield", thereby forming a fixed intelligent model (preset yield prediction model) that can be used for future prediction.

[0053] Specifically, in some embodiments, before inputting the multi-dimensional feature set into the preset yield prediction model, the process includes: acquiring a training sample set, which contains initial multi-source data of multiple historical batches of tobacco leaves and the actual post-drying yield rate corresponding to each batch; extracting features from the initial multi-source data of each training sample in the training sample set to obtain a multi-dimensional feature set of the training sample set; constructing an initial model based on a gradient boosting decision tree algorithm, and initializing the initial prediction value of the initial model based on the central tendency statistics of all actual post-drying yield rates in the training sample set; inputting the multi-dimensional feature set of the training sample set into the initial model, using the actual post-drying yield rate as a supervision signal for training, and during the training process, constructing multiple decision trees iteratively, with each new decision tree used to fit the residual between the overall prediction result of the previous round and the actual post-drying yield rate; stopping training when the loss function value of the initial model on the training sample set is less than a preset threshold, or when the number of iterations reaches a preset maximum value, thus obtaining the preset yield prediction model.

[0054] The training sample set is a dataset used to "teach" the model. Each sample (i.e., each row of data) contains a historical batch of "questions" (initial multi-source data) and a corresponding "standard answer" (actual post-drying yield). The actual post-drying yield is calculated by the ratio of the actual weight of dried tobacco after drying to the weight of fresh tobacco. It is a real yield ratio verified in practice and is the target of model learning. In machine learning, it is called a "label" or "supervision signal".

[0055] The gradient boosting decision tree algorithm used in this application is a powerful ensemble machine learning algorithm. Its core idea is to construct multiple simple decision trees sequentially. Each new tree is dedicated to correcting the prediction errors of all previous trees combined. Through this "teamwork" approach, the most accurate prediction is gradually approached. The initial prediction value refers to the simplest and most basic prediction value made by the model for all samples before formally training the first tree.

[0056] The central tendency statistic in this application describes an indicator of the degree to which a set of data clusters around a central point, such as the mean or median. The residual in this application refers to the difference between the "true value" (the actual baked output rate in this application) and the "current model prediction value" in each iteration. The residual acts as a "compass" for model learning, indicating the target that the next tree needs to strive to correct. The loss function value in this application is a mathematical function that quantifies the overall prediction error of the current model across all training samples, such as mean squared error. It calculates the average of the squared differences between the predicted and true values ​​of all samples. In other words, the smaller the loss function value, the more accurate the model prediction. The goal of training the model in this application is to minimize the loss function value.

[0057] Figure 2 This is a flowchart of a training method for a preset yield prediction model according to a specific embodiment of this application. The following will be combined with... Figure 2 The training process of the pre-set yield prediction model will be explained in detail.

[0058] Specifically, such as Figure 2 As shown, the training of the model first requires preparing the training sample set. This step is consistent with the processing of the prediction data described above, that is, reading the "initial multi-source data" of all historical batches of tobacco leaves, performing the same preprocessing and feature extraction operations, generating a "multi-dimensional feature set" for each historical batch of tobacco leaves, resulting in a large table. Each row of the table is a feature vector of a sample, and the last column of the table is the "actual post-drying yield" (supervision signal) of that sample. Then, the model is initialized. In this embodiment, a gradient boosting decision tree algorithm (e.g., using Python's LightGBM or XGBoost library) is used to create a model framework. Subsequently, the average value of the "actual post-drying yield" of all training samples is calculated, for example, 0.195. At this point, the model sets an initial, very simple prediction rule: for any input, it is first predicted as 0.195, that is, the initial prediction value F0(x) = 0.195.

[0059] Furthermore, after determining the initial model and initial predicted values, iterative training and residual fitting of the model begin. This application's embodiment uses the first iteration as an example to illustrate the model iteration process. First, the model uses the initial predicted value F0(x) to predict all samples and calculates the residuals between the predicted and true values. Then, the first decision tree T1 is constructed. The learning objective of decision tree T1 is not to directly predict the output rate, but to learn how to predict these residuals based on features. It will find the best split point in the feature set (e.g., "if the color saturation is <0.6, then the average residual is +0.02"), forming a simple rule tree. Then, the prediction result of the first decision tree T1 is added to the initial prediction with a small learning rate (e.g., η=0.1) to form a new model: F1(x)=F0(x)+η*T1(x). Thus, the above process is repeated. The second decision tree T2 is used to fit the new residuals between F1(x) and the true values, and then the model is updated: F2(x)=F1(x)+η*T2(x), and so on.

