Machine vision-based online detection method and system for tear resistance of thermal paper

Abstract

The application discloses a machine vision-based online detection method and system for tear resistance of thermal paper, and relates to the technical field of tear resistance detection of thermal paper, and comprises the following steps: collecting a paper image; recording an audio waveform generated when the thermal paper is hit; extracting surface features of the thermal paper in the paper image; obtaining audio features of the thermal paper; constructing an online detection model for the tear resistance of the thermal paper, collecting the surface features and the audio features of the thermal paper with known quality and performing deep learning; analyzing the surface features and the audio features through the online detection model for the tear resistance of the thermal paper to determine whether the tear resistance of the thermal paper is qualified; and the application is used to solve the problems that the existing detection technology for the tear resistance of the thermal paper needs to damage the thermal paper for tear resistance detection and adopts a spot-check sampling method for detection, thereby reducing the production capacity of the thermal paper and failing to guarantee the yield of the thermal paper.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of tear resistance detection of thermal paper, in particular to a method and system for online detection of tear resistance of thermal paper based on machine vision. BACKGROUND

[0002] The tear resistance detection technology of thermal paper refers to a set of methods and means for quantitatively measuring the ability of thermal paper to resist external tearing expansion. Usually, a controllable force is applied to measure the force value required for the paper to continue tearing, thereby objectively evaluating the mechanical durability and physical quality.

[0003] The existing tear resistance detection technology of thermal paper usually needs to take the thermal paper off the production line after the production of thermal paper is completed, and then tear the thermal paper by a tensile testing machine to evaluate the tear resistance of the thermal paper. This method can only evaluate the tear resistance of the thermal paper by sampling, and needs to destroy the thermal paper, which also leads to a decrease in production capacity. At the same time, sampling actually reflects the whole by part, which has certain errors and cannot ensure the accuracy of the detection results. In some cases, unqualified thermal paper cannot be correctly detected, and the yield cannot be guaranteed. The existing tear resistance detection technology of thermal paper also has the problems of destroying the thermal paper for tear resistance detection and detecting by sampling, which leads to a decrease in the production capacity of the thermal paper and the yield of the thermal paper cannot be guaranteed. SUMMARY

[0004] The present application aims to at least solve one of the technical problems in the prior art. A light source and a camera are installed on the thermal paper production line to collect paper images. A paper beating device is installed on the thermal paper production line to beat the thermal paper and record the waveform of the sound produced when the thermal paper is beaten, which is named audio waveform. The surface features of the thermal paper in the paper image are extracted by feature analysis of the paper image. The audio features of the thermal paper are obtained by feature analysis of the audio waveform. Then, an online detection model for the tear resistance of thermal paper is constructed. The surface features of the thermal paper of known quality are collected and deep learning is performed to obtain surface reference features. The audio features of the thermal paper of known quality are collected and deep learning is performed to obtain audio reference features. Finally, the surface features and audio features are analyzed by the online detection model for the tear resistance of thermal paper to determine whether the tear resistance of the thermal paper is qualified, thereby solving the problems of the existing tear resistance detection technology of thermal paper, which needs to destroy the thermal paper for tear resistance detection and detects by sampling, leading to a decrease in the production capacity of the thermal paper and the yield of the thermal paper cannot be guaranteed.

[0005] To achieve the above object, in a first aspect, the application provides a method for online detection of tear resistance of thermal paper based on machine vision, comprising the following steps:

[0006] A light source and a camera are installed on the thermal paper production line to collect paper images;

[0007] A paper beating device is installed on the thermal paper production line to beat the thermal paper and record the waveform of the sound generated when the thermal paper is beaten, named audio waveform;

[0008] The paper image is analyzed for features, and the surface features of the thermal paper in the paper image are extracted;

[0009] The audio waveform is analyzed for features, and the acoustic features of the thermal paper are obtained;

[0010] An online detection model of the tear resistance of thermal paper is constructed, the surface features and acoustic features of thermal paper of known quality are collected and deep learning is performed;

[0011] The surface features and acoustic features are analyzed by the online detection model of the tear resistance of thermal paper to determine whether the tear resistance of the thermal paper is qualified.

