Tobacco leaf moisture content detection system fusing multi-source information

Abstract

The invention provides a tobacco leaf moisture content detection system fusing multi-source information, comprising: a data acquisition module, the data acquisition module comprising sensors arranged at a plurality of measurement points in a curing barn and used for acquiring data of the plurality of measurement points in the curing barn; and the processing module is used for receiving and processing the data acquired by the data acquisition module to obtain the tobacco leaf moisture content of each measurement point, and the tobacco leaf moisture content is obtained by calculating the weight data after temperature correction or inputting the temperature and humidity data into a trained neural network for prediction. According to the system, the data of a plurality of measuring points in the curing barn are acquired through the data acquisition module with relatively low cost, the tobacco leaf moisture content of each measuring point is calculated by utilizing the processing module according to the data, and during calculation, on one hand, the weight data can be corrected by utilizing the temperature data, so that the accurate tobacco leaf moisture content is obtained; on the other hand, temperature and humidity data can be used for prediction, and the system is low in cost, fast in detection and high in accuracy.

Description

Technical Field

[0001] This invention relates to the field of tobacco processing technology, and in particular to a tobacco moisture content detection system that integrates multi-source information. Background Technology

[0002] During the tobacco curing process, monitoring the moisture content of the tobacco leaves in the curing barn is crucial for ensuring tobacco quality, optimizing processing efficiency, and reducing energy consumption. Tobacco leaf moisture content is a core indicator for guaranteeing tobacco quality and can be used to precisely control the curing process. It also enables energy conservation, cost optimization, and reduction of ineffective curing time, thus lowering fuel costs. Furthermore, tobacco leaves with adequate moisture content are less prone to mold or caking during storage. Therefore, monitoring the moisture content of the tobacco leaves in the curing barn is of paramount importance.

[0003] Currently, the detection of tobacco leaf moisture content mainly relies on manual testing or near-infrared moisture detection. Manual testing depends on technicians regularly sampling and measuring offline, which has the highest accuracy rate, but it is labor-intensive, inefficient, and time-consuming, resulting in a significant delay in the test results. Near-infrared moisture detectors can achieve online detection, avoiding errors caused by human experience. However, the near-infrared spectrum is easily affected by the color and stacking morphology of tobacco leaves, and the instruments are expensive and require regular calibration and maintenance. Summary of the Invention

[0004] In view of the above-mentioned deficiencies of the prior art, the present invention provides a tobacco leaf moisture content detection system that integrates multi-source information to solve the technical problems of long detection time, low detection efficiency and high cost in the prior art.

[0005] To achieve the above and other related objectives, the present invention provides a tobacco leaf moisture content detection system that integrates multi-source information, comprising: a data acquisition module, the data acquisition module including sensors installed at multiple measurement points in the curing barn for collecting data from multiple measurement points in the curing barn; and a processing module for receiving and processing the data collected by the data acquisition module to obtain the tobacco leaf moisture content at each measurement point, wherein the tobacco leaf moisture content is calculated using weight data corrected for temperature, or predicted by inputting temperature and humidity data into a trained neural network.

[0006] In one embodiment of the present invention, the data acquisition module includes: a first sensor for acquiring first temperature data at a measurement point; a second sensor for acquiring weight data at a measurement point; the processing module obtains the tobacco leaf moisture content at each measurement point based on the first temperature data, the weight data, and a preset initial moisture content of tobacco leaves and a deviation fitting relationship, wherein the deviation fitting relationship is a fitting relationship between the normalized deviation of the weighing value and the temperature value obtained based on a polynomial regression equation.

[0007] In one embodiment of the present invention, the processing module includes: a first deviation calculation unit, which calculates the normalized temperature deviation at the current moment based on the measured temperature value at the current moment and the measured temperature value at the previous moment in the first temperature data; a second deviation calculation unit, which calculates the normalized weight deviation at the current moment based on the normalized temperature deviation at the current moment and the deviation fitting relationship; a weight correction unit, which calculates the weight correction value at the current moment based on the normalized weight deviation at the current moment, the measured weight value at the current moment in the weight data, and the weight correction value at the previous moment; and a moisture content calculation unit, which calculates the moisture content of the tobacco leaves at each measurement point based on the weight correction value at the current moment, the measured weight value at the initial moment in the weight data, and the initial moisture content of the tobacco leaves.