[0060] Furthermore, after each iteration, the control system calculates the loss function value (such as mean squared error) of the model on the current entire training set. Training continues until either of the following conditions is met: the loss function value drops to a preset level (less than a preset threshold), indicating that the model is accurate enough; or the number of iterations reaches a preset maximum value (such as 1000 times), to prevent infinite loops and overfitting.

[0061] After training stops, the final model F_final(x) is a complex prediction system composed of those hundreds of decision trees and an initial prediction value. This system is saved and becomes the preset yield prediction model in this application embodiment.

[0062] Therefore, this application embodiment uses the iterative learning mechanism of the gradient boosting decision tree algorithm to transform scattered multi-source historical data into a mathematical model with high-precision prediction capabilities. It can automatically mine and grasp the complex nonlinear relationship between the initial characteristics of tobacco leaves and the final yield, thereby realizing a fundamental shift from relying on human experience to data-driven intelligent prediction, and providing reliable core algorithm support for subsequent accurate yield prediction.

[0063] In step S104, based on fresh weight data and predicted yield, the predicted yield of the tobacco leaves to be dried is calculated using a preset single-batch yield prediction formula.

[0064] Among them, the predicted yield after drying is the final output, which is the expected weight of dried tobacco leaves after the current batch of tobacco leaves is dried. It is the most practical and guiding decision-making basis provided by the entire method for tobacco farmers and purchasing units. The preset single-batch yield prediction formula in this application embodiment is a predefined mathematical calculation rule. Its input is "fresh weight data" and "predicted yield rate", and its output is "predicted yield after drying".

[0065] Optionally, in some embodiments, the preset formula for predicting the yield of a single batch is: multiplying the fresh weight data by the predicted yield to calculate the predicted yield after drying.

[0066] In other words, the predicted yield after drying (dry weight) = fresh weight data × predicted yield rate, where the predicted yield rate represents the "retention ratio" from fresh tobacco leaves to dried tobacco leaves. Multiplying the initial total weight (fresh weight) by this retention ratio naturally yields the final retained weight (dry weight). For example, if the predicted yield rate is 0.20, then the predicted yield after drying of 350 kg of fresh leaves is 70 kg.

[0067] The embodiments of this application strictly rely on the initial objective measurement of "fresh weight" for the final yield, ensuring that the prediction results do not deviate from physical reality, improving the credibility of the prediction results, and the multiplication operation is not prone to errors, the calculation speed is fast, and the results are easy to understand and verify. This simplicity makes the execution of the whole method very stable and reliable.

[0068] Therefore, in this embodiment of the application, the fresh weight data (a precise measurement baseline value) directly collected in step S101 and the predicted yield rate (a ratio value between 0.15 and 0.25) output by the model in step S103 are used to call the preset single batch yield prediction formula to calculate a specific weight value, that is, the predicted yield after drying, and this result is presented to the user (such as displayed on the screen or mobile app).

[0069] Therefore, the embodiments of this application transform the abstract "output rate" into the concrete "output", so that the prediction results can be directly used for economic benefit analysis, logistics arrangement and sales planning, realizing the last mile leap from data analysis to business decision-making. Tobacco farmers can clarify their expected income based on the "predicted output after drying", and purchasing units can formulate precise purchasing, warehousing and transportation plans accordingly.

[0070] Optionally, in some embodiments, after calculating the predicted post-drying yield of the tobacco leaves to be cured using a preset single-batch yield prediction formula, the method further includes: obtaining the predicted post-drying yields of all tobacco leaf batches with collected initial multi-source data within a preset area, and summing them to obtain a first yield estimate; estimating the yield of tobacco leaf batches without collected initial multi-source data within the preset area using a scaling method based on the tobacco planting area and historical yield per unit area, and summing the estimated yield with the first yield estimate to obtain a second yield estimate; determining at least one climate influence factor based on historical climate data for the same period, correcting the second yield estimate based on the climate influence factor, and outputting the final predicted total post-drying yield for the region.

