Traditional Chinese medicinal material drying process intelligent detection and regulation equipment based on machine vision

Abstract

The invention discloses intelligent detection and regulation equipment for a traditional Chinese medicinal material drying process based on machine vision. The intelligent detection and regulation equipment comprises a detection module, a control module and an execution module, the detection module comprises a machine vision unit, a sensor and a dielectric property detection unit, and can obtain drying process information in multiple dimensions; a deep fuzzy control network system of the control module is combined with a deep neural network and a fuzzy control strategy, and a fuzzy control rule base can be automatically optimized and generated; the execution module adjusts parameters of the drying machine according to the instruction and forms closed-loop control through a feedback sensor; the equipment further comprises data acquisition software, a human-computer interaction interface and remote monitoring and fault diagnosis software. The problems that in the prior art, parameter detection is not comprehensive, real-time dynamic adjustment cannot be achieved, and the control precision is low are solved, intelligent and precise control is achieved, and the drying quality and the production efficiency of traditional Chinese medicinal materials are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of Chinese medicinal material processing equipment, and in particular to intelligent detection and control equipment for a Chinese medicinal material drying process based on machine vision. Background Art

[0002] In the field of modern precision processing of traditional Chinese medicine, the optimization of "multi-stage variable process" currently relies mainly on empirical experiments, and the research dimension is limited to the changes in physical and chemical characteristics before and after processing, seriously ignoring the dynamic change information of key physical and chemical characteristic indicators in the intermediate stages of the processing process.

[0003] Currently, most methods use response surface methodology or artificial neural networks to predict the quality of dry products. Although these methods can better predict the quality of dry products, due to their open-loop control nature, they cannot dynamically adjust process parameters based on the real-time material status detection. This causes the processing parameters to often deviate from the optimal values, which directly has an adverse impact on the homogenization level and quality stability of the product.

[0004] In response to the control requirements of nonlinear systems, the fuzzy control strategy uses membership functions to achieve adaptive adjustment of process parameters, showing unique advantages in drying rate and quality balance control.

[0005] However, the formulation of traditional fuzzy rules is highly subjective and difficult to process multi-dimensional sensor data. The control breadth and accuracy need to be improved. In addition, existing equipment lacks detection of the coupling relationship between volume shrinkage and dielectric properties during the drying process of traditional Chinese medicine, resulting in the inability to accurately judge the moisture migration status inside the material.

[0006] Therefore, how to provide an intelligent detection and control device for the drying process of traditional Chinese medicine based on machine vision is an urgent problem that needs to be solved by those skilled in the art. Summary of the Invention

[0007] One purpose of the present invention is to propose an intelligent detection and control device for the drying process of traditional Chinese medicine based on machine vision. The present invention can solve the problems in the existing technology of incomplete parameter detection of the drying process of traditional Chinese medicine, inability to adjust process parameters in real time and dynamically, and low control accuracy, thereby realizing intelligent and precise control of the drying process of traditional Chinese medicine and improving the drying quality and production efficiency of traditional Chinese medicine.

[0008] According to an embodiment of the present invention, a machine vision-based intelligent detection and control device for a Chinese medicinal material drying process includes a detection module, a control module, and an execution module;

[0009] The detection module is connected to the control module via a shielded data transmission line to reduce external electromagnetic interference and ensure the accuracy and stability of data transmission. The detection module is used to obtain key information during the drying process of traditional Chinese medicine, and includes a machine vision detection unit, a sensor detection unit, and a dielectric property detection unit;

[0010] The machine vision detection unit uses an industrial-grade high-resolution camera equipped with a high-resolution image sensor, and the camera frame rate is not less than 30fps. It can collect images of Chinese medicinal materials drying in real time and clearly, and transmit the image data to the control module through a dedicated image data transmission line. The convolutional neural network algorithm based on deep learning is used to automatically extract color and texture features in the image through multiple convolution layers and pooling layers. The shooting angles of the three camera installation positions are independent of each other and cover different perspectives of the Chinese medicinal materials drying area. Stereoscopic vision principles and three-dimensional reconstruction algorithms can be used to process and analyze images from different perspectives to accurately calculate the volume shrinkage rate of the material.