[0008] In one embodiment of the present invention, the data acquisition module includes: a third sensor for acquiring second temperature data of the measurement point; a fourth sensor for acquiring humidity data of the measurement point; and a fifth sensor for acquiring air velocity of the measurement point. The processing module obtains the tobacco leaf moisture content of each measurement point based on the second temperature data, the humidity data, the air velocity, and the location of the measurement point. The tobacco leaf moisture content of any measurement point is predicted by a trained neural network after feature weighting of the data of that measurement point and several measurement points within its preset range.

[0009] In one embodiment of the present invention, the processing module includes: a feature extraction unit, which obtains the temperature and humidity features of each measurement point based on the second temperature data and the humidity data; an influence factor calculation unit, which obtains the influence factor of each measurement point based on the second temperature data, the air velocity, and the location; a filtering unit, which obtains multiple second measurement points based on the first measurement point to be detected and several measurement points within a preset range; a weight calculation unit, which obtains the weight of each second measurement point based on the influence factors of all the second measurement points; a first feature fusion unit, which performs feature weighting based on the temperature and humidity features and weights of all the second measurement points to obtain the feature vector of the first measurement point; and a BP neural network, which predicts the tobacco moisture content of the first measurement point based on the feature vector of the first measurement point.

[0010] In one embodiment of the present invention, the feature extraction unit includes: a preprocessing component for preprocessing the second temperature data and the humidity data, the preprocessing including data cleaning, noise reduction and normalization; and an extraction component for obtaining the temperature and humidity features of each measurement point based on the temperature and humidity statistics of each measurement point within a preset time window after preprocessing, the preset time window including the current time window and the previous time window, and the statistics including mean, variance, maximum value and minimum value.

[0011] In one embodiment of the present invention, the influence factor calculation unit includes: a first calculation component, which obtains the temperature gradient of each measurement point based on the second temperature data of each measurement point and the average temperature of the drying oven; a second calculation component, which obtains the distance between each measurement point and the ventilation opening based on the location of each measurement point; and a third calculation component, which obtains the influence factor of each measurement point according to the following formula based on the air velocity, temperature gradient, and distance to the ventilation opening of all measurement points: In the formula, F i d i v i ΔT i α, β, and γ represent the influence factor, distance from the vent, air velocity, and temperature gradient of the i-th measurement point, respectively, with α, β, and γ being preset weighting coefficients.

[0012] In one embodiment of the present invention, α∈[0.5,0.6], β∈[0.15,0.2], and γ∈[0.2,0.25].

[0013] In one embodiment of the present invention, the processing module further includes a second feature fusion unit, which obtains multi-dimensional features of the first measurement point based on the feature vector of the first measurement point and a preset baking stage code; the BP neural network predicts the tobacco moisture content of the first measurement point based on the multi-dimensional features of the first measurement point.

[0014] In one embodiment of the present invention, the input layer of the BP neural network is used to receive multi-dimensional features of the first measurement point, the hidden layer consists of three layers of neurons, and the output layer is the predicted tobacco moisture content of the first measurement point. The number of neurons in each layer of the hidden layer is 128, 64, and 32, respectively.

[0015] The beneficial effects of this invention are as follows: This invention proposes a tobacco leaf moisture content detection system that integrates multi-source information. This system collects data from multiple measurement points in the curing barn using a low-cost data acquisition module. Based on this data, a processing module calculates the tobacco leaf moisture content at each measurement point. During the calculation, on the one hand, temperature data can be used to correct the weight data to obtain an accurate tobacco leaf moisture content. On the other hand, temperature and humidity data can be used for prediction. This system is low-cost, fast, and highly accurate. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a schematic diagram of a detection system provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of a drying room provided according to an embodiment of the present invention; Figure 3 This is a schematic diagram of a first structure of a detection system provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of a second structure of the detection system provided in an embodiment of the present invention; Figure 5 This is a schematic diagram of a third structure of the detection system provided in an embodiment of the present invention.

