Re-dried tobacco strip grading method and system based on computer vision
By using computer vision technology and multi-sensor detection, combined with image preprocessing and grading algorithms, the problems of low efficiency and insufficient accuracy in tobacco sheet grading have been solved, realizing the automation and precision of tobacco sheet grading and improving the quality control capabilities of tobacco production lines.
Patent Information
- Application Number
- CN202511760207.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-27
- Publication Date
- 2026-02-17
AI Technical Summary
Existing methods for grading tobacco sheets rely on manual sampling, which is inefficient and subjective. Furthermore, existing automated systems have low recognition rates and high false detection rates in complex production environments, making it difficult to achieve efficient and accurate quality control.
A computer vision-based method for grading re-dried tobacco flakes is adopted. The production environment is detected by multiple sensors, and images are simultaneously acquired by moving dark chamber and light source camera. Noise reduction, enhancement, cropping and color correction preprocessing are performed. Combined with anomaly detection and grading algorithms, automated and accurate tobacco flake grading is achieved.
It improves the accuracy and consistency of tobacco grading, reduces human error and labor intensity, enhances the manageability and data utilization value of the system, and realizes full-process automation from data collection to result output.
Smart Images
Figure CN121544579A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of tobacco processing and quality control technology, specifically to a method and system for grading re-dried tobacco flakes based on computer vision. Background Technology
[0002] Quality control during the production of sheet tobacco is crucial for ensuring product consistency. Traditional methods for grading sheet tobacco rely primarily on manual sampling and unpacking for inspection, which is not only inefficient but also susceptible to human error, leading to subjectivity and inconsistency in grading results. Furthermore, with the expansion of production scale and the increase in automation, the demand for efficient and accurate sheet tobacco grading technology is growing.
[0003] In recent years, the rapid development of computer vision technology and machine learning algorithms has provided new solutions for automated tobacco grading. Image processing-based grading methods can achieve rapid and objective grading by acquiring and analyzing tobacco images. However, most existing automated grading systems rely on basic image processing techniques, such as edge detection and morphological processing. While these techniques improve grading efficiency to some extent, they are still insufficient when dealing with complex production environments and diverse anomalies. For example, the detection and handling of anomalies such as box covering, exposure, partial occlusion, and broken leaves often suffer from low recognition rates and high false detection rates in existing systems, limiting the application effectiveness and scope of automated grading systems. Summary of the Invention
[0004] This invention addresses the problems of inaccurate and inefficient grading of tobacco flakes in existing technologies by proposing a computer vision-based method and system for grading re-dried tobacco flakes. The method and system of this invention automate, improve efficiency, and enhance accuracy in tobacco flake data acquisition and grading, effectively identifying and handling anomalies, and are suitable for online quality control in tobacco production lines.
[0005] On the one hand, the present invention provides a computer vision-based method for grading re-dried tobacco flakes, characterized by comprising:
[0006] S1. Working condition detection, including detection of pressure head position, smoke box position, obstacles, and press working status;
[0007] S2. Once the operating conditions are met, start the dark box to move to the vertical position above the smoke box;
[0008] S3. Start the light source and camera to acquire a clear image of the smoke.
[0009] S4. Perform preprocessing operations such as denoising, enhancement, cropping, and color correction on the acquired images;
[0010] S5. Detect abnormal conditions, including: box cover, exposure, partial obstruction, broken leaves, etc.
[0011] S6. Input the grading algorithm into the image under normal conditions, and output the tobacco grade and matching degree information;
[0012] S7. Visualize the tobacco analysis results and store them in the database, aligned with the image data.
[0013] On the other hand, the present invention also provides a computer vision-based grading system for re-dried tobacco flakes, comprising: a working condition detection module, which detects whether the pressure head is at the top position through a limit sensor; detects whether the tobacco box has reached a designated position through a photoelectric sensor; detects whether there is an obstacle above the tobacco box through a photoelectric sensor; detects whether the press is in working condition through a pressure sensor; and transmits the detection results to a lower-level computer module; a lower-level computer module, which, after the working conditions are met, receives the signal from the working condition detection module and transmits the signal to a dark box movement module and an upper-level computer module respectively; and a dark box movement module, which, after receiving the signal transmitted by the lower-level computer module, transmits the signal to a dark box movement module and an upper-level computer module respectively. The dark box is pneumatically controlled to move horizontally above the tobacco box; the host computer module receives the signal transmitted from the slave computer and transmits the signal to the data acquisition module; the data acquisition module activates the light source and camera to acquire clear tobacco images and transmits the tobacco images to the host computer module and the data processing module; the data processing module performs preprocessing operations such as noise reduction, enhancement, cropping, and color correction on the acquired images; detects abnormal conditions, including: box covering, overexposure, partial occlusion, broken leaves, etc.; inputs a grading algorithm into normal images and outputs tobacco grade and matching degree information; visualizes the tobacco analysis results and transmits them to the host computer module.
