Cigarette steel seal identification detection method based on laser imaging

Abstract

The invention discloses a cigarette steel seal identification detection method based on laser imaging, and the method mainly comprises the steps: a deep learning-based identification detection model can automatically learn the deep features of a steel seal identification, thereby achieving the efficient and accurate positioning and recognition. Firstly, image data of laser scanning imaging of cigarettes are labeled and enhanced, and a training data set containing diversified scenes and styles is constructed to improve the generalization ability of a model. In the model training, the learning rate can be dynamically adjusted, so that the model can be quickly converged at the initial stage of training, and the detection precision is further improved. And deploying the trained model on a production line, automatically detecting the steel seal mark and evaluating the integrity and definition of the steel seal mark, thereby realizing efficient monitoring on the quality of cigarette products. The detection efficiency and precision of the steel seal marks on the cigarettes are greatly improved, the manual detection cost can be remarkably reduced, and the intelligent and automatic progress of the tobacco manufacturing process is promoted.

Description

Technical Field

[0001] The invention relates to the field of cigarette manufacturing, and in particular to a cigarette steel stamp mark detection method based on laser imaging. Background Art

[0002] In the tobacco production process, the steel-stamped logo on cigarettes is a key identification that records important data such as production batches and manufacturer information. It is the core basis for tracing product sources and preventing counterfeiting. Traditional detection methods mainly rely on manual detection and ordinary image processing technology, which has obvious shortcomings. Manual detection methods rely on staff to visually check the integrity and clarity of the steel-stamped logo. However, this method is inefficient and easily affected by subjective factors and fatigue of the inspectors, resulting in poor consistency. Although ordinary image processing technologies, such as edge detection and morphological operations, can identify steel stamps in simple environments, in actual production environments, factors such as cigarette movement, lighting changes, background clutter, and the use of special processes will greatly reduce the detection accuracy of the steel-stamped logo on cigarettes, and are prone to false detection and missed detection. Summary of the invention

[0003] In view of the above, the present invention aims to provide a cigarette stamp mark detection method based on laser imaging to solve the technical problems mentioned above.

[0004] The technical solution adopted by the present invention is as follows:

[0005] The present invention provides a method for detecting cigarette steel stamp marks based on laser imaging, which includes:

[0006] Acquire a laser scanning image of the cigarette;

[0007] After preprocessing the laser scanning image, the preprocessed laser scanning image is expanded and enhanced in combination with the scene characteristics of the actual cigarette scanning to obtain diversified laser imaging samples;

[0008] After labeling the laser imaging samples, training a marker detection model based on a deep learning architecture;

[0009] The trained logo detection model is used to perform steel stamp logo detection on the actual scanned cigarette laser imaging, and abnormal results are fed back; wherein the abnormal results at least indicate that the steel stamp logo is missing or blurred.

[0010] In at least one possible implementation manner, the preprocessing includes: removing noise from the original laser scanning image.

[0011] In at least one possible implementation, the extension and enhancement includes: injecting preset noise into the preprocessed laser scanning image to simulate the imaging error that occurs when actually scanning a cigarette.

[0012] In at least one possible implementation, the extension and enhancement also includes: simulating different viewing angles, positions and size changes when actually scanning cigarettes through random rotation, horizontal flipping, scaling, cropping and translation.

[0013] In at least one possible implementation, the extension and enhancement also includes: simulating the imaging effect when actually scanning a cigarette by adjusting the brightness, contrast, and saturation.

[0014] In at least one possible implementation, the training process of the identification detection model includes: using an optimizer and optimizing the model training through a learning rate decay strategy.

[0015] In at least one possible implementation, the labeling includes: marking the cigarette body, the imaging background, and the position and shape of the steel stamp mark.

[0016] In at least one possible implementation, obtaining the laser scanning image of the cigarette includes: performing rotational scanning from above the cigarette using a laser beam to obtain a scanning image of the surface of the cigarette.

[0017] In at least one possible implementation, the method of obtaining a laser scanning image of a cigarette specifically includes: placing the cigarette on a preset rotating mechanism, the rotating mechanism having at least two adjacently arranged unidirectional rotating guide rods, and the cigarette being located between the two guide rods, and the cigarette being rotated by the rotation of the guide rods.

[0018] In at least one possible implementation manner, the rotating mechanism is configured on a production line.