[0018] Explanation of reference numerals in the attached figures: 100, Data acquisition module; 101, First sensor; 102, Second sensor; 103, Third sensor; 104, Fourth sensor; 105, Fifth sensor; 200, Processing module; 211, First deviation calculation unit; 212, Second deviation calculation unit; 213, Weight correction unit; 214, Moisture content calculation unit; 221, Feature extraction unit; 222, Influence factor calculation unit; 223, Screening unit; 224, Weight calculation unit; 225, First feature fusion unit; 226, Second feature fusion unit; 227, BP neural network. Detailed Implementation

[0019] The following specific embodiments illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. It should be noted that, unless otherwise specified, the following embodiments and features can be combined with each other. In addition to the specific methods, equipment, and materials used in the embodiments, based on the knowledge of the prior art and the description of the present invention by those skilled in the art, any prior art methods, equipment, and materials similar to or equivalent to the methods, equipment, and materials in the embodiments of the present invention can be used to implement the present invention.

[0020] It should be understood that the terminology used in the embodiments of this invention is for describing specific implementations and not for limiting the scope of protection of this invention. Unless otherwise defined, all technical and scientific terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art.

[0021] The structures, proportions, and sizes illustrated in the accompanying drawings are solely for illustrative purposes and to aid those skilled in the art in understanding and reading the invention. They are not intended to limit the implementation of the invention and therefore have no substantial technical significance. Any modifications to the structure, changes in proportions, or adjustments to size, without affecting the effectiveness and objectives of the invention, should still fall within the scope of the technical content disclosed herein. Furthermore, the drawings only show components relevant to the invention and are not drawn according to the actual number, shape, and size of components in practice. In actual implementation, the type, quantity, and proportion of each component can be arbitrarily changed, and the component layout may be more complex.

[0022] In the following description, numerous details are explored to provide a more thorough explanation of embodiments of the invention. However, it will be apparent to those skilled in the art that embodiments of the invention may be practiced without these specific details. In some embodiments, well-known structures and devices are shown in block diagram form rather than in detail to avoid obscuring embodiments of the invention.

[0023] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functions, and operations that may be implemented in the methods and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0024] Please see Figure 1 , Figure 1 An embodiment of the present invention provides a tobacco leaf moisture content detection system that integrates multi-source information, comprising a data acquisition module and a processing module. The data acquisition module includes sensors installed at multiple measurement points within the curing barn to collect data from these points. The processing module receives and processes the data acquired by the data acquisition module to obtain the tobacco leaf moisture content at each measurement point. The tobacco leaf moisture content is calculated using weight data corrected for temperature, or predicted by inputting temperature and humidity data into a trained neural network.

[0025] This system collects data from multiple measurement points in the curing barn using a low-cost data acquisition module. Based on this data, the processing module calculates the moisture content of the tobacco leaves at each measurement point. During the calculation, on the one hand, temperature data can be used to correct the weight data to obtain an accurate moisture content of the tobacco leaves. On the other hand, temperature and humidity data can be used for prediction. This system is low-cost, fast, and highly accurate.

[0026] Understandably, measurement points are pre-selected within the curing barn. For example, measurement points can be set up at the upper, middle, and lower sections of the curing barn, as well as at key locations near heat sources and ventilation openings, to ensure comprehensive coverage of the curing barn space and obtain representative data.

[0027] Data acquisition modules can include, for example, temperature sensors, humidity sensors, wind speed sensors, and load cells. Alternatively, sensors with multiple data acquisition functions can be selected, such as those capable of simultaneously acquiring temperature and humidity data. When choosing sensors, prioritize those with high precision and stability to ensure accurate data acquisition.

[0028] The sensor can be equipped with wireless communication capabilities, allowing the collected data to be transmitted wirelessly to the processing module, avoiding the cumbersome wiring and maintenance difficulties of wired transmission. For example, a communication module based on LoRa technology can be used to ensure stable transmission of the collected data to the processing module even in the complex environment of the drying oven.

[0029] The processing module can be a local computer, an industrial control computer, or other modules with data processing capabilities, or it can be a cloud server. In addition, a display module can be set up, which can be a computer monitor, a central screen, or a webpage displaying relevant data.

