A method for positioning cigarette packet cardboard based on image analysis

Abstract

The application discloses a positioning method of cigarette paper based on image analysis. It relates to the technical field of image processing, and comprises the following steps: projecting a structured light pattern to a target ink area; applying pulse energy to the target ink area on the surface of the cigarette paper to cause selective photothermal excitation, so that transient thermal elastic deformation is generated; synchronously collecting structured light fringe images before and after the application of the pulse energy; based on the collected reference fringe image and the deformed fringe image, a dynamic response feature field (DRSF) is calculated to represent the transient thermal elastic deformation of the target ink area caused by photothermal excitation; the DRSF is subjected to image segmentation to identify a deformed area, and geometric features are extracted from the deformed area; and the position or attitude of the target ink area is determined according to the geometric features and system calibration parameters. By utilizing the selective absorption of energy by the ink and the physical response generated thereby, the application overcomes the limitations of traditional visual positioning under specific conditions, and can simultaneously realize early warning of microscopic defects.

Description

Technical Field

[0001] This invention relates to the field of image processing, and in particular to a method for positioning cigarette pack cardboard based on image analysis. Background Technology

[0002] As a crucial component of the modern packaging industry, cigarette pack cardboard relies heavily on precise positioning during its high-speed automated production processes, such as printing, cutting, coding, and folding. This precise positioning is essential for ensuring product quality, improving production efficiency, and reducing material waste. Traditional cigarette pack cardboard positioning methods primarily depend on visible light-based machine vision technology, such as template matching and feature point extraction algorithms that identify printed marks, edge contours, or specific patterns. However, these methods often face numerous challenges in practical industrial applications: the printed patterns on cigarette pack cardboard are becoming increasingly complex, and color contrast may be low; lighting conditions in the production environment are prone to fluctuation; and the cardboard surface may have minor scratches, stains, or reflections due to lamination. These factors can all interfere with the stable extraction of visual features, leading to decreased positioning accuracy or even positioning failure. Furthermore, the printing quality of the cigarette pack cardboard directly affects the appearance and function of the final product. Conventional machine vision defect detection mainly focuses on macroscopic defects visible on the surface. For some microscopic, hidden or early defects involving the interaction between ink and paper substrate, such as tiny breaks, pinholes, uneven thickness inside the ink layer, or potential printing quality hazards caused by microscopic inhomogeneity of the paper substrate, existing technologies often cannot effectively and quickly detect them online.

[0003] Although existing technologies utilize photothermal effects for material analysis or non-destructive testing, most of these methods are designed for the steady-state or quasi-steady-state thermal properties of bulk materials. Their equipment, principles, testing speed, and accuracy requirements differ significantly from the specific needs of online high-speed positioning and micro-defect detection for thin, fast-moving printed materials such as cigarette pack paper. Furthermore, they fail to reveal the technological potential of using the transient physical response induced by selective photothermal excitation for both high-precision positioning and early warning of micro-defects. Summary of the Invention

[0004] This invention provides a positioning method for cigarette pack cardboard based on image analysis to solve the technical problems of insufficient positioning robustness and accuracy in the face of complex industrial environments such as low contrast, complex background, light fluctuation or surface contamination in the prior art.

[0005] In view of the above problems, the present invention provides a method for locating cigarette pack cardboard based on image analysis, the method comprising the following steps:

[0006] S100: Project a structured light pattern onto the target ink area on the surface of the cigarette pack cardboard;

[0007] S200: Obtain a first structured light stripe image of the target ink area under the illumination of the structured light pattern as a reference stripe image;

[0008] S300: Apply pulse energy to the target ink area on the surface of the cigarette pack card to induce selective photothermal excitation, thereby generating transient thermoelastic deformation;

[0009] S400: At a preset response time of the transient thermoelastic deformation of the target ink region caused by the selective photothermal excitation, a second structured light stripe image of the target ink region under the illumination of the structured light pattern is acquired as a deformation stripe image.

[0010] S500: Based on the reference stripe image and the deformation stripe image, calculate the dynamic response feature field characterizing the transient thermoelastic deformation of the target ink region caused by photothermal excitation;

[0011] S600 performs image segmentation on the dynamic response feature field to identify the deformation region corresponding to the target ink region;

[0012] S700: Extract at least one geometric feature from the identified deformed region;

[0013] S800: Based on the at least one geometric feature and in conjunction with preset system calibration parameters, determine the position or orientation of the target ink area on the cigarette pack cardboard.

[0014] Preferably, the method further includes the following steps:

[0015] S900: Based on the dynamic response feature field calculated in step S500, analyze the abnormal features in the dynamic response feature field that are inconsistent with the response mode of the preset target ink area in a defect-free state.