[0011] The sensor detection unit includes temperature sensors, humidity sensors, and wind speed sensors evenly distributed at different locations inside the dryer. It also includes a pinhole sensor that can penetrate into the material to detect the internal temperature of the material, and a high-precision weighing sensor installed under the material tray. Each type of sensor transmits detection data to the control module via analog and digital signal transmission lines, where the analog signal transmission line adopts differential transmission to reduce noise interference during signal transmission.

[0012] The dielectric property detection unit is based on the nonlinear functional relationship between the moisture content and dielectric constant of the material. By setting a non-contact dielectric induction probe inside the dryer, the dielectric constant of the material is measured using a high-frequency excitation signal, and the moisture content inside the material is calculated using a pre-established mathematical model. At the same time, the data is transmitted to the control module through a dedicated data transmission line;

[0013] The control module includes a deep fuzzy control network system, which organically combines a deep neural network with a fuzzy control strategy. Its input layer receives the multi-dimensional data transmitted by the detection module. After data normalization, the hidden layer uses a multi-layer perceptron structure composed of at least three layers of neurons to perform deep feature extraction and learning on the data using a back propagation algorithm. The output layer outputs process parameter adjustment instructions to the execution module based on the learning results.

[0014] The execution module is connected to the control module via a high-speed control signal transmission line, and is used to adjust the temperature, humidity and wind speed of the dryer according to the instructions of the control module. The temperature adjustment device is a heating device, and the heating power is adjusted by a thyristor power regulator to achieve precise temperature control. The humidity adjustment device uses a dehumidifier and a humidifying device, and the humidity is adjusted by controlling the operating time and power of the equipment. The wind speed adjustment device adjusts the wind speed by changing the fan speed through a fan frequency controller.

[0015] Furthermore, the multi-angle mounting bracket used in the machine vision inspection unit is provided with at least three camera mounting positions. The camera shooting angles on each camera mounting position are independent of each other and cover different viewing angles of the Chinese medicinal materials drying area. The angle range between adjacent camera mounting positions is 30 degrees to 150 degrees, ensuring all-round image acquisition of the Chinese medicinal materials drying process.

[0016] Furthermore, the temperature sensor in the sensor detection unit adopts a high-precision thermocouple sensor with a measurement accuracy of ±0.5°C. The cold end of the temperature sensor adopts automatic compensation technology. The humidity sensor adopts a capacitive humidity sensor with a measurement accuracy of ±2%RH and a temperature compensation function. The wind speed sensor adopts an ultrasonic wind speed sensor with a measurement error of less than ±0.1m / s. The measurement accuracy is improved through a dual-transmit and dual-receive measurement method.

[0017] Furthermore, the dielectric property detection unit adopts a non-contact measurement method, and measures the dielectric constant of the material by setting a dielectric induction probe inside the dryer. The operating frequency range of the dielectric induction probe is 1MHz to 100MHz, which can adapt to the dielectric property detection needs of different Chinese medicinal materials.

[0018] Furthermore, the deep neural network of the deep fuzzy control network system adopts a multi-layer perceptron structure, the hidden layer contains at least three layers of neurons, the number of neurons in each layer is optimized according to the dimension of the input data, and the number of neurons is iteratively optimized using a genetic algorithm. The method for optimizing the number of hidden layer neurons of the deep fuzzy control network system includes:

[0019] The fitness function of the number of neurons and control accuracy is iteratively calculated by genetic algorithm. The fitness function is Fitness=1 / (MSE+ε), where MSE is the mean square error and ε is a minimum constant, to ensure optimal network performance.

[0020] Furthermore, the deep fuzzy control network system automatically optimizes and generates a fuzzy control rule base through learning and training of a large amount of Chinese medicinal material drying data by a deep neural network, replacing the traditional subjective fuzzy rule formulation method. The optimized fuzzy control rule base is classified, stored and called according to different types of Chinese medicinal materials. At the same time, an incremental learning algorithm is used to continuously update and optimize the fuzzy control rule base during the operation of the equipment.