[0030] Please see Figure 2 , Figure 2 The schematic diagram of the curing barn shows that one side of the barn has an openable and closable door for easy access to and from the barn, and for putting in or taking out tobacco leaves. The other side, opposite the door, is connected to the air inlet A and air outlet B of the heat pump box. The high-temperature gas heated by the heat pump box enters the curing barn through the air inlet to cure the tobacco leaves. The cooled gas flows back into the heat pump box through the air outlet, is heated again, and then flows back into the curing barn through the air inlet to achieve circulation.

[0031] Inside the curing barn, a frame-like loading steel frame serves as the main support, and tobacco leaves to be dried are held and fixed using tobacco clamps. A weighing sensor can be installed between the tobacco clamps and the support frame to obtain the weight of the tobacco leaves held by the clamps; other sensors such as temperature sensors, humidity sensors, and wind speed sensors can be installed on the tobacco clamps or the loading steel frame.

[0032] Please see Figure 3In a specific embodiment of the present invention, the data acquisition module includes a first sensor and a second sensor, wherein the first sensor is used to acquire first temperature data at the measurement point; and the second sensor is used to acquire weight data at the measurement point. The processing module obtains the tobacco leaf moisture content at each measurement point based on the first temperature data, weight data, and a preset initial moisture content and deviation fitting relationship of the tobacco leaves. The deviation fitting relationship is a fitting relationship between the normalized deviation of the weighing value and the temperature value obtained based on a polynomial regression equation.

[0033] In this embodiment, the reason for correcting the weight data is that temperature changes can affect the accuracy of weight data acquisition. This effect is very significant and directly affects the accuracy of the weight data. Since the moisture content is calculated directly from the weight data, if uncorrected weight data is used directly, the result will be very inaccurate. Therefore, correction processing is required.

[0034] In practical setup, for example, the weight data of the tobacco leaves currently in the cigarette clip can be collected at both ends of a cigarette clip using a weighing sensor (second sensor). The corresponding measurement point can be the center of gravity of the entire cigarette clip after the tobacco leaves are clamped, or it can be the center point of the cigarette clip. For this cigarette clip, a separate temperature sensor (first sensor) can be set to collect the corresponding first temperature data. In this way, the weight data and the first temperature data are data obtained for the same measurement point (cigarette clip), and the subsequent processing module processes the data corresponding to each measurement point.

[0035] It should be noted that the first temperature data here is only used to distinguish it from the second temperature data in another embodiment; they are essentially the same temperature data.

[0036] In a specific embodiment of the present invention, the processing module includes a first deviation calculation unit, a second deviation calculation unit, a weight correction unit, and a moisture content calculation unit.

[0037] The first deviation calculation unit calculates the current measured temperature value n from the first temperature data. meas (t) and the measured temperature value n at the previous moment meas (t-1), to obtain the normalized temperature deviation Δn at the current time. norm (t), the specific calculation formula is as follows: .

[0038] The second deviation calculation unit normalizes the deviation Δn based on the current temperature. norm (t), the deviation fitting relationship f, are used to obtain the weight normalization deviation Δm at the current time. norm (t), can be expressed by the formula: Δm norm (t)=f(Δnnorm (t)); The deviation fitting relationship f can be obtained experimentally. This relationship reflects that when the ambient temperature of the weighing sensor is higher than a certain nominal temperature, the weighing value will drift at the reference point and will continue to change as the temperature continues to rise.

[0039] The weight correction unit normalizes the deviation Δm based on the weight at the current moment. norm (t), the current measured value m in the weight data obs and the weight correction value m from the previous moment cor (t-1), to obtain the weight correction value m at the current time. cor (t); its specific calculation formula is as follows: .

[0040] The moisture content calculation unit uses the current weight correction value (m) as a reference. cor (t) The initial measured weight m0 and the initial moisture content w0 of the tobacco leaves in the weight data are used to obtain the moisture content w of the tobacco leaves at each measurement point. t The moisture content mentioned here refers to the wet basis moisture content, which is the percentage of water mass to the total mass of the material (including water and dry matter).

[0041] .