[0016] S910: Based on the abnormal features obtained from the analysis in step S900, identify the microscopic defects on the cigarette pack cardboard related to the printing quality of the target ink, the interface state between the target ink and the cigarette pack cardboard substrate, or the cigarette pack cardboard substrate itself, and generate early warning information.

[0017] Preferably, the analysis of the anomalous features of the dynamic response feature field in step S900 includes at least one of the following operations:

[0018] The currently acquired dynamic response feature field is registered and compared with the reference dynamic response feature field template of the response mode of the target ink area in a defect-free state, and the area where the difference between the two exceeds the preset tolerance threshold is identified as the abnormal feature.

[0019] Within the dynamic response feature field, the signal amplitude that characterizes the phase change or the converted physical deformation caused by the transient thermoelastic deformation is detected, and local regions whose signal amplitude is outside the statistical average of its neighboring pixels or outside the predefined normal value range are defined as the abnormal features.

[0020] Calculate the spatial gradient distribution of the dynamic response feature field, and define the regions where the gradient magnitude exceeds a preset gradient threshold, the regions where the gradient direction change rate exceeds a preset direction change rate threshold, or the regions where a breakpoint or jump point is detected in the gradient distribution as the abnormal features.

[0021] For ink regions in the dynamic response feature field that are expected to exhibit spatially continuous or uniform responses, the actual spatial continuity or uniformity index of their signals is evaluated, and the detected breaks or discontinuities, or regions with uniformity indices lower than preset standard values, are defined as the abnormal features.

[0022] The technical solution provided in this application has at least the following technical effects or advantages:

[0023] This invention detects the external manifestation of transient thermoelastic deformation (DRSF) resulting from the selective absorption of pulsed energy by a target ink region. Since DRSF primarily depends on the physical response characteristics of the ink and its interaction with the substrate, this invention demonstrates higher positioning robustness and potential accuracy improvements in situations difficult to handle with traditional vision methods, such as low-contrast printing, complex background texture interference, fluctuating ambient lighting conditions, partial transparent stain coverage, and the reflective effects of specific ink materials. Particularly for small or irregularly shaped ink features, DRSF-based analysis promises to achieve more accurate edge delineation and pose determination.

[0024] This invention can detect microscopic defects that are difficult to find or require complex post-processing algorithms to infer indirectly by deeply analyzing the local or global anomaly features of DRSF while performing localization.

[0025] Since DRSF (Different Radiant Fluid Sampling) is a direct reflection of the physical response of ink materials after selective energy absorption, it is not only related to the printing quality of the ink itself but may also be subtly affected by upstream processes such as ink formulation, printing parameters, and paper characteristics. Therefore, through long-term, continuous monitoring and statistical analysis of DRSF, "byproduct" data regarding production process stability can be indirectly obtained. For example, a systematic drift in the average DRSF response intensity may indicate batch variations in ink or inappropriate printing pressure; an increased frequency of specific defect patterns in DRSF may suggest that printing equipment requires maintenance or adjustment. This information can provide valuable reference for continuous optimization of the production process and preventative maintenance.

[0026] This invention uses optical methods for excitation and measurement, which is a non-contact detection method that will not cause physical damage to the surface of the cigarette pack paper, making it suitable for products with high surface quality requirements. Attached Figure Description

[0027] Figure 1 This is a flowchart of a cigarette pack card positioning method based on image analysis according to the present invention. Detailed Implementation

[0028] This invention discloses a positioning method for cigarette pack cardboard based on image analysis. The method selectively excites the ink area on the cigarette pack cardboard by photothermal stimulation and captures its transient thermoelastic deformation by combining structured light measurement technology, thereby constructing a dynamic response feature field and achieving precise positioning of the target ink area based on this field.

[0029] The above technical solutions will now be described in detail with reference to the accompanying drawings and specific embodiments to provide a better understanding of them. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. It should be understood that the present invention is not limited to the exemplary embodiments used only to explain the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention. Furthermore, it should be noted that, for ease of description, only the parts related to the present invention are shown in the drawings, not all of them.

[0030] In one specific embodiment, the image analysis-based positioning method for cigarette pack cardboard provided by the present invention should include the following configuration:

[0031] Pulsed energy source unit: Used for selective photothermal excitation of the target ink area. For example, it can be a near-infrared pulsed laser or a high-power pulsed light-emitting diode (LED) array. This unit features a tunable wavelength, an adjustable pulse width (e.g., from 1 nanosecond to 100 nanoseconds), and an adjustable pulse energy (e.g., from 0.01 joules per square centimeter to 1 joule per square centimeter, sufficient to induce measurable thermoelastic deformation in the target ink, but well below its ablation threshold). The unit is also equipped with appropriate optical components, such as lenses, mirrors, scanning galvanometers, or fiber optic couplers, for precisely focusing or directing the pulse energy to the target ink area on the cigarette pack cardboard surface.