[0021] Furthermore, the temperature control device, humidity control device and wind speed control device of the execution module are all provided with feedback sensors for feeding back the actual adjusted temperature, humidity and wind speed data to the control module to form a closed-loop control system. The sampling frequency of the feedback sensor is not lower than the frequency of the output instructions of the control module to ensure timely feedback of the adjustment effect.

[0022] Furthermore, it also includes data acquisition software, which is deployed in the processor of the control module and is used to collect and transmit the sensor and camera data of the detection module in real time, and calibrate and filter the data. The data acquisition software uses multi-threading technology to achieve synchronous collection of multi-channel data and uses the Kalman filter algorithm to denoise the collected data.

[0023] Furthermore, it also includes a human-computer interaction interface, which displays detection data, process parameters and equipment operating status in real time through a graphical interface, supports manual setting of initial parameters and control strategies of the drying process, and has data storage, query and report generation functions. The human-computer interaction interface adopts a touch-type operation mode, and the interface layout follows ergonomic design to improve the convenience of operation.

[0024] Furthermore, it also includes remote monitoring and fault diagnosis software, which is connected to the control module through a network communication module, supports operators to remotely access equipment, adjust parameters and control equipment through the network, can monitor the equipment operation status in real time, issue an alarm and provide fault diagnosis information when the equipment fails. The remote monitoring and fault diagnosis software adopts a B / S architecture, supports multiple users to monitor online at the same time, and uses a fault tree analysis algorithm for fault diagnosis.

[0025] The beneficial effects of the present invention are:

[0026] 1. The present invention integrates machine vision detection, multiple sensor detection and dielectric property detection to obtain information on the drying process of traditional Chinese medicine from multiple dimensions such as appearance characteristics, environmental parameters, physical properties and moisture content. Compared with traditional single detection methods, it can more comprehensively and accurately reflect the drying status of the material, providing a reliable basis for precise control.

[0027] 2. The deep fuzzy control network system of the present invention combines deep neural networks with fuzzy control strategies, automatically optimizes the fuzzy control rule base, realizes efficient processing of multidimensional data, and can dynamically adjust drying process parameters according to real-time detection data, effectively solving the problem that open-loop control cannot be adjusted in real time, improving the control accuracy of the drying process, and ensuring the homogenization level and quality stability of the product.

[0028] 3. In the present invention, a shielded data transmission line is used between the detection module and the control module, and the sensor analog signal transmission adopts a differential transmission method, which effectively reduces electromagnetic interference, ensures accurate and stable data transmission, and improves the reliability of equipment operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0030] Figure 1 This is a schematic diagram of the framework structure of a machine vision-based intelligent detection and control device for the drying process of traditional Chinese medicine proposed in the present invention. DETAILED DESCRIPTION

[0031] The present invention will now be described in further detail with reference to the accompanying drawings, which are simplified schematic diagrams that illustrate the basic structure of the present invention in a schematic manner.

[0032] An intelligent detection and control device for the drying process of traditional Chinese medicine based on machine vision, including a detection module, a control module and an execution module;

[0033] The detection module is used to obtain multi-dimensional key information during the drying process of traditional Chinese medicine, including a machine vision detection unit, a sensor detection unit, and a dielectric property detection unit;

[0034] like Figure 1 As shown, the machine vision inspection unit includes three industrial-grade high-resolution cameras, which are installed on the top and both sides of the dryer respectively. The angles between adjacent cameras are 60° and 120°, covering the top, front and rear perspectives of the material drying area. The cameras transmit the real-time collected images to the control module through the GigE interface and a dedicated image data transmission line.

[0035] When astragalus with an initial weight of 5 kg and an initial moisture content of about 28% is placed in the dryer, the camera collects images of the astragalus at intervals of 200 ms.

[0036] In the image preprocessing stage, histogram equalization and median filtering are used to remove noise. In the feature extraction stage, a convolutional neural network based on ResNet-50 is used to extract color and texture features. Color features are extracted from RGB and HSV spaces, and texture features are extracted using LBP local binary pattern. The basic formula is:

[0037]

[0038] Among them, (x c ,y c ) is the center pixel coordinate, g c is the gray value of the center pixel, g p (p=0,1,…,7) is the grayscale value of 8 pixels in the neighborhood with the center pixel as the center and a radius of 1. s(x) is the sign function. When x≥0, s(x)=1, and when x≤0, s(x)=0.