[0042] In this embodiment, a direct weighing sensor is used to acquire the weight data of the tobacco leaves, thereby directly calculating the moisture content of the tobacco leaves. This addresses the weighing drift error caused by temperature changes and prolonged high temperatures. Given that the overall trend of temperature fluctuations affecting the weighing value of the pressure strain gauge weighing sensor (i.e., the deviation fitting relationship f) is known, the weight data of the weighing sensor can be corrected to obtain accurate tobacco leaf weight changes. Based on this, an accurate tobacco leaf moisture content can be obtained.

[0043] Please see Figure 4 In a specific embodiment of the present invention, the data acquisition module includes a third sensor, a fourth sensor, and a fifth sensor. The third sensor is used to acquire second temperature data at the measurement point; the fourth sensor is used to acquire humidity data at the measurement point; and the fifth sensor is used to acquire air velocity at the measurement point. The processing module obtains the tobacco leaf moisture content at each measurement point based on the second temperature data, humidity data, air velocity, and the location of the measurement point. The tobacco leaf moisture content at any measurement point is predicted using a trained neural network after feature weighting of data from that measurement point and several measurement points within a preset range. Compared to... Figure 3In the embodiment described above, instead of directly measuring the weight change of the tobacco leaves, the embodiment directly utilizes second temperature data, humidity data, air velocity, and other data, combined with a BP neural network for inference, fundamentally avoiding the situation where the weighing sensor is inaccurate.

[0044] In a specific embodiment of the present invention, the processing module includes a feature extraction unit, an influence factor calculation unit, a screening unit, a weight calculation unit, a first feature fusion unit, and a BP neural network.

[0045] In one specific embodiment of the present invention, the feature extraction unit obtains the temperature and humidity characteristics of each measurement point based on the second temperature data and humidity data. Specifically, it may include a preprocessing component and an extraction component.

[0046] The preprocessing component preprocesses the second set of temperature and humidity data, including data cleaning, noise reduction, and normalization. Data acquired through the data acquisition module may contain noise, outliers, and other interference; preprocessing yields higher-quality data. The preprocessing component can clean the data by setting reasonable thresholds to remove obviously erroneous data points; it can also remove random noise using a mean filtering algorithm; and it can normalize the data by mapping data from different ranges to the [0,1] space. After these processes, the obtained temperature and humidity characteristics will be more accurate.

[0047] The extraction component obtains the temperature and humidity characteristics of each measurement point based on the pre-processed temperature and humidity statistics within a preset time window. Since the temperature and humidity data collected by the data acquisition module is continuous time-series data, and temperature and humidity data can be collected at relatively short intervals depending on the sensor's sampling frequency, processing each collected temperature and humidity data to obtain the corresponding tobacco leaf moisture content would be inaccurate due to data fluctuations, consume significant computational resources, and be unnecessary because tobacco leaf moisture content changes slowly. Therefore, the extraction component introduces the concept of a time window, for example, using 30 minutes as a time window length. When calculating the temperature and humidity characteristics of a measurement point at a certain moment, the calculation can be performed based on the temperature and humidity collected at that measurement point at the current moment and in the 30 minutes preceding the current moment.

[0048] It's important to note that the time window and the detection frequency of tobacco leaf moisture content can be different. For example, if the tobacco leaf moisture content is measured every 10 minutes, the data used at 0:00 could be from 23:30 to 0:00, and the data used at 0:10 could be from 23:40 to 0:10. Alternatively, the time window and the detection frequency can be the same, i.e., measuring the tobacco leaf moisture content every 30 minutes. If 0:00 to 0:30 is defined as a time window, the tobacco leaf moisture content will generally be measured at 0:30.

[0049] In a specific embodiment of the present invention, the preset time window includes the current time window and the previous time window. For example, the current time window is s and the previous time window can be recorded as s-1. This can capture the changing trend of environmental parameters over time.

[0050] In a specific embodiment of the present invention, the statistics include the mean, variance, maximum value, and minimum value. In addition, combinations of other statistics may also be included. The processed temperature feature vector is denoted as T, T = (t mean,s ,t var,s ,t max,s ,t min,s ,t mean,s-1 ,t var,s-1 ,t max,s-1 ,t min,s-1 The humidity eigenvector is denoted as H, where H = (h mean,s ,h var,s ,h max,s ,h min,s ,h mean,s-1 ,h var,s-1 ,h max,s-1 ,h min,s-1 ), where mean, var, max, and min in the subscripts represent the mean, variance, maximum, and minimum values, respectively, and s and s-1 represent the current time window and the previous time window, respectively. Understandably, each measurement point can extract a temperature and humidity feature; for example, the temperature and humidity feature of the i-th measurement point can be denoted as [T i H i ].