[0014] One or more technical solutions provided by this invention have at least the following technical effects or advantages:
[0015] This invention utilizes multiple sensors to monitor the press head position, smoke box position, presence of obstacles, and press operating status in real time during the working condition detection phase, ensuring a stable and safe production environment. Subsequently, the automatic movement of the dark box and the synchronized activation of the light source and camera make the image acquisition process efficient and reliable. Preprocessing operations such as noise reduction, enhancement, cropping, and color correction on the acquired images significantly improve image quality, providing a clear and accurate data foundation for subsequent grading algorithms. This not only improves the accuracy and consistency of grading but also significantly reduces errors and labor intensity caused by human operation.
[0016] Furthermore, this invention possesses powerful anomaly detection capabilities, automatically identifying issues such as box covering, exposure, partial occlusion, and broken leaves, effectively enhancing quality control capabilities. The grading algorithm, through analysis of the grade and matching degree of normal images, ensures the scientific rigor and reliability of the tobacco grading results. Aligned storage of visual analysis results with image data facilitates subsequent data retrieval and traceability, enhancing system manageability and data utilization value. Overall, by integrating multiple modules working collaboratively, this system achieves full automation from data acquisition and processing to result output, not only improving production efficiency and product quality but also providing strong technical support for the intelligent upgrading of the tobacco industry. Attached Figure Description
[0017] Figure 1 A flowchart illustrating a computer vision-based grading method for re-dried tobacco flakes, provided as an embodiment of the present invention;
[0018] Figure 2 A schematic diagram of a computer vision-based grading system for re-dried tobacco flakes provided in an embodiment of the present invention;
[0019] Figure 3 This is a pre-processed image of tobacco leaves provided in an embodiment of the present invention;
[0020] Figure 4 This is a schematic diagram of an abnormal situation provided in an embodiment of the present invention;
[0021] Figure 5 The diagram illustrates three grades of tobacco sheets provided in this embodiment of the invention. Detailed Implementation
[0022] This invention provides a computer vision-based method and system for grading re-dried tobacco flakes. In a production line environment, this system enables efficient monitoring of tobacco flake production quality through working condition detection, automatic image data acquisition and processing, anomaly detection, and tobacco flake grading. This solves the problems of inaccuracy and low efficiency associated with existing manual sampling, unpacking, and grading methods.
[0023] Example 1: As Figure 1 As shown, a computer vision-based method for grading re-dried tobacco flakes includes:
[0024] S1. Working condition detection, including detection of pressure head position, smoke box position, obstacles, and press working status;
[0025] Preferably, S1 includes: detecting whether the pressure head is at the top position using a limit sensor; detecting whether the smoke box has reached the designated position using a photoelectric sensor; detecting whether there is an obstacle above the smoke box using a photoelectric sensor; and detecting whether the press is in working condition using a pressure sensor.
[0026] Specifically, the limit sensor is an Omron E3Z series sensor with an operating voltage of 24VDC and a detection accuracy of 0.01 mm; the photoelectric sensor is a TCRT5000 with an operating distance between 10 and 50 mm; and the pressure sensor is a Schneider PX200 with a range of 0-1000 PSI, an accuracy of ±0.1% FS, an operating voltage of 5V DC, and an output signal of 4-20mA analog current. All sensor data is transmitted to the lower-level module via an industrial Ethernet interface.
[0027] S2. Once the operating conditions are met, start the dark box to move to the vertical position above the smoke box;
[0028] Preferably, the operating conditions include the pressure head being at the top position, the smoke box reaching the designated position, no obstacles above the smoke box, and the press being in a non-working state. After the operating conditions are met, the dark box is pneumatically controlled to move horizontally above the smoke box.
[0029] Specifically, a SY2B series double-acting pneumatic cylinder is selected, with a stroke length set to 1000 mm to ensure that the dark box moves to a position aligned with the center of the smoke box, and the working pressure is set to 0.8 MPa; a VSO series proportional valve is equipped to precisely control the airflow, thereby achieving acceleration, uniform speed movement, and deceleration of the dark box to avoid vibration or impact during movement; a POS series high-precision position sensor is installed to monitor the position of the dark box in real time and feed the position information back to the PLC to achieve closed-loop control, improving the accuracy and repeatability of movement; the entire pneumatic control system adopts a high-temperature resistant and interference-resistant industrial Ethernet interface to ensure high-speed and reliable signal transmission; an emergency stop button and a pressure overload protection device are provided to enhance safety during operation.