[0039] At the same time, the three-view images are reconstructed using a stereo vision algorithm to calculate the volume shrinkage of the material. Assuming that the initial volume is V0 and the current volume is V t , then the volume shrinkage rate η is calculated as:

[0040]

[0041] Through this algorithm, the volume shrinkage calculation accuracy can reach ±1.5%.

[0042] When angelica with an initial weight of 4kg and a moisture content of 30% was dried, images were also collected through the camera. After 1.5 hours of drying, it was found that the color of the angelica changed from dark brown to light brown, and the LBP characteristics and volume shrinkage rate data also changed accordingly, providing visual data support for drying process monitoring.

[0043] The dielectric property detection unit includes dielectric induction probes symmetrically arranged on both sides of the dryer material layer, which measure the material's dielectric constant ε through non-contact measurement. The material's internal moisture content w is estimated based on a pre-processed BP neural network model.

[0044] During the BP neural network training process, the mean square error (MSE) is used as the loss function, and the formula is:

[0045]

[0046] Where n is the number of samples, y i is the actual moisture content of the i-th sample, is the predicted moisture content of the i-th sample, as predicted by the BP neural network. Calculations show that the internal moisture content of the astragalus root has dropped to 15%, close to the target value. Using this model, the internal moisture content of the material can be detected with an accuracy of ±1.2%. This data is transmitted to the control module via the RS485 bus.

[0047] During the angelica drying process, when the drying time reaches 3 hours, the dielectric property detection unit detects the dielectric constant of the angelica. After calculation by the BP neural network model, the moisture content of the angelica is reduced to 18%, thereby providing real-time and accurate feedback on the drying degree of the angelica.

[0048] like Figure 1 As shown, the control module includes a deep fuzzy control network system, and its hardware core is an industrial control computer;

[0049] The specific software architecture is as follows:

[0050] Input layer data processing: During the drying process of astragalus and angelica, the control module receives 11-dimensional data output from the detection module, including color characteristics, texture characteristics, volume shrinkage, dryer environmental parameters, material internal temperature, weight, and moisture content. This data must be normalized before being input into the hidden layer. The normalization formula is:

[0051]

[0052] Where x is the raw data, μ is the mean, reflecting the central tendency of the data, and σ is the standard deviation, measuring the degree of dispersion. Through normalization, data from different ranges are uniformly mapped to a specific interval, making it easier for neural networks to process.

[0053] Deep fuzzy control network architecture: It adopts 3 fully connected hidden layers with 128, 64, and 32 neurons respectively. Nonlinear mapping is achieved through the ReLU activation function, and the output layer is a 3D vector.

[0054] The fuzzy control rule base is generated by deep neural network through genetic algorithm optimization. In the genetic algorithm, the individual fitness function Fitness is defined as:

[0055]

[0056] MSE is the mean squared error between the predicted output and the actual optimal output. A smaller MSE value indicates that the predicted result is closer to the actual value, and the corresponding individual fitness is higher. This means that individuals are more likely to be selected, crossovered, and mutated during the genetic algorithm iteration process, thereby gradually optimizing the fuzzy control rule base. The initial rule base contained 50 basic rules, and after 10,000 iterations of training, the rule accuracy increased to 92%. The network training used the Adam optimizer, and its parameter update formula is:

[0057] m t =β1m t-1 +(1-β1)g t

[0058]

[0059] Among them, m t and υ t is the first-order moment estimate and the second-order moment estimate, which are used to record the mean and uncentered variance of the gradient; β1 and β2 are the moment estimate decay rates, which are usually close to 1, g t is the gradient at the current moment; α is the learning rate, which controls the step size of parameter update; ∈ is a minimum constant used to prevent the denominator from being 0; θ t is the parameter of the network at time t, which is continuously updated through the above formula to optimize the network parameters in the direction of minimizing the loss function.