[0019] Compared with the prior art, the main design concept of the present invention is that the deep learning-based logo detection model can automatically learn the deep features of the steel stamp logo, so as to achieve efficient and accurate positioning and identification. First, the image data of the laser scanning imaging of the cigarette is annotated and enhanced, and a training data set containing diverse scenes and styles is constructed to improve the generalization ability of the model. In model training, the learning rate can also be dynamically adjusted to make the model converge quickly in the early stage of training, and fine optimization can be performed in the later stage to further improve the detection accuracy. The trained model is deployed on the production line to automatically detect the steel stamp logo and evaluate its integrity and clarity, so as to achieve efficient monitoring of the quality of cigarette products. The present invention not only greatly improves the efficiency and accuracy of detecting steel stamp logos on cigarettes, but also can significantly reduce the cost of manual inspection, and promote the intelligent and automated process of the tobacco manufacturing process. In general, the scheme of the present invention has at least the following advantages:

[0020] (1) Efficient and automated detection: The deep learning-based detection algorithm can automatically detect the steel stamp on each cigarette, significantly improving the detection efficiency. No human intervention is required during the process, which reduces labor costs and avoids erroneous detection caused by human factors.

[0021] (2) Strong robustness and adaptability to complex environments: There are various external interference factors in the on-site detection environment, such as the rapid flow of tobacco, the high-speed operation of the production line, changes in lighting conditions, changes in cigarette posture, etc. In the embodiment of the present invention, through the deep learning architecture, the deep features of the stamped logo can be accurately extracted under the influence of various complex environmental factors, and it has strong robustness.

[0022] (3) High precision and high stability: The steel-stamped logo may be difficult to be correctly identified by traditional methods due to blur, deformation, wear, special processes, etc. The detection method of the present invention can not only accurately identify the existence of the steel-stamped logo, but also judge its integrity and clarity.

[0023] (4) Dynamic learning and self-optimization: When new cigarette stamp styles appear, the detection capability can be continuously adjusted and optimized through incremental learning, avoiding the errors caused by the fixed templates of traditional algorithms that cannot adapt to new changes. This dynamic learning mechanism does not rely on fixed templates or specific styles, but can detect various types of stamp logos by learning different stamp styles, sizes and shapes, so that the present invention can cope with the diversified and complex changes in stamp logos in the long-term production process, ensuring the long-term stability and adaptability of the present invention.

[0024] (5) The present invention also uses an optimizer, which enables the model to converge quickly at a high learning rate during the training process, gradually reduce the learning rate in the later stage of training, fine-tune the network parameters, and improve the stability and accuracy of the model. This adaptive learning strategy enables the model to always maintain a high generalization ability when processing different cigarette stamp logos, reduce overfitting, and improve the practical application ability of the model.

[0025] (6) Real-time detection and data feedback: The solution of the present invention is real-time detection. During the production process, the steel stamp logo of each cigarette can be detected in real time, and the detection results can be fed back in time to help the production line make quick decisions. This instant feedback mechanism greatly improves the flexibility of the production line, can quickly respond to any quality problems, and ensures the efficiency of production and the high quality of products. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention will be further described below with reference to the accompanying drawings, in which:

[0027] Figure 1 A schematic diagram of a cigarette stamp mark detection method based on laser imaging provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0028] Embodiments of the present invention are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be interpreted as limiting the present invention.

[0029] The overall concept of the present invention can be referred to as follows: First, a laser scanning device can be used to collect rotating cigarette images in real time to ensure that the steel stamp mark is clearly visible while the cigarette is flowing stably; the collected scanned images are then subjected to data denoising, and image enhancement methods are applied to expand the image diversity. Next, the pre-processed and enhanced cigarette scanned images are annotated to ensure that the laser imaging characteristics of the area where the steel stamp mark is located can be accurately calibrated, thereby generating a training data set. The data set is used to train detection models such as CNN architectures and optimize the training process, thereby improving the convergence speed and fine optimization capabilities of the model. The trained model is deployed to perform real-time detection of cigarettes, predicting the position and shape of the steel stamp mark in each cigarette image; and when an abnormal steel stamp mark is detected, such as missing or unclear, it will actively feedback and trigger an alarm, which can assist the production line in quality control while reminding the production line operator to deal with it in time. Of course, the real-time detection data can also be stored locally or on a cloud platform for later product quality analysis and traceability, and a test report can also be provided to help optimize the production process.

[0030] The above concept will be described in detail below. Figure 1 As shown, the present invention proposes an embodiment of a cigarette stamp mark detection method based on laser imaging, which specifically includes:

[0031] Step S1, obtaining a laser scanning image of a cigarette;

[0032] The present invention adopts laser scanning technology, and uses a laser beam to perform rotational scanning from above the cigarette to obtain a scanned image of the cigarette surface. In this process, the laser beam emitted by the laser emitting device scans the cigarette surface in a vertical direction, and the laser beam interacts with the cigarette surface and is reflected back to the laser receiving device. Through rotational scanning, the surface of the cigarette can be fully covered and sufficient surface details can be collected. Compared with traditional camera shooting methods, the laser scanning image can ensure that the steel-printed logo can be clearly captured under different angles and lighting conditions. The implementation method of rotational scanning can be to place the cigarette to be tested on a pre-configured rotating mechanism, which can have at least two adjacently arranged unidirectional rotating guide rods, and the cigarette is placed between the two guide rods. The cigarette is driven to rotate by the rotation of the guide rods and relying on friction, thereby realizing the rotational scanning requirements of the laser scanner. This method can efficiently obtain a scanned image of the cigarette surface, avoid the shortcomings of traditional manual inspection or other mechanical scanning methods in efficiency and accuracy, and provide accurate raw data for subsequent steel-printed logo detection.