[0032] Structured light projection unit: Used to project structured light patterns onto a target area. For example, it can be a digital light processing (DLP) projector or a laser interference-based fringe projection device. This unit can stably project structured light patterns with preset spatial frequencies and modulation, typically high-contrast sinusoidal fringe patterns. Projection parameters, such as fringe period and projection angle, can be configured according to measurement requirements.

[0033] High-speed image acquisition unit: Used to capture structured light patterns. For example, it can be a high-speed, high-resolution complementary metal-oxide-semiconductor (CMOS) camera or a scientific-grade CMOS (sCMOS) camera with a global shutter. The camera is equipped with an imaging objective (such as a telecentric lens) with a suitable working distance and magnification, capable of clearly capturing the striped image projected by the structured light projection unit and modulated on the surface of the cigarette pack card at a sufficiently high frame rate (e.g., thousands of hertz or even higher).

[0034] Synchronization control unit: Used to coordinate the operating timing of various units in the system. For example, it can be a high-precision programmable digital delay generator (DDG) or a control board based on a field-programmable gate array (FPGA). This unit is responsible for precisely coordinating the excitation timing of the pulse energy source unit, the pattern refresh of the structured light projection unit (if multi-frame phase shift technology is used), and the exposure acquisition timing of the high-speed image acquisition unit, ensuring that the operation of each unit achieves synchronization accuracy at the sub-microsecond or even nanosecond level.

[0035] Data processing unit: Used to process the acquired image data and execute positioning algorithms. For example, it can be an industrial computer or embedded processing system configured with a high-performance central processing unit (CPU), graphics processing unit (GPU), and sufficient memory and storage space. This unit comes pre-installed or is configured with dedicated algorithm modules.

[0036] like Figure 1 A flowchart of a method for locating cigarette pack cardstock based on image analysis is shown, including the following steps:

[0037] S100: Project a structured light pattern onto the target ink area;

[0038] Specifically, the structured light projection unit stably projects a sinusoidal stripe pattern with a preset spatial frequency and high contrast onto the target ink area and its surrounding area on the surface of the cigarette pack cardboard. Parameters such as the period, direction, and projection angle of the stripes are configured according to measurement accuracy and field of view requirements.

[0039] S200: Before the pulse energy is applied, acquire a first structured light stripe image of the target ink region as a reference stripe image;

[0040] Specifically, under the command of the synchronization control unit, the high-speed image acquisition unit captures a frame of digital image at a reference moment before the pulse energy source applies the excitation pulse. This image records the original modulation state of the structured light pattern in the target ink area when it is not excited, i.e., the reference stripe image.

[0041] S300: Apply pulse energy to the target ink area on the surface of the cigarette pack card to induce selective photothermal excitation; further, the applied pulse energy originates from a near-infrared pulsed laser or a high-power pulsed light-emitting diode (LED) array.

[0042] Specifically, the pulse width of the pulse energy source unit, for example, is controlled between 1 nanosecond and 100 nanoseconds, and the pulse energy, for example, is controlled between 0.01 joules and 1 joule per square centimeter, is precisely adjusted to ensure that the energy is sufficient to induce thermoelastic deformation of the target ink that can be clearly measured by subsequent optical systems, while remaining well below the ablation threshold of the ink material to avoid damage to the cigarette pack paper. This selective excitation utilizes the differences in light absorption characteristics and thermophysical properties between the target ink and the surrounding paper substrate at specific wavelengths.

[0043] S400: After the pulse energy is applied, a second structured light stripe image of the target ink region is acquired as a deformation stripe image at a preset time when transient thermoelastic deformation occurs in the target ink region; further, the preset time for acquiring the deformation stripe image is a time when the deformation reaches its peak or has significant characteristics, which is estimated in advance based on the thermophysical properties of the target ink material (such as thermal conductivity, heat capacity, coefficient of thermal expansion) and the parameters of the applied excitation pulse (such as energy density, pulse width), or a specific time within the optimal detection time window determined by experimental calibration.

[0044] Specifically, the synchronization control unit precisely coordinates to ensure that after the pulsed energy source applies the excitation pulse, the high-speed image acquisition unit captures another frame of digital image, i.e., the deformation fringe image, at the aforementioned preset deformation moment. The time interval between the acquisition moment of the reference fringe image and the acquisition moment of the deformation fringe image (i.e., the time interval obtained by subtracting the reference moment from the deformation moment) is controlled within an extremely short range (e.g., several microseconds to hundreds of microseconds). This time interval needs to be long enough for the thermoelastic deformation to be clearly identified and quantified by the high-speed image acquisition unit and subsequent DRSF calculation steps, but it also needs to be as short as possible to effectively capture this highly transient response and minimize the impact of low-frequency environmental vibrations (such as the vibration of a mechanical platform) on the measurement results.