[0071] It is understood that the method for predicting tobacco curing yield based on multi-source data fusion in this application embodiment can also achieve "prediction of total regional curing yield", realizing a powerful capability to expand from serving "micro-decision" for individual tobacco farmers to serving "macro-decision" for enterprises. Through the strategy of "from point to surface, statistical estimation, and climate correction", it cleverly uses partial data to infer the overall situation, realizing the leap from "single-point perception" to "regional insight".

[0072] In this embodiment, the first yield estimate is the sum of yields predicted based on tobacco leaf batches for which data has already been collected, using the method described in the above embodiments (steps S101-S104). The scaling method in this embodiment is a statistical inference method for estimating the yield of unknown portions based on area and average yield per unit area. The second yield estimate in this embodiment is the sum of the first yield estimate and the estimated yield of uncollected batches; it is a preliminary yield prediction covering the entire preset area. The climate impact factor in this embodiment refers to an indicator or coefficient extracted from historical climate data that can quantify its impact on tobacco yield. The final regional total post-drying yield prediction in this embodiment refers to the final regional total yield prediction obtained after correction by the climate impact factor.

[0073] Specifically, regarding the first yield estimate of this application embodiment, all batches located in the "preset area" (such as a county) that have been sent for drying and whose data have been collected in the current harvest season can be queried from the database. The "predicted yield after drying" field of these batches is summed to obtain a specific weight value, denoted as ∑P_known.

[0074] Then, obtain the tobacco planting area A_known covered by the collected batches, calculate the yield per unit area of ​​the collected area, and the formula for calculating the historical yield per unit area is: Yield_per_Acre=∑P_known / A_known; Where Yield_per_Acre is the historical yield per unit area, ∑P_known is the first estimated yield, and A_known is the tobacco planting area.

[0075] Then, obtain the total planting area A_total for the entire region, and calculate the uncollected area using the following formula: A_unknown = A_total - A_known; Where A_unknown represents the area not collected, A_total represents the total planting area of ​​the entire region, and A_known represents the tobacco planting area.

[0076] The formula for estimating the total yield of the uncollected portion (the yield of tobacco batches within the preset area for which initial multi-source data was not collected) is as follows: ∑P_unknown=Yield_per_Acre×A_unknown; Where ∑P_unknown represents the total output of the uncollected portion, Yield_per_Acre represents the historical output per unit area, and A_unknown represents the uncollected area.

[0077] Therefore, the second production estimate is determined based on the sum of the first production estimate (∑P_known) and the total production of the uncollected portion (∑P_unknown).

[0078] Furthermore, regarding the climate impact factor in this application embodiment, it is possible to obtain climate data from the same historical period (such as the same period last year), compare the climate data of the current harvest season with the historical period, and convert the climate difference into a yield impact coefficient based on predefined rules or models. For example, if the average temperature of the current season is 2°C higher than the historical period and this temperature is within the suitable range, it is determined to be favorable, and the climate impact factor is set to +0.02 (representing a 2% increase in yield); or, if the rainfall of the current season is 30% lower than the historical period and reaches the drought standard, it is determined to be unfavorable, and the climate impact factor is set to -0.05 (representing a 5% decrease in yield).

[0079] Therefore, the revised calculation of the final regional total post-drying yield forecast can be expressed as the following formula: The predicted total yield after drying in the region = the second estimated yield × (1 + climate impact factor). For example, if the second production estimate is 10,000 tons and the climate impact factor is -0.03, then the final total regional production = 10,000 × (1 - 0.03) = 9,700 tons.

[0080] As described above, this application embodiment utilizes data aggregation and statistical techniques to rapidly aggregate predicted yields from collected batches in a regional database. Through proportional estimation, it infers the "total" (overall yield of the entire region) based on the "sample" (collected batches). It uses a geographic information system to obtain the boundaries of the "preset area" and the distribution of planting area within it for management and analysis. Through a climate data API interface, it obtains standardized historical and current climate data from the meteorological department to establish a quantitative relationship model between climate variables (temperature, precipitation, sunshine) and tobacco yield. This model can be a simple empirical lookup table or a small regression model. Through data fusion and correction algorithms, it fuses information from different sources (production data, area data, climate data) and performs the final correction calculation.