[0033] Step S2, after preprocessing the laser scanning image, the preprocessed laser scanning image is expanded and enhanced in combination with the scene characteristics of the actual cigarette scanning to obtain a variety of laser imaging samples;

[0034] Preprocessing and data enhancement of cigarette scanned images are key steps to improve the performance of subsequent algorithm models. First, in the data preprocessing stage, first, noise in the scanned image can be removed through denoising techniques (such as median filtering and Gaussian filtering) to retain cigarette details and edge information; second, the image size can be unified through cropping and scaling, irrelevant background can be removed, and the consistency of input data can be ensured.

[0035] Regarding data enhancement, image diversity can be increased in a variety of ways to help the model generalize better. In combination with the characteristics of cigarettes, the present invention proposes that enhancement methods may include: random rotation, translation, scaling, flipping, etc., to simulate cigarettes at different viewing angles, positions, and scales. In addition, by adjusting the brightness, contrast, saturation, and injecting noise of the image, the imaging errors under different scanning environment conditions can be simulated to enhance the robustness of the model. Through the above-mentioned preprocessing and enhancement methods, the quality and diversity of cigarette scanned images can be significantly improved, providing richer data samples for subsequent deep learning training, thereby improving the accuracy and generalization ability of the model.

[0036] Step S3, annotating the laser imaging samples and training a marker detection model based on a deep learning architecture;

[0037] Data annotation of cigarette scanned images and deep learning model training are the core links to achieve accurate cigarette recognition and classification. In the data annotation stage, it is first necessary to assign the correct label to the cigarette part in the laser scanned image. In some embodiments, the annotation method includes manually drawing a bounding box (target detection). For cigarette scanned images, the tool LabelImg can be used for annotation, and the cigarette area can be accurately calibrated in the laser imaging to ensure the quality and consistency of the training data. In addition, the background part of the image also needs to be labeled so that the model can distinguish between cigarettes and other objects or noise. In addition to the annotation of cigarettes and background, the position, concave and convex shape of the steel stamp logo on the cigarette paper are also marked.

[0038] After the labeling is completed, the deep learning model training stage begins. For the laser scanning images of cigarettes, the present invention proposes the use of deep learning models, including convolutional neural networks (CNN) and their variants (such as U-Net, Mask R-CNN, etc.). These models are good at extracting features of targets such as cigarettes, backgrounds, and steel prints from scanned images and classifying or segmenting them. During the training process, a suitable loss function (such as cross entropy loss, Dice loss, etc.) can be used to guide model learning and optimize the objective function. During training, select a suitable optimizer (such as Adam or SGD), and optimize the training process through strategies such as learning rate decay to improve the convergence speed and robustness of the model.

[0039] Finally, after multiple rounds of training and verification, the model performance is evaluated. In actual operation, the model can be evaluated by indicators such as precision, recall, and F1 score. If the model performs well, the model performance can be further improved through hyperparameter tuning, model fusion, etc., to ensure its accuracy and real-time performance in actual applications.

[0040] Step S4: Use the trained logo detection model to perform steel stamp logo detection on the actual scanned cigarette imaging, and feed back abnormal results, which at least indicate that the steel stamp logo is missing or blurred.

[0041] The data storage and analysis method after cigarette surface recognition includes several key steps. First, the identified defect information (such as missing or blurred steel stamp logo) and related imaging data are stored in real time in a local database or cloud storage system, such as a relational database (such as MySQL) or a non-relational database (such as MongoDB). The stored data includes laser imaging files, detection time, production batch, defect type and location, etc.; then data preprocessing and cleaning are performed to remove redundant or invalid records to ensure data accuracy and consistency.

[0042] In the data analysis stage, trend analysis is used to identify the frequency and time distribution of defects, and correlation analysis is used to explore the relationship between defects and production conditions. Machine learning methods can also be used for predictive analysis to predict the types and frequencies of defects that may occur in the future. At the same time, anomaly detection can also identify potential problems. In other embodiments, the analysis results can be presented in the form of visual charts or reports to help operators and quality control personnel quickly identify problems in production, and provide real-time feedback to the production system for adjusting production parameters and improving production processes. Ultimately, all analysis results and reports will be archived to facilitate subsequent quality traceability and optimization decisions, while ensuring the security and long-term effectiveness of the data.