[0060] When it is detected that the astragalus is drying too fast, the control module calculates and adjusts the temperature, humidity and wind speed parameters based on the detection data through the deep fuzzy control network to ensure that the drying process meets the requirements.

[0061] Similarly, when drying angelica, the control module uses the deep fuzzy control network to calculate and dynamically adjust the drying process parameters based on the angelica drying status data, thereby ensuring the drying effect of angelica.

[0062] Closed-loop control mechanism: The feedback sensor of the execution module transmits the actual adjustment parameters back to the control module at a frequency of 10Hz. The PID algorithm dynamically adjusts the control output, forming a "detection-decision-execution-feedback" closed loop. The calculation formula of the PID algorithm control quantity υ(t) is:

[0063]

[0064] Among them, K p is the proportional coefficient, which is used to adjust the control amount proportionally according to the size of the current deviation e(t); k i K is the integral coefficient, which eliminates the steady-state error of the system by integrating the deviation e(t); d is the differential coefficient, which adjusts the control quantity according to the rate of change of the deviation e(t) to improve the dynamic performance of the system; e(t) is the deviation between the set value and the actual value at time t, that is, e(t) = set value - actual value. In this device, the proportional coefficient K p =8, integral coefficient k i =0.1, differential coefficient K d =0.2.

[0065] During the astragalus drying process, if the actual temperature value differs from the target value by more than 5%, the PID compensation control is triggered to quickly restore the temperature to the set range. During the angelica drying process, the feedback sensor monitors and feeds back data in real time to ensure that the drying environment remains stable at all times and meets the angelica drying requirements.

[0066] like Figure 1 As shown, the execution module adjusts the dryer parameters according to the control module instructions, specifically including:

[0067] Temperature control device: Electric heating element combined with thyristor power regulator, adjustment accuracy ±1℃, response time <30s.

[0068] Humidity control device: The dehumidification unit is a rotary dehumidifier with a dehumidification capacity of 60kg / h;

[0069] The humidification unit is an ultrasonic humidifier with a humidification capacity of 5kg / h and a humidity control accuracy of ±3%RH.

[0070] Wind speed adjustment device: Centrifugal fan with frequency converter, wind speed adjustment range 0.5-5m / s, accuracy ±0.05m / s.

[0071] Specifically, the thyristor power regulator adjusts the output power by controlling the trigger angle α. The relationship between the output power P and the trigger angle is:

[0072]

[0073] Among them, V in is the input voltage, I in is the input current; α is the trigger angle of the thyristor. By changing the trigger angle, the conduction time of the thyristor can be adjusted, thereby controlling the output power and achieving the adjustment of the dryer temperature.

[0074] During the preheating phase of drying astragalus, the temperature sensor reports that the actual temperature rises from room temperature to the set value of 50°C within 2 minutes. During this process, the temperature regulator precisely adjusts the heating power according to the control module's instructions to ensure a steady temperature rise. When drying angelica, the temperature regulator also adjusts the temperature in real time according to the control module's instructions to meet the requirements of the angelica at different drying stages.

[0075] The frequency converter used in the wind speed control device adjusts the fan speed n by changing the motor input power frequency f. The relationship is:

[0076]

[0077] Among them, n is the fan speed; f is the motor input power frequency; s is the slip rate, which is the ratio of the difference between the asynchronous motor speed and the synchronous speed to the synchronous speed; p is the number of motor pole pairs, which determines the synchronous speed of the motor. By changing f through the frequency converter, the fan speed n can be adjusted, thereby controlling the wind speed in the dryer.

[0078] During the constant-speed drying stage of Astragalus, the wind speed regulating device maintains a stable wind speed of 2m / s; during the drying process of Angelica sinensis, the wind speed regulating device adjusts the wind speed in real time according to the instructions of the control module to ensure that the Angelica sinensis is heated evenly during the drying process and improve the drying quality.

[0079] like Figure 1 As shown, the software system specifically includes:

[0080] Data acquisition software: uses multi-threading technology to achieve 16-channel data synchronous acquisition, integrates Kalman filtering algorithm to remove high-frequency noise, and data transmission delay is less than 50ms.