[0081] Therefore, the method for predicting tobacco curing yield based on multi-source data fusion in this application does not require 100% data collection coverage. Through scientific statistical sampling and proportional extrapolation, it can efficiently and cost-effectively estimate the yield of the entire region, solving the problem of macro-prediction under incomplete data conditions. The introduction of climate influence factors as correction terms makes the prediction results no longer static, but dynamically responsive to interannual climate fluctuations, realizing dynamic calibration and environmental adaptability of the prediction model, and greatly improving the accuracy and reliability of regional predictions in years with abnormal climate. Combined with the above-mentioned single-batch prediction, it serves not only the production arrangements of individual tobacco farmers, but also the meso-level management of cooperatives and townships, and can provide data-driven decision-making basis for the procurement plans, warehousing and logistics, market regulation and industrial policy formulation of provincial tobacco companies and regulatory departments. Agricultural yield statistics in related technologies are often lagging. This application realizes the dynamic output and updating of regional total yield predictions before the curing season, winning valuable decision-making time for all relevant parties.

[0082] Optionally, in some embodiments, after calculating the predicted post-drying yield of the tobacco leaves to be dried using a preset single-batch yield prediction formula, the method further includes: collecting the actual post-drying dry weight of the tobacco leaves after drying; calculating the actual post-drying yield rate of the tobacco leaves to be dried based on the actual post-drying dry weight and fresh weight data; constructing a new training sample by combining the initial multi-source data of the tobacco leaves to be dried with the actual post-drying yield rate of the tobacco leaves to be dried, and adding the new training sample to the training sample set; and updating the preset yield prediction model using a preset incremental learning strategy, wherein the preset incremental learning strategy independently configures training parameters, including at least one of the following: learning rate, number of iterations, or loss function threshold.

[0083] In this application embodiment, the actual weight of dried tobacco after drying refers to the actual weight of dried tobacco leaves measured after the current batch of tobacco leaves has completed drying. The actual yield rate after drying is the true output ratio calculated based on factual data, and the formula for calculating the actual yield rate after drying is as follows: Actual yield after drying = actual weight of dried tobacco after drying / fresh weight data; The new training sample in this application embodiment refers to a new, high-quality data pair consisting of the complete "question" (initial multi-source data) and the "new standard answer" (actual post-baking yield) of the current batch.

[0084] The preset incremental learning strategy in this application is a machine learning technique that allows the model to continuously update its parameters using newly generated data without forgetting old knowledge, rather than retraining from scratch.

[0085] Specifically, the actual weight of the dried tobacco after drying can be measured at the exit of the curing barn using an electronic weighbridge. This data is automatically recorded and entered into the control system, and then correlated with the aforementioned prediction records. In this embodiment, the actual yield after drying is calculated by retrieving the fresh weight data of the current batch from the database and comparing it with the collected actual weight of the dried tobacco after drying, resulting in an accurate yield value.

[0086] Then, the control system packages all the initial multi-source data (images, moisture, fresh weight, regional data, etc.) collected in step S101 for the current batch together with the calculated actual post-drying yield to form a new data record that can be used for training.

[0087] Regarding the preset incremental learning strategy in the embodiments of this application, those skilled in the art will preset a set of independent training parameters for the incremental learning process, including at least one of the following: learning rate, number of iterations, or loss function threshold. Preferably, the learning rate can be set to a smaller value (e.g., 0.01) to prevent new data from causing excessive impact on the model. Preferably, the number of iterations can be set to a smaller value (e.g., 50 times) because it is only for fine-tuning and does not require large-scale reconstruction. The embodiments of this application do not specifically limit the loss function threshold; those skilled in the art can set a reasonable accuracy according to actual needs to improve the target.

[0088] Understandably, gradient boosting decision tree frameworks (such as XGBoost and LightGBM) inherently support incremental fitting based on existing models. They don't erase the original model; instead, they use new data as new learning material, continuing to grow new trees or adjust the weights of leaf nodes based on the existing tree structure. Therefore, the control system adds new training samples to the training set and then uses this new data, following the configured incremental learning strategy, to perform a new round of training on the deployed, pre-defined yield prediction model. This fine-tunes the model parameters to better fit the combined patterns of the new and old data.