[0045] S500: Based on the reference stripe image and the deformation stripe image, calculate the dynamic response feature field DRSF, which characterizes the transient thermoelastic deformation of the target ink region caused by photothermal excitation; further, the process of calculating the dynamic response feature field DRSF can adopt the principle of single-frame or multi-frame Fourier transform profilometry (FTP).

[0046] Specifically, taking single-frame Fourier transform profilometry as an example, the calculation steps include:

[0047] S510: Perform two-dimensional fast Fourier transform (2D-FFT) on the preprocessed reference fringe image and deformed fringe image respectively to transform the image from the spatial domain to the frequency domain, thereby obtaining their respective complex form spectra, which are called the reference spectrum and the deformed spectrum.

[0048] S520: In the obtained reference spectrum and deformation spectrum, based on the fringe spatial frequency and direction information preset by the structured light projection unit, the first-order spectral component corresponding to the fundamental frequency of the fringe is identified and located. This first-order spectral component is usually represented as a pair of conjugate symmetrical energy concentration regions in the spectrum that deviate from the DC component (spectral center). By constructing a two-dimensional bandpass filter (e.g., Gaussian window, Hanning window, or ideal bandpass filter) with its passband center aligned with the center of the selected first-order spectral component and its bandwidth optimized, this bandpass filter is then applied to the reference spectrum and deformation spectrum respectively to selectively extract the required fundamental frequency information, i.e., the first-order spectral component.

[0049] S530: The first-order spectral component extracted after bandpass filtering in step S520 is shifted in the frequency domain so that its center coincides with the origin of the spectrum (the position of the DC component). Then, a two-dimensional inverse fast Fourier transform (2D-IFFT) is performed on the centered first-order spectrum (corresponding to the reference state and the deformed state, respectively). Two complex images with the same size as the original image are obtained, referred to as the reference complex analytic signal and the deformed complex analytic signal.

[0050] S540: From the reference complex analytic signal and the deformed complex analytic signal, the reference wrapped phase map and the deformed wrapped phase map are extracted respectively by calculating the argument of the complex value at each pixel (i.e., using the real part and imaginary part, the correct phase angle is calculated using the arctangent function atan2). The values ​​of these two wrapped phase maps are restricted to a range from a negative value of pi to a positive value of pi (e.g., -π to +π radians), or a range from zero to twice the value of pi (e.g., 0 to 2π radians), exhibiting a periodic sawtooth distribution.

[0051] S550: Calculate the difference between the phase value of each pixel in the deformation-wrapped phase map and the phase value of the corresponding pixel in the reference-wrapped phase map to obtain the original, wrapped differential phase map. This differential phase map intuitively reflects the phase change caused by thermoelastic deformation, constituting a preliminary form of the Dynamic Response Feature Field (DRSF) described in this invention, or its basic data.

[0052] Furthermore, in order to optimize or further quantify the dynamic response feature field DRSF, the differential phase map in the form of a wrapper obtained in the aforementioned step S550 is subjected to subsequent processing.

[0053] Specifically, this follow-up processing may include at least one of the following optional operations:

[0054] First, phase unwrapping is performed on the wrapped differential phase map. Due to the periodicity of the wrapped phase, the differential phase obtained by direct subtraction may have abrupt changes with a period of integer multiples of pi (i.e., phase ambiguity). Phase unwrapping algorithms (such as branching based on path integrals, quality graph-guided algorithms based on reliability ranking, or least squares methods) aim to eliminate these abrupt changes and recover a continuous "unwrapped differential phase map" that represents the total phase change. This unwrapped differential phase map can serve as a more accurate Dynamic Response Feature Field (DRSF).

[0055] Secondly, the differential phase map in the form of a package or the expanded differential phase map obtained after phase expansion processing is converted into a physical deformation map that directly represents the surface normal displacement based on the phase-physical deformation transformation relationship (e.g., a system calibration coefficient representing the physical displacement corresponding to a unit phase change) determined in advance through system calibration (e.g., using a step of known height or a standard deformation sample for measurement). This physical deformation map makes the deformation information more physically meaningful and constitutes a form of the Dynamic Response Feature Field (DRSF) described in this invention. Its grayscale values ​​or pseudo-color mapping can intuitively display the actual, quantified magnitude and spatial distribution of the transient thermoelastic deformation of the target ink area.

