Intelligent green tea strip tidying processing device and method based on image recognition

Abstract

The invention discloses an intelligent green tea carding processing device based on image recognition, which comprises a light supplementing module and an image acquisition module, the light supplementing module is a light box with a single luminous surface, and is mounted above a carding machine body through a bracket, the luminous surface of the light supplementing module is vertically downward and is positioned right above a pot body of the carding machine body, and the image acquisition module is connected with the light supplementing module. A camera window is formed in the middle of the light-emitting face of the box body, the image acquisition module is a high-definition color camera and is installed in the lamp box, a lens of the high-definition color camera faces downwards and extends out of the camera window, a control cabinet is arranged beside the carding machine body, and a computing server is arranged in the carding machine body. A PLC (Programmable Logic Controller) used for receiving a control instruction of the computing server and controlling the carding machine body to work is also arranged in the carding machine body; according to the device, the image acquisition mold and the light supplementing mold are arranged above the carding machine body, and calculation of the calculation server and control of the PLC are matched, so that online and nondestructive detection of the moisture of the tea leaves in a moving state is realized, and the precision of a prediction model is high.

Description

Technical Field

[0001] This invention relates to the field of tea processing technology, specifically to an intelligent green tea processing device and method based on image recognition. Background Technology

[0002] Shaping is the core process in processing needle-shaped premium green tea. Its process parameters (such as temperature, time, and vibration frequency) directly affect the appearance, color, and taste of the tea leaves. Moisture content is the most critical quality control indicator in the shaping process. However, the traditional shaping process mainly adopts the "tea-by-eye" method, where workers observe the color and shape changes of the tea leaves with the naked eye and judge the moisture content by touch, thereby adjusting the process parameters. This method is highly subjective, and the judgment of color by human senses is easily affected by external ambient light. Different workers, and even the same worker at different times and under different environments, will have significant differences in their judgment of the quality and moisture content of the same online product. This leads to unstable product quality, low standardization, high technical requirements for workers, and high labor intensity. Moreover, traditional shaping processes mostly lack real-time online moisture monitoring methods. When detecting the moisture content of tea leaves, it is usually necessary to stop the machine and take samples for offline testing, which is time-consuming and has poor timeliness, resulting in serious delays in process adjustments. A small number of tea processing enterprises use technologies such as near-infrared spectroscopy for online moisture detection, but the equipment purchase and maintenance costs are very expensive, which is not conducive to the promotion and application of such technologies in the vast majority of small and medium-sized tea processing enterprises.

[0003] To address these issues, the present invention provides an image recognition-based intelligent tea processing device and method for green tea, which solves the aforementioned problems. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies by providing an image recognition-based intelligent tea processing device and method for sorting and processing green tea, thereby solving the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: an intelligent green tea processing device based on image recognition, comprising a supplementary lighting module and an image acquisition module. The supplementary lighting module is a single-sided light box, which is mounted above the main body of the tea processing machine via a bracket. Its light-emitting surface faces vertically downward and is located directly above the pot of the main body of the tea processing machine. A camera window is provided in the middle of the light-emitting surface of the box. The image acquisition module is a high-definition color camera, which is installed inside the light box. Its lens is set downward and protrudes from the camera window. A control cabinet is provided next to the main body of the tea processing machine, which contains a computing server and a PLC controller for receiving control commands from the computing server and controlling the operation of the main body of the tea processing machine.

[0006] As a preferred embodiment of the present invention, the light box is provided with a USB window and an operation window on both the front and rear sides, and a flip cover is mounted on the operation window by means of a hinge.

[0007] As a preferred embodiment of the present invention, a detachable camera cover is threaded onto the camera window. The camera cover is a cylindrical shell, and the lens of the high-definition color camera is located inside the cavity of the camera cover and does not contact its inner wall.

[0008] As a preferred embodiment of the present invention, the light box is made of acrylic and its internal light source is a flexible LED light source.

[0009] As a preferred embodiment of the present invention, the high-definition color camera adopts an industrial-grade camera with a resolution of not less than 1920x1080 pixels.

[0010] A method for intelligent sorting and processing of green tea based on image recognition, comprising an intelligent sorting and processing device for green tea based on image recognition, including the following steps: S1. Data Acquisition and Preprocessing: During the data processing, time-series images of tea leaves are continuously acquired through the image acquisition module. Simultaneously, manual sampling is performed at set time intervals. The actual moisture content of the tea leaves is determined using standard methods and used as the ground truth label for model training. The acquired time-moisture-image raw data is preprocessed, including outlier detection and removal using algorithms such as Isolation Forest, and missing data is filled by linear interpolation. Finally, the data is standardized, and all valid data are randomly divided into training and test sets according to the proportion.