[0081] The data acquisition software is deployed in the processor of the industrial control computer and can realize data processing through the following hardware interaction methods:

[0082] The detection module is connected via a PCI-E data acquisition card built into an industrial control computer. The PCI-E data acquisition card includes 16 analog signal input channels and 8 digital signal input channels. The analog signal sensor of the sensor detection unit is connected via a shielded BNC cable, and the industrial camera of the machine vision detection unit is connected via an RJ45 Ethernet interface.

[0083] The data acquisition software realizes multi-threaded synchronous acquisition through the driver of the PCI-E data acquisition card, performs A / D conversion on the analog signal output by the sensor with a conversion accuracy of 24 bits, and performs real-time decoding on the GigE image data output by the camera;

[0084] When the Kalman filter algorithm is used to suppress sensor noise, matrix operations of the state equation and the observation equation are implemented based on the floating-point operation unit of the processor.

[0085] Specifically, in the Kalman filter algorithm, the state prediction formula is:

[0086]

[0087] The state update formula is:

[0088]

[0089] in, is the estimated value of the state at time t predicted at time t-1; is the estimated value of the state at time t-1; F t B is the state transfer matrix, which describes the transfer relationship of the system state from time t-1 to time t; t is the control matrix, reflecting the control quantity u t Impact on system status; u t is the control vector at time t; is the predicted covariance matrix; is the covariance matrix at time t-1; Q t is the process noise covariance matrix, which is used to describe the statistical characteristics of the system process noise; K t is the Kalman gain, which is used to weigh the weight of the predicted value and the measured value; H t is the observation matrix, which describes the conversion relationship from system state to observation value; t is the observed value at time t; R t is the observation noise covariance matrix, which describes the statistical characteristics of the observation noise; I is the identity matrix. In this device, the process noise covariance matrix Q = 0.01, the measurement noise covariance matrix R = 0.05, and the data transmission delay is less than 50ms.

[0090] Human-machine interface: Touchscreen display, real-time display of drying curve, material status and equipment alarm information. Supports manual setting of drying stage parameters, and has historical data query and report export functions.

[0091] Specifically, the touch display screen is connected to the graphics card output port of the industrial control computer via an HDMI interface, the screen size is 15.6 inches, the resolution is 1920×1080, and it supports 10-point capacitive touch;

[0092] The touch screen has a built-in ARM coprocessor for independently processing the acquisition and preprocessing of touch signals, and transmitting the operation instructions to the processor of the industrial control computer through the USB3.0 interface;

[0093] The data storage function is realized by the built-in 512GB solid-state hard drive of the industrial control computer, which supports real-time storage of detection data and process parameters. The storage format is a binary file. The query and report generation functions are realized through the multi-threaded scheduling of the processor, and report export supports Excel and PDF formats.

[0094] Remote monitoring and fault diagnosis software: This software utilizes a B / S architecture, enabling remote connectivity via a 4G / Ethernet module and supporting mobile app and PC access. The fault diagnosis module uses a fault tree analysis algorithm, with 50 preset fault modes, an alarm response time of less than 10 seconds, and a diagnostic accuracy rate of ≥95%.

[0095] Specifically, in fault tree analysis, the probability of the top event P(T) is calculated by the probability of the bottom event P(Xi ) calculation, if the fault tree is an AND gate structure, then:

[0096]

[0097] If it is an OR gate structure, then:

[0098]

[0099] Among them, P(T) is the probability of the top event occurring; P(X i ) is the probability of the i-th bottom event occurring; n is the number of bottom events in the AND or OR gate structure. By calculating the top event probability, we can assess the system failure risk and locate the cause of the failure based on the fault tree logic.