[0056] S600: Perform image segmentation on the dynamic response feature field (DRSF) to identify the deformed region corresponding to the target ink region; further, the process of performing image segmentation on the dynamic response feature field (DRSF) (e.g., the differential phase map, unfolded differential phase map, or physical deformation map obtained in the aforementioned steps) is intended to separate the target ink region that has undergone significant deformation from the background.

[0057] Specifically, the segmentation process may include: determining one or more appropriate thresholds based on prior knowledge of the expected DRSF signal intensity (i.e., the amplitude of pixels in the DRSF image) generated by the target ink under specific excitation conditions, or through statistical analysis of the DRSF image (such as histogram analysis). Pixels in the DRSF image with signal amplitudes higher than (or lower than, depending on the deformation direction and the definition of the DRSF signal) than the preset thresholds are identified as belonging to the target deformation region, while the remaining pixels are considered as background regions. This generates a binary target region mask image, where the target region pixel value is 1 (or 255) and the background region pixel value is 0. To improve the accuracy and robustness of the segmentation results, morphological image processing techniques can be further applied to the initially segmented binary target region mask image. For example, opening operations (i.e., erosion followed by dilation) can remove small isolated noise points and fine burrs that may exist in the image; closing operations (i.e., dilation followed by erosion) can fill small holes that may exist inside the target region and connect broken fine parts in the target region. These morphological optimization operations help to obtain a more complete and regular contour of the target deformation region.

[0058] S700: Extract at least one geometric feature from the identified deformed region; further, extract at least one geometric feature for localization from the identified deformed region (e.g., the binary mask image obtained in step S600 or its corresponding DRSF region).

[0059] Specifically, at least one of the following geometric features can be extracted:

[0060] First, edge detection algorithms (such as the Canny algorithm, the Sobel operator, etc.) or contour tracking algorithms (such as the Suzuki algorithm) are used to extract the precise boundary contour point set of the target deformation region. Based on this contour point set, its length (i.e., perimeter) and the area enclosed by the contour can be further calculated.

[0061] Secondly, calculate the image moments of the target deformed region. For example, the zeroth moment can be used to calculate the area of ​​the region (which can be verified with the area calculation of the contour mentioned above); the first moment is used to calculate the pixel coordinates of the geometric centroid of the region in the image; the second central moment can be used to calculate the direction of the principal axis of inertia of the region (usually expressed as an angle relative to the image coordinate axis, representing the overall orientation of the region) and shape description parameters such as the eccentricity of the region.

[0062] Third, if the target ink features themselves contain clear and predictable geometric shapes (e.g., straight line segments in barcodes, circular or rectangular elements in trademark logos), then in areas of significant deformation in DRSF, methods such as Hough transform (used to detect straight lines, circles, etc.) and least squares fitting can be used to directly detect and fit these geometric primitives, thereby obtaining their precise parameters (e.g., the position and angle of straight lines, the center coordinates and radius of circles, the vertex coordinates of rectangles, etc.).

[0063] Fourth, if the target deformation area in DRSF exhibits clearly identifiable corners (such as the vertices of a rectangular pattern), endpoints (such as the endpoints of barcode lines), or specific inflection points, algorithms such as Harris corner detection, Shi-Tomasi corner detection, or FAST corner detection can be applied to extract the precise pixel coordinates of these key feature points.

[0064] S800: Based on the at least one geometric feature and in conjunction with preset system calibration parameters, determine the position or orientation of the target ink area on the cigarette pack cardboard.

[0065] Specifically, the system calibration must first be completed through preliminary work. This includes camera calibration (obtaining the camera's intrinsic parameter matrix and distortion coefficients) and hand-eye calibration or world coordinate system calibration (establishing a precise transformation relationship between the image pixel coordinate system and the physical world coordinate system of the cigarette pack cardboard surface, for example, by calculating the homography matrix or a more complete projection transformation model). Then, one or more geometric features extracted in step S700 (e.g., the pixel coordinates of the centroid and the attitude angle in the image obtained from principal axis analysis, or a set of pixel coordinates of multiple feature points) are used as input. Using the obtained system calibration parameters (i.e., camera parameters and coordinate transformation relationship), through corresponding mathematical operations (such as direct application of coordinate system transformation formulas, or PnP solving for multiple feature points, affine / projective transformation fitting, etc.), they are transformed from the image coordinate system to the physical coordinate system of the cigarette pack cardboard. Therefore, the planar position of the target ink feature in the physical coordinate system of the cigarette pack cardboard can be calculated (e.g., world coordinates, attitude angle (e.g., rotation angle in the world coordinate system)). If a template-matching-based positioning strategy is adopted, and a standard defect-free sample DRSF is pre-made as a reference template, this step can also be achieved by calculating the optimal alignment transformation (e.g., translation, rotation, scaling parameters) between the currently acquired DRSF (or its extracted features) and the reference DRSF template. These transformation parameters indicate the deviation of the current target ink feature from the standard pose of the template, and can thus be used for precise positioning and possible subsequent production calibration.