[0089] Therefore, this application takes into account the impact of climate, soil, and planting pattern changes on tobacco production, which would cause a static model to gradually become outdated. Thus, the embodiments of this application make the model a "living entity," capable of tracking and adapting to these changes, ensuring that prediction accuracy does not decrease over time, and resolving the contradiction between model staticity and dynamic environmental changes. Through a pre-set yield prediction model, when a new tobacco variety is introduced into a region, after curing several batches of that variety's leaves, the model can quickly learn the characteristics of that variety, thereby providing increasingly accurate predictions for subsequent similar tobacco leaves in that region. This enhances the personalization and adaptability of the prediction method, realizing personalized services "tailored to local conditions and specific varieties." Through closed-loop feedback, predictions are continuously tested with reality, and the model is corrected using the test results, greatly enhancing the long-term value and vitality of the model and prediction method.

[0090] The following specific embodiments illustrate the implementation process of the method for predicting tobacco curing yield based on multi-source data fusion according to the embodiments of this application.

[0091] In this embodiment, a tobacco-growing area exhibits distinct climatic characteristics and concentrated tobacco cultivation, facilitating data collection and model validation. This area comprises approximately 50 tobacco farmers, with a total planting area of ​​about 1000 mu (approximately 67 hectares). During a curing season, an estimated 500 batches of tobacco leaves are delivered to clustered curing barns for curing. The objective of this embodiment is to accurately predict the post-curing yield of a single batch of tobacco leaves before they are delivered to the curing barns, and further estimate the total yield of the entire area.

[0092] Figure 3 This is a flowchart illustrating a specific embodiment of a method for predicting tobacco curing yield based on multi-source data fusion, as shown in the following diagram. Figure 3 As shown, the method for predicting tobacco curing yield based on multi-source data fusion includes the following steps.

[0093] Step S301: Collect initial multi-source data; In this embodiment, when tobacco farmers transport a batch of Yunyan 87 middle leaves with a fresh weight of 350 kg to the curing barn, the control system (integrating the method of predicting tobacco curing yield based on multi-source data fusion in this application embodiment) automatically collects initial multi-source data, including: image data, physical parameters (moisture content data and fresh weight data) and regional reference data (historical output data, climate data and planting area).

[0094] Specifically, an industrial camera installed at the entrance of the curing barn captures images of tobacco leaves within a standard light source box to analyze their color and texture. A handheld near-infrared moisture meter is used for rapid, non-destructive measurement of the tobacco leaves, yielding an average moisture content of 82%. An electronic weighbridge automatically records the fresh weight of the tobacco leaves as 350.0 kg. The control system retrieves regional reference data from a cloud database, accessing data from 100 recently cured batches of the same variety in the current area, calculating an average actual yield of 0.195. Historical data from the same period during the middle and late stages of tobacco growth (maturity and harvesting period) in the current production area are obtained via meteorological API, showing an average temperature of 23.5°C, rainfall of 180 mm, and sunshine duration of 125 hours. The total planting area in the current area is also obtained from an agricultural management department's database, which indicates a total planting area of ​​1000 mu.

[0095] Step S302, data processing and feature extraction; The initial multi-source data in step S301 is processed to extract features that the model can understand.

[0096] Specifically, the visual features of the tobacco leaf images, including color and texture features, were calculated using the OpenCV library. The average greenness of the tobacco leaves was calculated, which is the average value of the S channel in the HSV color space of 65, and the proportion of yellow pixels was 15%, to quantify the maturity of the tobacco leaves. The contrast ratio was calculated to be 2.8 through the gray-level co-occurrence matrix, which can reflect the degree of leaf wrinkling.

[0097] The physical characteristics of tobacco leaves, including the uniformity of moisture distribution and leaf density, are analyzed in this application. The standard deviation of multiple moisture measurements is calculated to be 1.5% (a higher value indicates greater uneven moisture distribution throughout the batch of tobacco leaves). The leaf density is calculated by dividing the fresh weight by the leaf area estimated from the image, yielding a weight per unit area of ​​0.45 kg / m². 2 .

[0098] The reference characteristics of tobacco leaves are based on geographical and temporal proximity, and the weighted historical output rate is calculated to be 0.197, which serves as a priori reference for prediction.

[0099] Ultimately, the control system generates a "multi-dimensional feature set" for the current batch of tobacco leaves, which includes information such as color, texture, moisture, density, and historical yield.

[0100] Step S303, Model Predictive Analysis; The multi-dimensional feature set obtained in step S302 is input into the preset yield prediction model of this application embodiment. After the model performs rapid inference, it outputs the predicted yield rate of the current batch of tobacco as 0.203.