[0051] The influence factor calculation unit obtains the influence factor for each measurement point based on the second temperature data, air velocity, and location. The reason for using an influence factor for each measurement point is that predicting tobacco leaf moisture content solely based on the temperature and humidity characteristics of each individual measurement point would not be very accurate. Only by comprehensively considering the temperature and humidity characteristics of all predicted points within a certain range and obtaining a weighted feature vector can the prediction result be more accurate.

[0052] In a specific embodiment of the present invention, the influence factor calculation unit includes a first calculation component, a second calculation component, and a third calculation component.

[0053] The first calculation component calculates the temperature gradient for each measurement point based on the second temperature data for each measurement point and the average temperature of the drying oven. The average temperature of the drying oven can be calculated from the temperatures of all measurement points. The temperature gradient is the difference between the temperature at that measurement point and the average temperature of the drying oven, expressed in °C.

[0054] The second calculation component calculates the distance between each measurement point and the vent based on the location of each measurement point. Since the measurement points are already set up, their locations are known and therefore do not need to be collected directly. These locations can be coordinates, which can be saved as preset data and retrieved when needed. Based on the coordinates of the measurement points and the vent, the distance between the measurement points and the vent can be calculated. This distance can also be pre-calculated and saved along with the coordinate information, thus eliminating the need for the second calculation component.

[0055] The third calculation component calculates the influence factor for each measurement point based on the air velocity, temperature gradient, and distance from the vent, using the following formula: In the formula, F i d i v i ΔT i α, β, and γ represent the influence factor, distance from the vent, air velocity, and temperature gradient of the i-th measurement point, respectively, with α, β, and γ being preset weighting coefficients.

[0056] In a specific embodiment of the present invention, the values ​​of α, β, and γ can be in the range of: α∈[0.5,0.6], β∈[0.15,0.2], γ∈[0.2,0.25]. More preferably, α=0.55, β=0.17, and γ=0.22.

[0057] The screening unit generates multiple second measurement points based on the first measurement point to be detected and several measurement points within its preset range. The screening unit uses three concepts: measurement point, first measurement point, and second measurement point. A measurement point refers to the location of sensors placed within the curing barn, and the data collected by these sensors constitutes the measurement point data. The first measurement point is one of the measurement points, specifically the one where the tobacco leaf moisture content needs to be calculated. The second measurement point encompasses the first measurement point and several other measurement points within its preset range; the number of second measurement points corresponding to each first measurement point may differ.

[0058] The aforementioned preset range can be, for example, a numerical value R in meters, representing several measurement points within a radius R centered on the first measurement point, which together constitute the second measurement point. The value of the radius R can be determined based on actual baking experience; for example, it can be set to 1 to 3 meters.

[0059] In this step, the "several measurement points within the preset range of the first measurement point" can refer to all measurement points within that preset range, or only a portion of them. Generally, we select all measurement points. If we want to select only some measurement points, we can use random selection or a specified rule. A specified rule could be, for example, selecting more points closer to the first measurement point and fewer points farther away. Increasing or decreasing the radius R can also affect the number of measurement points.

[0060] The weight calculation unit obtains the weight of each second measurement point based on the influence factors of all second measurement points. For example, the weight calculation unit can calculate the weight of each second measurement point using the following formula: In the formula, w i Fi and Fi are the weight and influence factor of the i-th second measurement point, respectively.

[0061] The first feature fusion unit performs feature weighting based on the temperature and humidity features and weights of all second measurement points to obtain the feature vector of the first measurement point. The first feature fusion unit can perform feature weighting, for example, according to the following formula: In the formula, w i The calculation result of the weight calculation unit, [T i H i ] represents the calculation result of the feature extraction unit, N is the total number of the second measurement points, and V is the feature vector of the first measurement point.

[0062] The BP neural network predicts the moisture content of the tobacco leaves at the first measurement point based on the feature vector of the first measurement point. For a detailed introduction to the BP neural network, please see subsequent content.