[0100] Working principle: First, the operator inputs the type of Chinese medicinal materials, target moisture content and drying stage division through the human-computer interaction interface, and the system automatically loads the corresponding deep fuzzy control rule library. For example, before drying Astragalus, the operator inputs the target moisture content of 12% through the human-computer interaction interface and sets the drying stage parameters. When drying Angelica, the operator inputs the target moisture content of 13% and the corresponding drying stage parameters. The system automatically loads the corresponding deep fuzzy control rule library to provide a control basis for the drying process. Subsequently, the machine vision detection unit collects images every 200ms, and the sensor detection unit and the dielectric property detection unit upload data every 100ms. The data is pre-processed by the data acquisition software and input into the deep fuzzy control network system. At this time, the control module calculates the optimal process parameters through the deep fuzzy control network according to the current material status. The instruction sending cycle is 1s. After that, the execution module completes the parameter adjustment within 10s after receiving the instruction. The feedback sensor verifies the adjustment effect in real time. If the deviation exceeds 5%, the PID compensation control is triggered. When the material weight change rate is less than 0.1% / min and the moisture content meets the standard, such as after 6.5h of drying for astragalus and 7h of drying for angelica, the control module sends a shutdown command and generates a drying process report at the same time to record the key parameters and quality index data of the entire drying process.

[0101] The above are only preferred specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with this technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solutions and inventive concepts of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. An intelligent detection and control device for the drying process of traditional Chinese medicine based on machine vision, characterized in that: It includes a detection module, a control module and an execution module; The detection module is connected to the control module via a shielded data transmission line to reduce external electromagnetic interference and ensure the accuracy and stability of data transmission. The detection module is used to obtain key information during the drying process of traditional Chinese medicine, and includes a machine vision detection unit, a sensor detection unit, and a dielectric property detection unit; The machine vision detection unit uses an industrial-grade high-resolution camera equipped with a high-resolution image sensor, and the camera frame rate is not less than 30fps. It can collect images of Chinese medicinal materials drying in real time and clearly, and transmit the image data to the control module through a dedicated image data transmission line. The convolutional neural network algorithm based on deep learning is used to automatically extract color and texture features in the image through multiple convolution layers and pooling layers. The shooting angles of the three camera installation positions are independent of each other and cover different perspectives of the Chinese medicinal materials drying area. Stereoscopic vision principles and three-dimensional reconstruction algorithms can be used to process and analyze images from different perspectives to accurately calculate the volume shrinkage rate of the material. The sensor detection unit includes temperature sensors, humidity sensors, and wind speed sensors evenly distributed at different locations inside the dryer. It also includes a pinhole sensor that can penetrate into the material to detect the internal temperature of the material, and a high-precision weighing sensor installed under the material tray. Each type of sensor transmits detection data to the control module via analog and digital signal transmission lines, where the analog signal transmission line adopts differential transmission to reduce noise interference during signal transmission. The dielectric property detection unit is based on the nonlinear functional relationship between the moisture content and dielectric constant of the material. By setting a non-contact dielectric induction probe inside the dryer, the dielectric constant of the material is measured using a high-frequency excitation signal, and the moisture content inside the material is calculated using a pre-established mathematical model. At the same time, the data is transmitted to the control module through a dedicated data transmission line; The control module includes a deep fuzzy control network system, which organically combines a deep neural network with a fuzzy control strategy. Its input layer receives the multi-dimensional data transmitted by the detection module. After data normalization, the hidden layer uses a multi-layer perceptron structure composed of at least three layers of neurons to perform deep feature extraction and learning on the data using a back propagation algorithm. The output layer outputs process parameter adjustment instructions to the execution module based on the learning results. The execution module is connected to the control module via a high-speed control signal transmission line, and is used to adjust the temperature, humidity and wind speed of the dryer according to the instructions of the control module. The temperature adjustment device is a heating device, and the heating power is adjusted by a thyristor power regulator to achieve precise temperature control. The humidity adjustment device uses a dehumidifier and a humidifying device, and the humidity is adjusted by controlling the operating time and power of the equipment. The wind speed adjustment device adjusts the wind speed by changing the fan speed through a fan frequency controller.

2. The intelligent detection and control device for the drying process of Chinese medicinal materials based on machine vision according to claim 1, characterized in that: The multi-angle mounting bracket used in the machine vision inspection unit is equipped with at least three camera mounting positions. The camera shooting angles on each camera mounting position are independent of each other and cover different viewing angles of the Chinese medicinal material drying area. The angle range between adjacent camera mounting positions is 30 degrees to 150 degrees, ensuring all-round image capture of the Chinese medicinal material drying process.