[0101] Step S304, Calculation and Optimization; The control system calls the preset single-batch yield prediction formula to calculate the predicted yield after drying. The preset single-batch yield prediction formula is: predicted yield after drying = fresh weight data × predicted output rate. Substituting the data into the formula, i.e. 350.0kg × 0.203 = 71.05kg, we get the expected yield of dry tobacco in the current batch of tobacco leaves as approximately 71.0 kg.

[0102] Furthermore, the control system aggregates the predicted yields of 200 batches already collected within the area, summing them to obtain the first estimated yield ∑P_known = 14200 kg; based on the 500 mu area covered by the collected batches, the yield per unit area is calculated as follows: 14200 / 500 = 28.4 kg / mu; Further estimate of the uncollected batches: 1000 - 500 = 500 mu; The production volume of the uncollected batches is: 28.4 × 500 = 14200 kg; The total estimated value for the region is: 14200 + 14200 = 28400 kg.

[0103] Furthermore, based on the climate data correction results, the control system determines that the average temperature during the current harvest season is higher than the historical average for the same period. Therefore, the climate impact factor is set to +0.02 (slightly favorable), and the final regional total post-drying yield is corrected as follows: 28400 × (1 + 0.02) = 28968 kg; This result can serve as a reference for tobacco company acquisition plans.

[0104] Finally, after the current batch of tobacco leaves was cured, the actual weight of the dried tobacco was 70.5 kg. The control system calculated the actual yield after curing as follows: 70.5 / 350.0=0.201; Then, the initial multi-source data of the current batch and the actual post-baking output rate are combined to form a new training sample, which is added to the training sample set of the model. An incremental learning strategy (configuration: learning rate 0.01, iterations 50 times) is used to fine-tune and update the model to make it more adaptable to actual production and more accurate with use.

[0105] Therefore, as can be seen from the above embodiments, the method for predicting tobacco curing yield based on multi-source data fusion in this application embodiment accelerates the prediction timeliness of tobacco curing yield, improves the prediction accuracy, realizes the fundamental transformation from "experience-driven" to "data-driven", and provides a practical and feasible technical path for the intelligent upgrading of tobacco agriculture.

[0106] The method for predicting tobacco curing yield based on multi-source data fusion proposed in this application involves collecting initial multi-source data of the tobacco leaves to be cured before they are fed into the curing equipment. This initial multi-source data includes at least: image data, moisture content data, fresh weight data, and regional reference data of the tobacco leaves. The initial multi-source data is preprocessed and feature extracted to obtain a multi-dimensional feature set for yield prediction. This multi-dimensional feature set is then input into a preset yield prediction model to obtain the predicted yield rate of the tobacco leaves to be cured. Based on the fresh weight data and the predicted yield rate, the predicted post-curing yield of the tobacco leaves is calculated using a preset single-batch yield prediction formula. This method solves the problems of large prediction errors, delayed prediction timing, and biased and low-accuracy prediction models in tobacco curing yield prediction. It achieves accurate prediction of single-batch and regional total yields before tobacco curing, providing forward-looking decision support for tobacco farmers' expected management and the adjustment of purchasing unit plans.

[0107] Next, referring to the accompanying drawings, an apparatus for predicting tobacco curing yield based on multi-source data fusion according to an embodiment of this application is described.

[0108] Figure 4 This is a block diagram of an apparatus for predicting tobacco curing yield based on multi-source data fusion, according to an embodiment of this application.

[0109] like Figure 4 As shown, the device 10 for predicting tobacco curing yield based on multi-source data fusion includes: a data acquisition module 100, a preprocessing module 200, a modeling module 300, and a calculation module 400.

[0110] Specifically, the acquisition module 100 is used to acquire initial multi-source data of the tobacco leaves to be baked before they are fed into the baking equipment. The initial multi-source data includes at least: image data of the tobacco leaves to be baked, moisture content data, fresh weight data and regional reference data. The preprocessing module 200 is used to preprocess and extract features from the initial multi-source data to obtain a multi-dimensional feature set for yield prediction. The modeling module 300 is used to input a multi-dimensional feature set into a preset yield prediction model to obtain the predicted yield of the tobacco leaves to be cured. The calculation module 400 is used to calculate the predicted yield of the tobacco leaves to be dried based on fresh weight data and predicted yield rate, using a preset single-batch yield prediction formula.