[0066] Further step S900: Preprocess the dynamic response feature field (DRSF) and establish a reference benchmark for defect judgment; using the dynamic response feature field DRSF calculated in the aforementioned step S500, identify microscopic defects related to the target ink or its underlying substrate by analyzing its abnormal characteristics, and generate early warning information.

[0067] Specifically, first, the dynamic response feature field DRSF generated in the aforementioned step S500 is obtained, for example, it can be a differential phase map, an expanded differential phase map, or a physical deformation map.

[0068] Secondly, to ensure the accuracy and consistency of subsequent defect analysis, necessary preprocessing can be performed on the currently acquired DRSF. For example, if subsequent analysis relies on comparison with a pre-stored template, it may be necessary to first perform precise spatial image registration between the current DRSF and the template to eliminate artifacts caused by minor positioning deviations or slight overall deformation of the DRSF itself (not caused by defects). Furthermore, if there is a drift in the overall signal intensity of the DRSF due to minor systematic fluctuations in the excitation source energy or the gain of the image acquisition unit, the signal intensity of the current DRSF can be normalized based on the signal intensity of certain known stable reference areas in the DRSF (e.g., a large area of ​​unprinted blank paper, whose DRSF response should be close to zero or a stable background value), or based on the statistical mean of the overall DRSF signal.

[0069] Next, establish or load reference benchmarks for defect determination. Reference benchmarks can be one of the following forms or a combination thereof:

[0070] a) DRSF template for healthy samples: DRSF images are pre-collected and averaged (or selected from typical samples) on a large number of qualified cigarette packs of the same type as those currently being tested, under standard excitation and measurement conditions consistent with the current testing conditions. This template represents the DRSF response pattern under healthy conditions.

[0071] b) Statistical characteristic model of DRSF: A model describing the normal parameter range of DRSF is learned through statistical analysis of a large number of healthy DRSF samples. For example, for each pixel or each pre-divided sub-region in DRSF, the statistical mean and standard deviation of its signal amplitude (such as phase difference or physical deformation) can be obtained; or, for the entire DRSF or a specific region therein, the normal distribution range of its gradient, the expected value range of texture features, etc., can be obtained.

[0072] c) Design expectations based on internal consistency: For certain target ink areas on cigarette pack paper that should exhibit uniform printing (such as large areas of solid color blocks) or regular repetition (such as barcode lines), the resulting DRSF (Dual Printed Solid State) should also show spatial uniformity, smoothness, or specific periodicity under defect-free conditions. This internal consistency based on design expectations can itself serve as a dynamic reference benchmark.

[0073] S910: Analyze the abnormal features in the Dynamic Response Feature Field (DRSF) that are inconsistent with the reference benchmark, identify micro-defects and generate early warning information;

[0074] Specifically, by comparing the preprocessed current DRSF with the reference benchmark established or loaded in step S900, or by analyzing whether the signal characteristics within the current DRSF itself deviate from expectations, anomalous features that may indicate the presence of microscopic defects are extracted and quantified. At least one of the following analysis operations can be used:

[0075] i) Anomaly feature extraction based on differences from the reference template: If a DRSF template of a healthy sample is loaded in step S900, a pixel-by-pixel difference map is calculated between the current DRSF (after registration and normalization if necessary) and the reference template DRSF. For example, the absolute value of the difference between corresponding pixel values ​​of the two images can be directly calculated, or the normalized correlation coefficient value can be calculated and low correlation regions can be identified. Then, a difference tolerance threshold is set, and pixels or connected regions in the difference map whose difference values ​​exceed this threshold are identified as anomaly features indicating potential defects.

[0076] ii) Feature extraction based on local anomalies in DRSF signal amplitude: Within the current DRSF (e.g., differential phase map or physical deformation map), for each pixel or predefined local neighborhood window, analyze whether its signal amplitude deviates significantly from the normal level. For example, the difference between the signal amplitude of this point and the statistical average of the signal amplitudes of pixels in its surrounding M×M neighborhood can be calculated. If this difference exceeds a preset fluctuation threshold, it is determined to be an anomaly. Alternatively, the signal amplitude of this point can be compared with the normal value range (e.g., mean plus or minus N times the standard deviation) of this location (or this type of region) obtained based on statistics from a large number of healthy samples. If it exceeds this range, it is determined to be an anomaly. Such amplitude anomalies may indicate that the ink layer is too thin or too thick, the printing is incomplete, or the thermophysical properties of the underlying substrate have changed.