[0011] S2, HSV color space dynamic feature extraction: The acquired RGB images are converted to the HSV (Hue, Saturation, Lightness) color space. Representative H, S, and V values ​​of the tea leaf region in each frame are extracted to construct a high-dimensional feature vector. In addition to the basic instantaneous H, S, and V values, a series of derived features that reflect dynamic color changes are calculated to construct a 14-dimensional feature vector, including but not limited to: ① Changes in hue / saturation / brightness (ΔH, ΔS, ΔV) ②Relative change rate of hue / saturation / brightness (ΔHrate, ΔSrate, ΔVrate) ③ Hue circumference difference (ΔHcircular) ④ Color space distance (d), where considering that saturation changes are most sensitive to moisture, ΔH, ΔS, and ΔV are assigned weights of 0.3, 0.4, and 0.3, respectively.

[0012] ⑤ The interaction between saturation and brightness: ISV = ΔS × ΔV

[0013] ⑥ The direction angle (θ) of the color change on the SV plane

[0014] All features are normalized to eliminate the influence of dimensions.

[0015] S3. Construction and Training of the Moisture Prediction Model: Using the 14-dimensional HSV dynamic feature vector as input and the measured moisture content as output, a machine learning algorithm is used to train the moisture prediction model. By comparing various algorithms, the Gradient Boosting Tree is selected as the core model. After parameter tuning, its optimal parameters can be set as: n_estimators=100, learning_rate=0.1, max_depth=6.

[0016] S4. Real-time moisture monitoring and intelligent control of process parameters: The trained optimal gradient boosting tree model is embedded into the software system of the computing server. During the real-time tea-leaving process, the system continuously collects tea images, extracts the 14-dimensional HSV features in real time, and inputs them into the deployed model to predict the current moisture content of the tea. The system compares the predicted moisture content with the preset optimized process curve, and the computing server generates control commands. Through the PLC, the heater and vibration motor of the tea-leaving machine are controlled to automatically adjust the pot temperature and vibration frequency, so that the process parameters always track the optimal curve. When the predicted moisture content reaches the target discharge moisture, the system automatically triggers the automatic feeding mechanism to discharge the tea. The remaining processing time can be estimated and displayed in real time.

[0017] Compared with the prior art, the present invention provides an intelligent green tea processing device and method based on image recognition, which has the following beneficial effects: 1. This image recognition-based intelligent green tea processing device, by setting up an image acquisition mold and a supplementary lighting mold above the main body of the processing machine, and in conjunction with the calculation of the computing server and the control of the PLC controller, realizes online and non-destructive detection of the moisture of tea leaves in motion. The prediction model has high accuracy, greatly shortens the detection time, and solves the problem of lagging process adjustment. 2. This image recognition-based intelligent green tea shaping and processing device, through the coordinated setup of the shaping machine body, image acquisition mold, supplementary lighting mold, PLC controller and computing server, realizes fully automated closed-loop control of the shaping process, greatly reducing reliance on manual experience and effectively ensuring the stability and consistency of product quality. 3. The core sensor of this image recognition-based intelligent green tea processing device is a high-definition color camera, which is far less expensive than near-infrared spectroscopy equipment. The entire device has a simple structure and is easy to maintain, making it particularly suitable for promotion in a wide range of small and medium-sized tea processing enterprises. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the overall structure of the present invention; Figure 2 This is a schematic diagram of the light box structure of the present invention; Figure 3 This is a schematic diagram of the camera window cover structure of the present invention.