[0063] Please see Figure 5 In a specific embodiment of the present invention, the processing module further includes a second feature fusion unit, which obtains multi-dimensional features of the first measurement point based on the feature vector of the first measurement point and the preset baking stage code; the BP neural network predicts the tobacco leaf moisture content of the first measurement point based on the multi-dimensional features of the first measurement point.

[0064] The tobacco curing process is generally divided into several stages, including the yellowing stage, the color-fixing stage, and the drying stage. The influence of temperature and humidity on the moisture content of tobacco leaves varies at each stage. Therefore, in the second feature fusion unit, the curing stages are encoded. One-heat encoding can be used, for example, encoding the yellowing stage as [1,0,0], the color-fixing stage as [0,1,0], and the drying stage as [0,0,1]. The encoded curing stage information is then integrated as a dimension into the feature vector V, which already contains spatial location weights. This results in a multi-dimensional feature vector that comprehensively reflects changes in the moisture content of the tobacco leaves. Taking the color-fixing stage as an example, the final multi-dimensional feature vector is: [V,0,1,0].

[0065] In a specific embodiment of the present invention, the input layer of the BP neural network receives the feature vector or multi-dimensional features of the first measurement point, the hidden layer consists of three layers of neurons, and the output layer is the predicted tobacco moisture content of the first measurement point. By traversing each first measurement point, the tobacco moisture content of all measurement points can be calculated. The number of neurons in each layer of the hidden layer is 128, 64, and 32, respectively. Complex nonlinear relationships between features are learned through a nonlinear activation function (such as the ReLU function). This design of the hidden layer achieves a good balance between model complexity and computational efficiency, effectively learning data features.

[0066] Understandably, the BP neural network mentioned above needs to be trained before it can be used for prediction. The training process will be explained in detail below using multi-dimensional features as an example.

[0067] (1) First, construct the dataset. Based on the above steps, obtain the multi-dimensional features of the first measurement point, and then perform manual sampling or... Figure 3 The method in the corresponding embodiment obtains the moisture content of the tobacco leaves at the time corresponding to the first measurement point. This multi-dimensional feature and moisture content constitute a set of data. Multiple sets of data are obtained to form a dataset. The dataset should cover complete data records of different varieties of tobacco leaves under different curing conditions.

[0068] (2) Construct a BP neural network and use the Sparrow Search Algorithm (SSA) to optimize the initial weights and thresholds of the BP neural network. SSA simulates the foraging behavior of a sparrow population and searches for the optimal solution in the search space. During the optimization process, the initial weights and thresholds of the BP neural network are used as the positions of the sparrows. By iteratively updating the positions of the sparrows, the combination of weights and thresholds that minimizes the prediction error of the model is found, so as to improve the convergence speed and prediction accuracy of the model.

[0069] (3) The optimized SSA-BP model is trained using the dataset to obtain a trained BP neural network. During the training process, the model parameters are continuously adjusted through the backpropagation algorithm, the error between the predicted value and the true value is calculated, and then the error is backpropagated to update the weights and thresholds until the model converges, that is, the error reaches the set small threshold or the number of iterations reaches the upper limit.

[0070] In a specific embodiment of the present invention, the processing module can visualize the prediction results on the display module, allowing users to easily monitor the changes in moisture content during the tobacco curing process in real time. Heat maps can visually display the distribution of moisture content in different locations within the curing barn; contour maps clearly show the distribution of areas with similar moisture content; and 3D maps can display the changes in moisture content over time and space from multiple dimensions. Users can interact with the data through functions such as zooming and rotating. In addition to heat maps, contour maps, and 3D maps, dynamic charts and animations of moisture content curves changing over time can be combined to allow users to more clearly observe the dynamic changes in moisture content. For example, the processing module can use WebGL technology to achieve efficient rendering of 3D maps and utilize visualization libraries such as D3.js to implement interactive displays of heat maps and contour maps, enhancing the user experience.

[0071] The above embodiments are merely illustrative of the principles and effects of the present invention and are not intended to limit the invention. Any person skilled in the art can modify or alter the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or alterations made by those skilled in the art without departing from the spirit and technical concept disclosed in the present invention should still be covered by the claims of the present invention.