[0030] S3. Start the light source and camera to acquire a clear image of the smoke.
[0031] Preferably, the light source is four ring-shaped strip lights installed around the inside of the top surface of the dark box to provide uniform and high-brightness illumination of the tobacco surface at a fixed tilt angle; the camera is an industrial area array image sensor. When the dark box is moved horizontally above the tobacco box, it sends a signal to the PLC. The PLC transmits the signal to the industrial control computer via network communication, and the industrial control computer controls the camera to acquire tobacco image data.
[0032] Specifically, the light source is selected as two TSD-LT70035F-W-2.0 and two TSD-LT50035F-W-2.0 light sources, with lengths of 700 mm and 500 mm respectively, and widths of 35 mm. The divergence angle is 120 degrees, the uniformity is above 85%, the input voltage is 24V, and the color temperature range is 6500-7500K. The industrial area array image sensor is selected as MER-531-20GM / CP, with a resolution of 2592×2048, a frame rate of 20FPS, and equipped with a lens model of HN-0826-20M-C1 / 1X, a lens focal length of 8 mm, and a large aperture of F2.6.
[0033] S4. Perform preprocessing operations such as denoising, enhancement, cropping, and color correction on the acquired images;
[0034] Preferably, the preprocessing operation specifically includes:
[0035] S41. The denoising process employs median filtering to reduce salt-and-pepper noise and high-frequency noise in the image;
[0036] Specifically, a kernel size of 3×3 pixels is selected. For each pixel, the median value of all pixels in its neighborhood is taken as the new value of that pixel. This eliminates isolated noise without blurring the edges, effectively reducing salt-and-pepper noise and high-frequency noise in the image, while preserving the detailed information of the smoke edges.
[0037] S42. The enhancement employs adaptive histogram equalization to enhance the contrast and details of the image;
[0038] Specifically, the parameters are set to a clip limit of 2.0 to prevent excessive enhancement from amplifying noise, and the tile grid size of the local area is set to 8×8 pixels to dynamically adjust the contrast in different image areas, thereby enhancing the overall contrast and detail of the image.
[0039] S43. The cropping is performed precisely on the image based on the preset image size and target area to remove redundant background and irrelevant parts, ensuring that the analysis area is focused on the tobacco itself;
[0040] Specifically, a centered Region of Interest (ROI) is defined, and the image output size is set to 1920×1440 pixels, covering the central part of the image, to ensure that the smoke sheet is located in the center of the cropped image.
[0041] S44. The color correction uses a white balance algorithm to adjust the color balance of the image, eliminate the influence of light source color bias on the image color, and ensure the true reproduction and consistency of image color.
[0042] Specifically, assuming the average value of all colors in the image should be close to neutral gray, the average values of the red, green, and blue channels are calculated, and the gain of each channel is adjusted to make the average values of the three channels equal, thus achieving color balance. A fine-tuning step based on a white reference target is introduced, using a pre-calibrated white board image to further correct color deviations, ensuring high consistency and accuracy of image colors. The final pre-processed image is as follows: Figure 2 As shown.
[0043] S5. Detect abnormal conditions, including: box cover, exposure, partial obstruction, broken leaves, etc.
[0044] Preferably, a labeled dataset containing abnormal situations such as box covering, exposure, partial occlusion, and broken leaves is constructed, and an anomaly detection model is trained based on a computer vision target detection model. The input image can generate bounding boxes containing abnormal areas and their corresponding category labels in real time. Different category label thresholds are set for bounding boxes. If the value exceeds the threshold, it is judged as an abnormal situation and removed without further tobacco grading.
[0045] Specifically, the abnormal situation image is as follows: Figure 4 As shown, the anomaly types were proportionally allocated as follows: 500 images of covered boxes, 500 images of exposure, 500 images of partial occlusion, and 500 images of broken leaves. The bounding boxes of the anomaly regions in each image were precisely drawn by professionals using annotation tools. The training set:validation set:test set ratio was 7:2:1. Specific model training parameters were set as follows: initial learning rate of 0.001, using the Adam Optimizer (Adaptive Moment Estimation Optimizer), batch size of 16, and 100 training epochs. Data augmentation techniques such as random scaling, rotation, flipping, and color dithering were applied to enhance the model's generalization ability and robustness. Test results showed that the anomaly detection model achieved a mean average precision (mAP) of 92% on the test set, with 95% accuracy for covered boxes, 90% for exposure, 93% for partial occlusion, and 88% for broken leaves. The model's latency in real-time detection was less than 50 milliseconds, meeting the requirements for real-time monitoring on the production line.