[0019] Reference numerals in the attached diagram: 1. Body of the slicing machine; 2. Support frame; 3. Light box; 4. Camera window; 5. High-definition color camera; 6. Control cabinet; 7. USB window; 8. Operation window; 9. Flip cover; 10. Camera window cover. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Example 1: Please refer to Figures 1-3 A smart green tea processing device based on image recognition includes a supplementary lighting module and an image acquisition module. The supplementary lighting module is a single-sided light box 3, which is mounted above the tea processing machine body 1 via a bracket 2. Its emitting surface faces vertically downwards and is located directly above the pot of the tea processing machine body 1. The light box 3 is made of acrylic, and its internal light source is a flexible LED light source. Its function is to provide constant intensity illumination to eliminate ambient light interference and ensure the consistency and reliability of the captured image color. A camera window 4 is opened in the middle of the emitting surface of the box 3. The image acquisition module is a high-definition color camera 5, which is installed inside the light box 3. Its lens is set downwards and protrudes from the camera window 4. The high-definition color camera 5 is an industrial-grade camera with a resolution of not less than 1920x1080 pixels, used for solid-state... The dynamic image sequence of tea leaves in the pot is continuously acquired at fixed time intervals. The front and rear sides of the light box 3 are provided with USB window 7 and operation window 8. The function of USB window 7 is to allow the USB interface of the internal components of the light box 3 to be connected to the USB plug of the external connection cable. The size of the operation window 8 must be sufficient to allow the user to put their hand in to operate the high-definition color camera 5. A flip cover 9 is installed on the operation window 8 by hinge rotation. When the device is not in use, the camera window cover 10 can be installed on the camera window 4 by thread. The camera window cover 10 is a columnar shell made of acrylic material. After the camera window cover 10 is installed, the lens of the high-definition color camera 5 is located inside its cavity and does not contact its inner wall. The purpose of setting the camera window cover 10 is to protect the high-definition color camera 5 after the device stops working. Next to the main body 1 of the beating machine is a control cabinet 6, which houses a computing server. The computing server is equipped with moisture prediction and control system software. Its hardware configuration must meet the requirements of machine learning model operation, such as Intel i7 CPU, NVIDIA graphics card, and 32G memory. The software environment is based on machine learning libraries such as Python and scikit-learn. It also contains a PLC controller to receive control commands from the computing server, thereby precisely controlling the heater, vibrating motor, and automatic feeding mechanism of the main body 1 of the beating machine. The PLC temperature sensor and frequency sensor collect relevant data of the main body 1 of the beating machine and analyze it with the data collected by the image acquisition module. The Siemens PLC200SMART EM AT04 module and EM AM06 analog module are used to realize data transmission and control of the main body 1 of the beating machine. The Siemens touch screen is used to realize the visualization, intelligent and precise operation of the main body 1 of the beating machine.

[0022] Example 2: When using this image recognition-based intelligent green tea processing device, the first step is data acquisition and preprocessing. During the tea processing, the image acquisition module continuously acquires time-series images of the tea leaves. Simultaneously, manual sampling is performed at set time intervals, and the actual moisture content of the tea leaves is determined using standard methods as ground truth labels for model training. The acquired time-moisture-image raw data is preprocessed, including using algorithms such as Isolation Forest to detect and remove outliers, and linear interpolation to fill in missing data. Finally, the data is standardized, and all valid data are randomly divided into training and test sets according to a certain proportion.

[0023] Next, dynamic features of the HSV color space are extracted. The acquired RGB images are converted to the HSV (Hue, Saturation, Lightness) color space, and representative H, S, and V values ​​of the tea leaf area in each frame are extracted to construct a high-dimensional feature vector. In addition to the basic instantaneous H, S, and V values, a series of derived features that reflect dynamic color changes are calculated to construct a 14-dimensional feature vector, including but not limited to: ① Changes in hue / saturation / brightness (ΔH, ΔS, ΔV) ②Relative change rate of hue / saturation / brightness (ΔHrate, ΔSrate, ΔVrate) ③ Hue circumference difference (ΔHcircular) ④ Color space distance (d), where considering that saturation changes are most sensitive to moisture, ΔH, ΔS, and ΔV are assigned weights of 0.3, 0.4, and 0.3, respectively.

[0024] ⑤ The interaction between saturation and brightness: ISV = ΔS × ΔV

[0025] ⑥ The direction angle (θ) of the color change on the SV plane

[0026] All features are normalized to eliminate the influence of dimensions.

[0027] Then, a moisture prediction model is constructed and trained, using the 14-dimensional HSV dynamic feature vector as input and the measured moisture content as output. A machine learning algorithm is used to train the moisture prediction model. By comparing various algorithms, the Gradient Boosting Tree is selected as the core model. After parameter tuning, its optimal parameters are set as follows: n_estimators=100, learning_rate=0.1, max_depth=6.

[0028] Finally, the trained optimal gradient boosting tree model is embedded into the software system of the computing server. During the real-time tea-leaving process, the system continuously collects tea images, extracts the 14-dimensional HSV features in real time, and inputs them into the deployed model to predict the current moisture content of the tea. The system compares the predicted moisture content with the preset optimized process curve, and the computing server generates control commands to control the heater and vibration motor of the tea-leaving machine through the PLC. The system automatically adjusts the pot temperature and vibration frequency so that the process parameters always track the optimal curve. When the predicted moisture content reaches the target outlet moisture content, the system automatically triggers the automatic feeding mechanism to unload the tea and can estimate and display the remaining processing time in real time.