[0077] iii) Feature extraction based on DRSF spatial gradient anomalies: First, calculate the spatial gradient distribution map of the current DRSF. The value of each pixel in this gradient map represents the degree and direction of change of DRSF near that point. Then, analyze whether there are anomalies in this gradient map. For example, detect whether the gradient amplitude increases abnormally in certain local areas (which may indicate burrs at the ink edge, the tip of a small crack, or discontinuities in the deformation field), or whether the gradient direction undergoes irregular, unexpected, and drastic changes, or whether the gradient distribution map itself exhibits breaks or jumps. These can all serve as anomalous features indicating defects.

[0078] iv) Feature extraction based on anomalies in DRSF spatial continuity or uniformity indicators: Focus on analyzing regions in DRSF that should exhibit spatial continuity (e.g., barcode lines) or uniformity (e.g., large color blocks) according to the cigarette pack design. For example, for the deformation bands corresponding to barcode lines in DRSF, image connectivity analysis algorithms (e.g., methods based on region growing or contour tracking) can be used to check whether they maintain topological integrity. If breaks or discontinuities are detected in the deformation bands, it may indicate ink line breaks. For expected uniform color block regions, their local standard deviation, entropy value, or specific texture descriptors (e.g., contrast, correlation, energy, etc. calculated based on the gray-level co-occurrence matrix) can be calculated. If these uniformity indicator values ​​are lower than preset quality standards or differ from the statistical values ​​of healthy samples by a value greater than a preset comparison threshold, it may indicate uneven ink coating, the presence of internal micro-impurities or pigment agglomerations, or non-uniform response patterns caused by uneven characteristics of the underlying paper substrate.

[0079] Finally, based on the abnormal features extracted and quantified through one or more of the above operations, the data processing unit will comprehensively determine whether microscopic defects exist and their possible types (such as broken lines, pinholes, uneven ink layers, poor interfaces, and abnormal substrates) according to pre-set defect judgment rules (e.g., comprehensively considering the size, shape, quantity, location, and degree of deviation from normal conditions of abnormal features, and possibly assigning different weights or judgment logic to different types of abnormal features). If a microscopic defect requiring attention is determined, a corresponding warning message will be generated. This warning message can be output in a visual manner (e.g., marking the defect location and displaying relevant parameters on an image on the operation interface), as an audible alarm, or as a digital signal, to guide subsequent quality control operations (e.g., product rejection, process parameter adjustment) or to generate detailed quality inspection reports for archiving and analysis.

[0080] Those skilled in the art should understand that the order of the above steps is not absolutely fixed. Without affecting the core technical effect, the order of some steps can be adjusted, or some steps can be combined or further subdivided. Furthermore, the method and system of this invention can also add other auxiliary processing steps according to the needs of actual application scenarios, such as confidence assessment of the final positioning result, historical data tracking and analysis, etc.

[0081] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method of locating a carton of cigarettes based on image analysis, characterised in that, The method comprises the following steps: S100: projecting a structured light pattern to a target ink area on the surface of the cigarette wrapper cardboard; S200: acquiring a first structured light fringe image of the target ink area under illumination of the structured light pattern as a reference fringe image; S300: applying pulsed energy to the target ink area on the surface of the cigarette wrapper cardboard to cause selective photothermal excitation, thereby generating transient thermoelastic deformation; S400: at a preset response moment of the transient thermoelastic deformation of the target ink area caused by the selective photothermal excitation, acquiring a second structured light fringe image of the target ink area under illumination of the structured light pattern as a deformation fringe image; S500: based on the reference fringe image and the deformation fringe image, calculating a dynamic response feature field representing the transient thermoelastic deformation of the target ink area caused by photothermal excitation; S600: performing image segmentation on the dynamic response feature field to identify a deformation area corresponding to the target ink area; S700: extracting at least one geometric feature from the identified deformation area; S800: determining the position or attitude of the target ink area on the cigarette wrapper cardboard according to the at least one geometric feature and in combination with preset system calibration parameters; The step S500 of calculating the dynamic response feature field comprises: S510: performing two-dimensional Fourier transform on the reference fringe image and the deformation fringe image respectively to obtain respective reference spectrum and deformation spectrum; S520: from the reference spectrum and the deformation spectrum, according to a preset spatial frequency and direction of the structured light fringe, identifying and extracting a first-order spectrum component corresponding to a fringe fundamental frequency respectively; S530: performing two-dimensional inverse Fourier transform on the extracted first-order spectrum component to obtain a reference complex analytic signal and a deformation complex analytic signal respectively; S540: by calculating the argument of each pixel point of the reference complex analytic signal and the deformation complex analytic signal, a reference wrapped phase image and a deformation wrapped phase image are obtained respectively; S550: calculating the pixel-by-pixel difference between the deformation wrapped phase image and the reference wrapped phase image to obtain a wrapped differential phase image, which constitutes the dynamic response feature field.