[0046] S6. Input the grading algorithm into the image under normal conditions, and output the tobacco grade and matching degree information;
[0047] Preferably, based on expert experience and production formula, a corresponding grade is assigned to a large number of tobacco flakes, a tobacco flake grading dataset containing different grades is constructed, and a tobacco flake grading model is obtained by training an image classification model based on computer vision. The tobacco flake grade and matching degree information are output in real time when a normal image is input.
[0048] Specifically, select three grades of tobacco: L01J-24, L12U-24, and L13U-24 (e.g. Figure 5 As shown, the images were proportionally allocated as follows: L01J-24 (1425 images), L12U-24 (1753 images), and L13U-24 (919 images). The grade of each image was determined based on expert experience and production formula. The training set to test set ratio was 9:1. The specific model training parameters were set as follows: the initial learning rate was 0.002, the Adam Optimizer was used, the batch size was set to 16, the number of training epochs was 500, and data augmentation techniques such as random scaling, rotation, flipping, and color dithering were applied to enhance the model's generalization ability and robustness. The test results showed that the anomaly detection model achieved an average accuracy of 100% on the test set, and the model's latency in real-time detection was less than 50 milliseconds, meeting the needs of real-time monitoring on the production line. The matching degree information is the prediction confidence level corresponding to each category.
[0049] S7. Visualize the tobacco analysis results and store them in the database, aligned with the image data.
[0050] Preferably, the tobacco grade and matching degree information output by the grading algorithm are presented in an intuitive chart form through a graphical user interface; a relational database management system is used to store the image data and its corresponding analysis results in a structured manner to ensure data consistency and retrieval.
[0051] Example 2: As Figure 2 As shown, this embodiment provides a computer vision-based grading system for re-dried tobacco flakes, used to execute a computer vision-based grading method for re-dried tobacco flakes, including:
[0052] The working condition detection module uses a limit sensor to detect whether the pressure head is at the top position; a photoelectric sensor to detect whether the smoke box has reached the designated position; a photoelectric sensor to detect whether there are obstacles above the smoke box; and a pressure sensor to detect whether the press is in working condition; and transmits the detection results to the lower-level computer module.
[0053] Once the operating conditions are met, the lower-level computer module receives the signal from the operating condition detection module and transmits the signal to the dark box moving module and the upper-level computer module respectively.
[0054] The dark box moving module receives the signal transmitted by the lower computer and then uses pneumatic control to move the dark box horizontally above the smoke box.
[0055] The host computer module receives the signal transmitted by the slave computer and then transmits the signal to the data acquisition module.
[0056] The data acquisition module activates the light source and camera to acquire clear images of the smoke, and transmits the smoke images to the host computer module and the data processing module.
[0057] The data processing module performs preprocessing operations such as denoising, enhancement, cropping, and color correction on the acquired images; detects abnormal conditions, including: box covering, overexposure, partial occlusion, broken leaves, etc.; inputs the grading algorithm into normal images and outputs tobacco grade and matching degree information; visualizes the tobacco analysis results and transmits them to the host computer module.
[0058] Through the foregoing detailed description of a computer vision-based method for grading re-dried tobacco flakes, those skilled in the art can clearly understand that the computer vision-based method for grading re-dried tobacco flakes in this embodiment corresponds to the method disclosed in Embodiment 3, and has corresponding functional modules and beneficial effects. For relevant details, please refer to the method section.
[0059] The specific embodiments of the present invention have been described in detail above. However, the present invention is not limited to the above embodiments. Within the scope of knowledge possessed by those skilled in the art, various changes can be made without departing from the spirit of the present invention.
Claims
1. A method for grading re-dried tobacco flakes based on computer vision, characterized in that, The application relates to a tobacco leaf grade detection system and method. S1. Working condition detection, including pressure head position, tobacco box position, obstacle, and press working state detection; S2. After the working condition is met, the dark box is moved to be vertically above the tobacco box; S3. A light source and a camera are started, and clear tobacco leaf images are collected; S4. The collected images are subjected to denoising, enhancement, cropping and color correction pretreatment operations; S5. Abnormal conditions are detected, including cover box, exposure, local shielding and broken leaf conditions; S6. Normal condition images are input into a grading algorithm, and tobacco leaf grade and matching degree information are output; S7. The tobacco leaf analysis result is visualized, and is stored in a database in alignment with the image data.