[0043] In summary, the main design concept of the present invention is that the deep learning-based logo detection model can automatically learn the deep features of the steel stamp logo, so as to achieve efficient and accurate positioning and identification. First, the image data of the laser scanning imaging of the cigarette is annotated and enhanced, and a training data set containing diverse scenes and styles is constructed to improve the generalization ability of the model. In model training, the learning rate can also be dynamically adjusted to make the model converge quickly in the early stage of training, and fine optimization can be performed in the later stage to further improve the detection accuracy. The trained model is deployed on the production line to automatically detect the steel stamp logo and evaluate its integrity and clarity, so as to achieve efficient monitoring of the quality of cigarette products. The present invention not only greatly improves the efficiency and accuracy of detecting steel stamp logos on cigarettes, but also can significantly reduce the cost of manual inspection, and promote the intelligent and automated process of the tobacco manufacturing process.

[0044] If the expressions expressing orientation are mentioned in the embodiments of the present invention, they are relative concepts based on the embodiments. In addition, "at least one" means one or more, and "plurality" means two or more. "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can represent the existence of A alone, the existence of A and B at the same time, and the existence of B alone. Among them, A and B can be singular or plural. The character " / " generally indicates that the previous and next associated objects are in an "or" relationship. "At least one of the following" and similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one of a, b and c can be represented by: a, b, c, a and b, a and c, b and c, or a and b and c, wherein a, b, c can be single or multiple.

[0045] The above describes in detail the structure, features and effects of the present invention based on the embodiments shown in the drawings, but the above is only a preferred embodiment of the present invention. It should be noted that the technical features involved in the above embodiments and their preferred methods can be reasonably combined and matched into a variety of equivalent schemes by those skilled in the art without departing from or changing the design ideas and technical effects of the present invention; therefore, the present invention is not limited to the scope of implementation shown in the drawings, and all changes made in accordance with the concept of the present invention, or modifications to equivalent embodiments with equivalent changes, which still do not exceed the spirit covered by the specification and drawings, should be within the protection scope of the present invention.

Claims

1. A method for detecting cigarette stamp marks based on laser imaging, characterized in that: include: Acquire a laser scanning image of the cigarette; After preprocessing the laser scanning image, the preprocessed laser scanning image is expanded and enhanced in combination with the scene characteristics of the actual cigarette scanning to obtain diversified laser imaging samples; After labeling the laser imaging samples, training a marker detection model based on a deep learning architecture; The trained logo detection model is used to perform steel stamp logo detection on the actual scanned cigarette laser imaging, and abnormal results are fed back; wherein the abnormal results at least indicate that the steel stamp logo is missing or blurred.

2. The method for detecting cigarette steel stamp marks based on laser imaging according to claim 1, characterized in that: The preprocessing includes: removing noise from the original laser scanning image.

3. The method for detecting cigarette steel stamp marks based on laser imaging according to claim 2, characterized in that: The expansion and enhancement includes: injecting preset noise into the preprocessed laser scanning image to simulate the imaging error that occurs when actually scanning cigarettes.

4. The method for detecting cigarette steel stamp marks based on laser imaging according to claim 3, characterized in that: The expansion and enhancement also includes: simulating different viewing angles, positions and size changes when actually scanning cigarettes through random rotation, horizontal flipping, scaling, cropping and translation.

5. The method for detecting cigarette steel stamp marks based on laser imaging according to claim 3, characterized in that: The expansion and enhancement also includes: simulating the imaging effect when actually scanning cigarettes by adjusting the brightness, contrast and saturation.

6. The method for detecting cigarette steel stamp marks based on laser imaging according to claim 1, characterized in that: The training process of the identification detection model includes: using an optimizer and optimizing the model training through a learning rate decay strategy.

7. The method for detecting cigarette steel stamp marks based on laser imaging according to claim 1, characterized in that: The marking includes: marking the cigarette body, imaging background, and the position and shape of the steel stamp mark.

8. The method for detecting cigarette stamp marks based on laser imaging according to any one of claims 1 to 7, characterized in that: The method of obtaining the laser scanning image of the cigarette includes: performing rotational scanning from above the cigarette by means of a laser beam to obtain a scanning image of the surface of the cigarette.

9. The method for detecting cigarette steel stamp marks based on laser imaging according to claim 8, characterized in that: The method of obtaining a laser scanning image of a cigarette specifically includes: placing the cigarette on a preset rotating mechanism, wherein the rotating mechanism has at least two adjacently arranged co-rotating guide rods, and the cigarette is located between the two guide rods, and the cigarette is rotated by the rotation of the guide rods.

10. The method for detecting cigarette steel stamp marks based on laser imaging according to claim 9, characterized in that: The rotating mechanism is arranged on a production line.

Images