[0029] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An image recognition-based intelligent tea strip processing device for green tea, comprising a light supplementing module and an image acquisition module, characterized in that: The light supplementing module is a single-side light emitting light box (3) which is installed above the body (1) of the stripper by a support (2), the light emitting surface of which is vertically downward and is located directly above the kettle body of the body (1), a camera window (4) is formed in the middle of the light emitting surface of the light box (3), the image collecting module is a high-definition color camera (5) which is installed inside the light box (3) with its lens downward and protruding from the camera window (4), a control cabinet (6) is arranged beside the body (1) of the stripper, a computing server is arranged inside the control cabinet (6), and a PLC controller for receiving the control instruction of the computing server and controlling the operation of the body (1) of the stripper is further arranged inside the control cabinet (6).

2. The image recognition based green tea smart processing device according to claim 1, wherein: USB windows (7) and operation windows (8) are formed in the front and rear sides of the light box (3), and a flip cover (9) is hingedly rotatably arranged on the operation window (8).

3. The image recognition based green tea smart processing device according to claim 1, wherein: A detachable camera window cover (10) is threadedly arranged on the camera window (4), the camera window cover (10) is a cylindrical shell, and the lens of the high-definition color camera (5) is located inside the cavity of the camera window cover (10) without contacting the inner wall thereof.

4. The image recognition based green tea smart processing device according to claim 1, wherein: The light box (3) is an acrylic light box, and the light source inside the light box is an LED flexible light source.

5. The image recognition based green tea smart processing device according to claim 1, wherein: The high-definition color camera (5) is an industrial-grade camera with a resolution not less than 1920x1080 pixels.

6. A green tea intelligent processing method based on image recognition, comprising a green tea intelligent processing device based on image recognition according to any one of claims 1-5, characterized in that, The method comprises the following steps: S1, data acquisition and preprocessing, in the striping process, the time sequence images of tea leaves are continuously acquired by the image collecting module, at the same time, the actual moisture content of tea leaves is manually sampled at a set time interval and measured by a standard method as a true value label for model training, the collected time-moisture-image original data are preprocessed, including abnormal value detection and elimination by using an isolation forest (Isolation Forest) algorithm, linear interpolation filling for missing data, and finally data standardization, all effective data are randomly divided into a training set and a test set according to a proportion; S2, HSV color space dynamic feature extraction, the collected RGB images are converted into an HSV (hue, saturation, value) color space, the representative H, S and V values of the tea leaf region in each frame of image are extracted, a high-dimensional feature vector is constructed, in addition to the basic H, S and V instantaneous values, a series of derived features capable of reflecting color dynamic changes are calculated, a 14-dimensional feature vector is constructed, including but not limited to: ① hue / saturation / value change amount (ΔH, ΔS, ΔV) ② relative change rate of hue / saturation / value (ΔHrate, ΔSrate, ΔVrate) ③ hue circular difference (ΔHcircular) ④ color space distance (d), wherein considering that the saturation change is most sensitive to moisture, the ΔH, ΔS and ΔV are respectively given a weight of 0.3, 0.4 and 0.3; ⑤ interaction of saturation and value ISV=ΔS×ΔV ⑥ direction angle (θ) of color change on the S-V plane All features are normalized to eliminate the dimensional influence. S3, moisture prediction model construction and training, taking the 14-dimensional HSV dynamic feature vector as input and the measured moisture content as output, using machine learning algorithm to train the moisture prediction model. By comparing various algorithms, the gradient boosting tree (Gradient Boosting Tree) is preferred as the core model, and after parameter optimization, the optimal parameters can be set as: n_estimators=100, learning_rate=0.1, max_depth=6; S4, real-time moisture monitoring and intelligent control of process parameters, the trained optimal gradient boosting tree model is embedded into the software system of the calculation server, during the real-time rolling process, the system continuously collects tea images, real-time extracts the 14-dimensional HSV features, and inputs them into the deployed model, real-time predicts the moisture content of the current tea, the system compares the predicted moisture content with the preset optimal process curve, the calculation server generates control instructions, through the PLC control of the heater and vibration motor of the rolling machine, automatically adjusts the temperature and vibration frequency of the kettle, so that the process parameters always track the optimal curve, when the predicted moisture content reaches the target out-kettle moisture, the system automatically triggers the automatic unloading mechanism to unload, and can real-time estimate and display the remaining processing time.

Images