2. The method for grading of redried tobacco leaves based on computer vision according to claim 1, wherein, S1. includes detecting whether the pressure head is at the top end position through a limit sensor; detecting whether the tobacco box reaches the specified position through a photoelectric sensor; detecting whether there is an obstacle above the tobacco box through the photoelectric sensor; and detecting whether the press is in the working state through a pressure sensor.
3. The method for grading of redried tobacco leaves based on computer vision according to claim 2, wherein, The working conditions include that the pressure head is at the top end position, the tobacco box reaches the specified position, there is no obstacle above the tobacco box and the press is in the non-working state. After the working conditions are met, the dark box is moved to be above the tobacco box through pneumatic control.
4. The method for grading of redried tobacco leaves based on computer vision as claimed in claim 1 wherein, The light source is four annular strip lamps which are installed around the inside of the top surface of the dark box and irradiate the surface of the tobacco leaf at a fixed inclination angle and high brightness. The camera is an industrial area array image sensor. When the dark box is moved to be above the tobacco box, a signal is sent to a PLC. The PLC transmits a signal to an industrial computer through network communication. The industrial computer controls the camera to collect tobacco leaf image data.
5. The method for grading of redried tobacco leaves based on computer vision as claimed in claim 1 wherein, The denoising adopts median filtering to reduce the salt and pepper noise and high-frequency noise in the image. The enhancement adopts adaptive histogram equalization to enhance the contrast and details of the image. The cropping accurately crops the image according to the preset image size and target area, removes the redundant background and irrelevant part, and ensures that the analysis area is concentrated on the tobacco leaf itself. The color correction adopts a white balance algorithm to adjust the color balance of the image, eliminate the influence of light source color deviation on the image color, and ensure the true restoration and consistency of the image color.
6. The method for grading of redried tobacco leaves based on computer vision as claimed in claim 1 wherein, In S5, a labeled data set containing cover box, exposure, local shielding and broken leaf abnormal conditions is constructed, and an abnormal detection model is trained based on a computer vision target detection model. The input image can generate a boundary box containing an abnormal area and a corresponding class label in real time. The threshold value of the boundary box of different class labels is set. If the threshold value is greater than the threshold value, the abnormal condition is determined and the tobacco leaf grading is not performed.
7. The method of grading of redried tobacco leaves based on computer vision as claimed in claim 1 wherein, In S6, a large number of tobacco leaves are assigned corresponding grades according to expert experience and production formula, a tobacco leaf grading data set containing different grades is constructed, and a tobacco leaf grading model is trained based on a computer vision image classification model. The normal condition image is input in real time to output the tobacco leaf grade and matching degree information.
8. The method for grading of redried tobacco leaves based on computer vision according to claim 7, wherein, In S7, the tobacco leaf grade and matching degree information output by the grading algorithm are presented in the form of intuitive charts through a graphical user interface. A relational database management system is adopted to structure the storage of the image data and the corresponding analysis result, so that the consistency and retrievability of the data are ensured.
9. A computer vision based grading system for redried tobacco leaf, characterized in that, A computer vision-based redrying tobacco sheet grading method according to any one of claims 1-8, comprising: a working condition detection module, which detects whether the pressure head is at the top end position through a limit sensor, whether the tobacco box reaches the designated position through a photoelectric sensor, whether there is an obstacle above the tobacco box through a photoelectric sensor, whether the press is in a working state through a pressure sensor, and transmits the detection results to the lower computer module; a lower computer module, which receives the signals from the working condition detection module after the working conditions are met, and transmits the signals to the dark box moving module and the upper computer module, respectively; a dark box moving module, which moves the dark box above the tobacco box through pneumatic control after receiving the signals transmitted by the lower computer; an upper computer module, which transmits signals to the data acquisition module after receiving the signals transmitted by the lower computer; a data acquisition module, which starts the light source and the camera, collects clear sheet tobacco images, and transmits the sheet tobacco images to the upper computer module and the data processing module; a data processing module, which performs denoising, enhancement, cropping, and color correction preprocessing operations on the collected images, detects abnormal conditions including covering the box, exposure, partial obstruction, and broken leaf conditions, inputs the normal condition images into a grading algorithm, outputs sheet tobacco grade and matching degree information, visualizes the sheet tobacco analysis results, and transmits them to the upper computer module.