3. The intelligent detection and control device for the drying process of Chinese medicinal materials based on machine vision according to claim 1 is characterized in that: The temperature sensor in the sensor detection unit adopts a high-precision thermocouple sensor with a measurement accuracy of ±0.5°C. The cold end of the temperature sensor adopts automatic compensation technology. The humidity sensor adopts a capacitive humidity sensor with a measurement accuracy of ±2%RH and a temperature compensation function. The wind speed sensor adopts an ultrasonic wind speed sensor with a measurement error of less than ±0.1m / s. The measurement accuracy is improved through a dual-transmit and dual-receive measurement method.

4. The intelligent detection and control device for the drying process of Chinese medicinal materials based on machine vision according to claim 1, characterized in that: The dielectric property detection unit adopts a non-contact measurement method and measures the dielectric constant of the material by setting a dielectric induction probe inside the dryer. The operating frequency range of the dielectric induction probe is 1MHz to 100MHz, which can meet the dielectric property detection needs of different Chinese medicinal materials.

5. The intelligent detection and control device for the drying process of Chinese medicinal materials based on machine vision according to claim 1, characterized in that: The deep neural network of the deep fuzzy control network system adopts a multi-layer perceptron structure, and the hidden layer contains at least three layers of neurons. The number of neurons in each layer is optimized according to the dimension of the input data, and the number of neurons is iteratively optimized using a genetic algorithm. The method for optimizing the number of hidden layer neurons of the deep fuzzy control network system includes: The fitness function of the number of neurons and control accuracy is iteratively calculated by genetic algorithm. The fitness function is Fitness=1 / (MSE+ε), where MSE is the mean square error and ε is a minimum constant, to ensure optimal network performance.

6. The intelligent detection and control device for the drying process of Chinese medicinal materials based on machine vision according to claim 5, characterized in that: The deep fuzzy control network system is trained through learning of a large amount of Chinese medicinal material drying data, and the deep neural network automatically optimizes and generates a fuzzy control rule base, replacing the traditional subjective fuzzy rule formulation method. The optimized fuzzy control rule base is classified, stored and called according to different types of Chinese medicinal materials. At the same time, an incremental learning algorithm is used to continuously update and optimize the fuzzy control rule base during the operation of the equipment.

7. The intelligent detection and control device for the drying process of Chinese medicinal materials based on machine vision according to claim 1, characterized in that: The temperature control device, humidity control device and wind speed control device of the execution module are all provided with feedback sensors for feeding back the actual adjusted temperature, humidity and wind speed data to the control module to form a closed-loop control system. The sampling frequency of the feedback sensor is not lower than the frequency of the output instructions of the control module to ensure timely feedback of the adjustment effect.

8. The intelligent detection and control device for the drying process of Chinese medicinal materials based on machine vision according to claim 1, characterized in that: It also includes data acquisition software, which is deployed in the processor of the control module and is used to collect and transmit the sensor and camera data of the detection module in real time, and to calibrate and filter the data. The data acquisition software uses multi-threading technology to achieve synchronous collection of multi-channel data and uses the Kalman filter algorithm to denoise the collected data.

9. The intelligent detection and control device for the drying process of Chinese medicinal materials based on machine vision according to claim 1, characterized in that: It also includes a human-computer interaction interface, which displays detection data, process parameters and equipment operating status in real time through a graphical interface, supports manual setting of initial parameters and control strategies of the drying process, and has data storage, query and report generation functions. The human-computer interaction interface adopts a touch-type operation mode, and the interface layout follows ergonomic design to improve the convenience of operation.

10. The intelligent detection and control device for the drying process of Chinese medicinal materials based on machine vision according to claim 1, characterized in that: It also includes remote monitoring and fault diagnosis software, which is connected to the control module through a network communication module, supports operators to remotely access equipment, adjust parameters and control equipment through the network, can monitor the equipment operation status in real time, issue an alarm when the equipment fails and provide fault diagnosis information. The remote monitoring and fault diagnosis software adopts a B / S architecture, supports multiple users to monitor online at the same time, and uses a fault tree analysis algorithm for fault diagnosis.

Images