[0111] It should be noted that the foregoing explanation of the method embodiment for predicting tobacco curing yield based on multi-source data fusion also applies to the apparatus for predicting tobacco curing yield based on multi-source data fusion in this embodiment, and will not be repeated here.

[0112] The device for predicting tobacco curing yield based on multi-source data fusion, as proposed in this application, uses an acquisition module to collect initial multi-source data of the tobacco leaves to be cured before they are fed into the curing equipment. This initial multi-source data includes at least: image data, moisture content data, fresh weight data, and regional reference data of the tobacco leaves. A preprocessing module performs preprocessing and feature extraction on the initial multi-source data to obtain a multi-dimensional feature set for yield prediction. A modeling module inputs the multi-dimensional feature set into a preset yield prediction model to obtain the predicted yield rate of the tobacco leaves to be cured. A calculation module calculates the predicted post-curing yield of the tobacco leaves based on the fresh weight data and the predicted yield rate using a preset single-batch yield prediction formula. This solves the problems of large prediction errors, delayed prediction timing, and biased and low-precision prediction models in tobacco curing yield prediction. It achieves accurate prediction of single-batch and regional total yields before tobacco curing, providing forward-looking decision support for tobacco farmers' expected management and purchasing unit plan adjustments.

[0113] Figure 5 A schematic diagram of the structure of an electronic device provided in an embodiment of this application. The electronic device may include: The memory 501, the processor 502, and the computer program stored on the memory 501 and capable of running on the processor 502.

[0114] When the processor 502 executes the program, it implements the method for predicting tobacco curing yield based on multi-source data fusion provided in the above embodiments.

[0115] Furthermore, electronic devices also include: Communication interface 503 is used for communication between memory 501 and processor 502.

[0116] The memory 501 is used to store computer programs that can run on the processor 502.

[0117] The memory 501 may include high-speed RAM (Random Access Memory) memory, and may also include non-volatile memory, such as at least one disk storage.

[0118] If the memory 501, processor 502, and communication interface 503 are implemented independently, then the communication interface 503, memory 501, and processor 502 can be interconnected via a bus to complete communication between them. The bus can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus, or an EISA (Extended Industry Standard Architecture) bus, etc. The bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 5 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0119] Optionally, in a specific implementation, if the memory 501, processor 502, and communication interface 503 are integrated on a single chip, then the memory 501, processor 502, and communication interface 503 can communicate with each other through an internal interface.

[0120] The processor 502 may be a CPU (Central Processing Unit), an ASIC (Application Specific Integrated Circuit), or one or more integrated circuits configured to implement the embodiments of this application.

[0121] This application also provides a computer program product on which a computer program is stored. When the program is executed by a processor, it implements the above-described method for predicting tobacco curing yield based on multi-source data fusion.

[0122] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0123] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "N" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0124] Any process or method described in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or N executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0125] It should be understood that the various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, the N steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or more of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (FPGAs), field-programmable gate arrays (FPGAs), etc.

[0126] Those skilled in the art will understand that all or part of the steps of the methods described in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it includes one or a combination of the steps of the method embodiments.

[0127] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. A method for realizing tobacco baking yield prediction based on multi-source data fusion, characterized in that, The method comprises the following steps: Collecting initial multi-source data of the to-be-cured tobacco leaves before being sent to a curing device, the initial multi-source data at least comprising image data, moisture content data, fresh weight data and regional reference data of the to-be-cured tobacco leaves; Performing pre-processing and feature extraction operations on the initial multi-source data to obtain a multi-dimensional feature set for yield prediction; Inputting the multi-dimensional feature set into a preset yield prediction model to obtain a predicted output rate of the to-be-cured tobacco leaves; Based on the fresh weight data and the predicted output rate, calculating a predicted post-curing yield of the to-be-cured tobacco leaves by a preset single-batch yield prediction formula.

2. The method for realizing tobacco curing yield prediction based on multi-source data fusion according to claim 1, characterized in that, The regional reference data at least comprises historical initial data of other tobacco farmers' cured tobacco leaves in a preset region and corresponding historical actual output rates, tobacco planting area of the preset region, historical unit area yield and historical climate data of the same period.