Claims

1. A tobacco leaf moisture content detection system integrating multi-source information, characterized in that, include: The data acquisition module includes sensors installed at multiple measuring points inside the curing oven for collecting data from these multiple measuring points. The processing module is used to receive and process the data collected by the data acquisition module to obtain the tobacco leaf moisture content at each measurement point. The tobacco leaf moisture content is calculated using weight data after temperature correction or predicted by inputting temperature and humidity data into a trained neural network. The data acquisition module includes: The first sensor is used to collect the first temperature data at the measurement point; The second sensor is used to collect weight data at the measurement point; The processing module obtains the tobacco leaf moisture content at each measurement point based on the first temperature data, the weight data, and the preset initial moisture content and deviation fitting relationship of the tobacco leaves. The deviation fitting relationship is a fitting relationship between the normalized deviation of the weighing value and the temperature value obtained based on the polynomial regression equation. The data acquisition module includes: The third sensor is used to collect the second temperature data at the measurement point; The fourth sensor is used to collect humidity data at the measurement point; The fifth sensor is used to collect the airflow velocity at the measurement point; The processing module obtains the tobacco leaf moisture content of each measurement point based on the second temperature data, the humidity data, the air velocity, and the location of the measurement point. The tobacco leaf moisture content of any measurement point is predicted by a trained neural network after feature weighting of the data of that measurement point and several measurement points within its preset range. The processing module includes: The feature extraction unit obtains the temperature and humidity features of each measurement point based on the second temperature data and the humidity data; The influence factor calculation unit calculates the influence factor for each measurement point based on the second temperature data, the air velocity, and the location. The screening unit obtains multiple second measurement points based on the first measurement point to be detected and several measurement points within its preset range; The weight calculation unit calculates the weight of each second measurement point based on the influence factors of all second measurement points. The first feature fusion unit performs feature weighting based on the temperature and humidity features and weights of all the second measurement points to obtain the feature vector of the first measurement point; A BP neural network is used to predict the moisture content of tobacco leaves at the first measurement point based on the feature vector of the first measurement point.

2. The tobacco leaf moisture content detection system integrating multi-source information according to claim 1, characterized in that, The feature extraction unit includes: A preprocessing component preprocesses the second temperature data and the humidity data, the preprocessing including data cleaning, noise reduction and normalization; The extraction component obtains the temperature and humidity characteristics of each measurement point based on the preprocessed temperature and humidity statistics within a preset time window. The preset time window includes the current time window and the previous time window, and the statistics include the mean, variance, maximum value, and minimum value.

3. The tobacco leaf moisture content detection system integrating multi-source information according to claim 1, characterized in that, The impact factor calculation unit includes: The first calculation component calculates the temperature gradient at each measurement point based on the second temperature data at each measurement point and the average temperature of the oven. The second calculation component calculates the distance between each measurement point and the vent based on the location of each measurement point. The third calculation component calculates the influence factor for each measurement point based on the air velocity, temperature gradient, and distance from the vent, using the following formula: ； In the formula, F i d i v i ΔT i α, β, and γ represent the influence factor, distance from the vent, air velocity, and temperature gradient of the i-th measurement point, respectively, with α, β, and γ being preset weighting coefficients.

4. The tobacco leaf moisture content detection system integrating multi-source information according to claim 3, characterized in that, The values ​​are α∈[0.5,0.6], β∈[0.15,0.2], and γ∈[0.2,0.25].

5. The tobacco leaf moisture content detection system integrating multi-source information according to claim 1, characterized in that, The processing module further includes a second feature fusion unit, which obtains multi-dimensional features of the first measurement point based on the feature vector of the first measurement point and the preset baking stage code; the BP neural network predicts the tobacco leaf moisture content of the first measurement point based on the multi-dimensional features of the first measurement point.

6. The tobacco leaf moisture content detection system integrating multi-source information according to claim 5, characterized in that, The input layer of the BP neural network is used to receive the multi-dimensional features of the first measurement point, the hidden layer consists of three layers of neurons, and the output layer is the predicted tobacco moisture content of the first measurement point. The number of neurons in each layer of the hidden layer is 128, 64, and 32, respectively.

Images