2. A method of locating a carton of cigarettes based on image analysis as claimed in claim 1, wherein, The step S500 further comprises at least one of the following operations on the wrapped differential phase image: performing phase unwrapping processing on the wrapped differential phase image to eliminate jumps of integer multiples of two times of pi, to obtain an unwrapped differential phase image, and taking this unwrapped differential phase image as the dynamic response feature field; or converting the wrapped differential phase image or the unwrapped differential phase image into a physical deformation variable image representing surface normal displacement variable according to a preset phase-physical deformation variable conversion relationship, and taking this physical deformation variable image as the dynamic response feature field.

3. A method of locating a carton of cigarettes based on image analysis as claimed in claim 1, wherein, The image segmentation of the dynamic response feature field in the step S600 includes: setting one or more threshold values according to signal amplitudes in the dynamic response feature field, identifying pixel points with signal amplitudes satisfying a preset condition as belonging to a target deformation region, and identifying the rest of the pixel points as background, thereby generating a binary target region mask image.

4. A method of locating a carton of cigarettes based on image analysis as claimed in claim 1, wherein, The step S700 of extracting at least one geometric feature from the identified deformation region includes at least one of the following operations: using an edge detection algorithm or a contour tracking algorithm to extract a boundary contour of the deformation region and calculating its perimeter or the area enclosed; calculating the image moment of the deformation region to obtain the pixel coordinates of its geometric centroid or the inertia principal axis angle representing its direction; if the target ink region contains an expected geometric shape, using a Hough transform or a least squares fitting method to detect and fit the geometric primitives of the geometric shape in the deformation region to obtain its parameters; An angle point detection algorithm is applied to extract the pixel coordinates of the corner points, end points or specific inflection points in the deformation region.

5. A method of locating a carton of cigarettes based on image analysis as claimed in claim 1, wherein, The pulse energy applied in the step S300 is derived from a near-infrared pulsed laser or a high-power pulsed light-emitting diode array, and the pulse width and pulse energy are adjusted to be sufficient to cause measurable thermal-elastic deformation of the target ink but far below the ablation threshold of the target ink; The time interval between the reference fringe image obtained in the step S200 and the deformation fringe image obtained in the step S400 is controlled to be within a range sufficient to cause quantitative thermal-elastic deformation of the target ink but as short as possible, so as to effectively capture the transient thermal-elastic deformation and suppress the influence of low-frequency environmental vibration. The preset time for obtaining the deformation fringe image is a deformation peak time estimated according to the thermophysical properties of the target ink material and the excitation parameters or a certain time within an optimal detection time window determined through experiments.

6. A method of locating a carton of cigarettes based on image analysis as claimed in claim 1, wherein, The method further includes the following steps: S900: based on the dynamic response feature field calculated in the step S500, analyzing abnormal features in the dynamic response feature field that are inconsistent with the response mode of the target ink region in a defect-free state; S910: according to the abnormal features analyzed in the step S900, identifying microscopic defects related to the printing quality of the target ink on the cigarette package paper, the interface state of the target ink and the cigarette package paper substrate, or the cigarette package paper substrate itself, and generating a warning message.

7. A method of locating a carton of cigarettes based on image analysis as claimed in claim 6, wherein, The step S900 of analyzing abnormal features of the dynamic response feature field includes at least one of the following operations: registering and comparing the currently obtained dynamic response feature field with a reference dynamic response feature field template of the response mode of the target ink region in a defect-free state, and identifying a region with a difference value exceeding a preset tolerance threshold as the abnormal feature; In the dynamic response feature field, the signal amplitude representing the phase change amount or the converted physical deformation amount caused by the transient thermoelastic deformation is detected, and a local region whose signal amplitude is outside the statistical average value of its neighboring pixels or outside a predefined normal value interval is defined as the abnormal feature; The spatial gradient distribution of the dynamic response feature field is calculated, and a region whose gradient amplitude exceeds a preset gradient threshold value or whose gradient direction change rate exceeds a preset direction change rate threshold value or in which a breakpoint or a jump point is detected in the gradient distribution is defined as the abnormal feature; For an ink region in the dynamic response feature field that is expected to exhibit a spatially continuous or uniform response, the actual spatial continuity or uniformity index of its signal is evaluated, and a region in which a break or discontinuous point is detected or in which the uniformity index is lower than a preset standard value is defined as the abnormal feature.

Images