3. The method for realizing tobacco curing yield prediction based on multi-source data fusion according to claim 2, characterized in that, After calculating the predicted post-curing yield of the to-be-cured tobacco leaves by the preset single-batch yield prediction formula, further comprising: Obtaining predicted post-curing yields of all tobacco leaf batches with collected initial multi-source data in the preset region, and summing up to obtain a first yield estimate value; Based on the tobacco planting area and the historical unit area yield, estimating the yield of tobacco leaf batches without collected initial multi-source data in the preset region by a scaling method, and summing up the estimated yield and the first yield estimate value to obtain a second yield estimate value; Based on the historical climate data of the same period, determining at least one climate impact factor, and correcting the second yield estimate value based on the climate impact factor, and outputting a final regional total post-curing yield prediction value.

4. The method for realizing tobacco curing yield prediction based on multi-source data fusion according to claim 1, characterized in that, Before inputting the multi-dimensional feature set into the preset yield prediction model, comprising: Obtaining a training sample set, the training sample set containing initial multi-source data of multiple historical batches of tobacco leaves and corresponding actual post-curing output rates of each batch; Performing feature extraction on the initial multi-source data of each training sample in the training sample set to obtain a multi-dimensional feature set of the training sample set; Based on a gradient boosting decision tree algorithm, constructing an initial model, and based on the centralized trend statistics of all actual post-curing output rates in the training sample set, initializing the initial prediction value of the initial model; Inputting the multi-dimensional feature set of the training sample set into the initial model, and taking the actual post-curing output rate as a supervision signal for training, and in the training process, constructing multiple decision trees by iteration, and each new decision tree is used to fit the residual between the last round of overall prediction results and the actual post-curing output rate; When the loss function value of the initial model on the training sample set is less than a preset threshold value, or the iteration number reaches a preset maximum value, stop training to obtain the preset yield prediction model.

5. The method for realizing tobacco curing yield prediction based on multi-source data fusion according to claim 4, characterized in that, After calculating the predicted post-curing yield of the to-be-cured tobacco leaves by the preset single-batch yield prediction formula, further comprising: Collecting actual post-curing dry tobacco weight of the to-be-cured tobacco leaves after curing; Based on the actual post-curing dry tobacco weight and the fresh weight data, calculating an actual post-curing output rate of the to-be-cured tobacco leaves; The initial multi-source data of the to-be-cured tobacco leaves and the actual post-curing yield of the to-be-cured tobacco leaves constitute a new training sample, and the new training sample is added to the training sample set; The preset incremental learning strategy is used to update the preset yield prediction model, wherein the preset incremental learning strategy independently configures training parameters, including at least one of a learning rate, an iteration number, or a loss function threshold. 6.The method of predicting tobacco curing yield based on multi-source data fusion according to claim 1, characterized in that, The preprocessing and feature extraction operation on the initial multi-source data includes: The preprocessing of the initial multi-source data includes at least one of data cleaning, format standardization, and outlier processing; The feature extraction on the preprocessed data includes: The visual features include color features and texture features of the tobacco leaves; The physical features include an index for characterizing moisture distribution uniformity and an index for characterizing tobacco leaf density. The reference features include historical yield similar to the current batch of tobacco leaves based on geographical proximity and time proximity.

7. The method for realizing tobacco curing yield prediction based on multi-source data fusion according to claim 1, characterized in that, The preset single-batch yield prediction formula is to multiply the fresh weight data by the predicted yield to calculate the predicted post-curing yield.

8. A device for realizing tobacco baking yield prediction based on multi-source data fusion, characterized in that, It includes: The acquisition module is configured to acquire initial multi-source data of to-be-cured tobacco leaves before being sent into a curing device, wherein the initial multi-source data at least includes image data, moisture content data, fresh weight data, and regional reference data of the to-be-cured tobacco leaves; The preprocessing module is configured to perform preprocessing and feature extraction operation on the initial multi-source data to obtain a multi-dimensional feature set for yield prediction; The modeling module is configured to input the multi-dimensional feature set into a preset yield prediction model to obtain a predicted yield of the to-be-cured tobacco leaves; The computing module is configured to calculate a predicted post-curing yield of the to-be-cured tobacco leaves based on the fresh weight data and the predicted yield through a preset single-batch yield prediction formula.

9. An electronic device, comprising: It includes: A memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor executes the program to implement the method for predicting tobacco curing yield based on multi-source data fusion according to any one of claims 1-7.

10. A computer program product comprising a computer program, characterized in that, The computer program is executed to implement the method for predicting tobacco curing yield based on multi-source data fusion according to any one of claims 1-7.

Images