[0012] Further, the light source and the camera installed on the thermal paper production line for collecting paper images comprise the following sub-steps:

[0013] The light source uses a non-flickering LED linear light source, and the camera uses a linear array camera;

[0014] The LED linear light source irradiates the thermal paper at a first angle, and the linear array camera is located directly above the thermal paper to take a picture of the thermal paper to obtain a paper image.

[0015] Further, the paper beating device installed on the thermal paper production line for beating the thermal paper and recording the waveform of the sound generated when the thermal paper is beaten comprises the following sub-steps:

[0016] The paper beating device comprises a front conveyor belt, a rear conveyor belt, a front fixing device, a rear fixing device and a beating device;

[0017] The front fixing device is installed above the front conveyor belt, and the rear fixing device is installed above the rear conveyor belt;

[0018] Before the heat-sensitive paper is tapped, the front conveying belt stops running, the front fixing device presses down to fix the front end of the heat-sensitive paper, the rear conveying belt stops running, the rear fixing device presses down to fix the rear end of the heat-sensitive paper, then the rear conveying belt is started, the rear fixing device can move synchronously with the rear conveying belt, at this time, the heat-sensitive paper between the front fixing device and the rear fixing device will produce a protrusion, the protrusion part is tapped by the tapping device and the waveform of the sound generated during tapping is collected, and the audio waveform is obtained.

[0019] Further, the paper image is subjected to feature analysis, and the surface features of the heat-sensitive paper in the paper image include the following sub-steps:

[0020] The paper image is subjected to grayscale processing to obtain a grayscale image.

[0021] The gray level co-occurrence matrix of the grayscale image is obtained, and the contrast and correlation in the gray level co-occurrence matrix are extracted as the surface features of the heat-sensitive paper.

[0022] Further, the audio waveform is subjected to feature analysis to obtain the acoustic features of the heat-sensitive paper, including the following sub-steps:

[0023] The audio waveform is stored in an audio graph, the X-axis of the audio graph represents time, and the Y-axis represents amplitude, the coordinates of the wave peaks of the audio waveform are obtained and named as peak points, and the coordinates of the wave troughs of the audio waveform are obtained and named as valley points.

[0024] The peak points and the valley points are numbered in chronological order, and are represented by symbols F n and G n respectively, wherein n is a non-zero natural number and n is the serial number of F and G, in general, the number of peak points and valley points is the same, and G n is the first valley point after F n .

[0025] F n and G n are connected by a straight line, the obtained auxiliary line is marked as L n , the slope of L n is obtained and marked as K n .

[0026] Two adjacent F n are connected by a straight line, and finally a polyline segment is obtained, which is named as a wave peak analysis polyline.

[0027] The time corresponding to F n is marked as T n , K n is associated with T n , T n is taken as the horizontal axis, and Kn A plane rectangular coordinate system is established for the longitudinal axis, named as a peak-valley analysis diagram, and K n According to T n Enter the peak-valley analysis diagram, and name the coordinate points in the peak-valley analysis diagram as peak-valley analysis points;

[0028] K n and T n constitute a coordinate point marked as P n , adjacent two P n are connected by a straight line, and finally a polyline segment is obtained, named as a peak-valley analysis polyline.

[0029] The peak analysis polyline and the peak-valley analysis polyline are the sound wave characteristics.

[0030] Further, an online detection model of tear resistance of thermal paper is constructed, the surface characteristics and the sound wave characteristics of thermal paper of known quality are collected and deep learning is performed, including the following sub-steps:

[0031] The online detection model of tear resistance of thermal paper is constructed, the surface characteristics of thermal paper of known quality are collected and deep learning is performed, and surface reference characteristics are obtained.

[0032] The sound wave characteristics of thermal paper of known quality are collected and deep learning is performed, and sound wave reference characteristics are obtained.

[0033] Further, the online detection model of tear resistance of thermal paper is constructed, the surface characteristics of thermal paper of known quality are collected and deep learning is performed, and surface reference characteristics are obtained, including the following sub-steps:

[0034] The online detection model of tear resistance of thermal paper is constructed, thermal paper of known quality is obtained as a sample paper, the quality of the sample paper is named as a sample quality, the surface characteristics of the sample paper are extracted and named as surface sample characteristics, and the sample quality corresponds to the surface sample characteristics;

[0035] The contrast and the correlation in the surface sample characteristics are respectively named as a sample contrast and a sample correlation;

[0036] The range of the sample contrast of the sample quality that is qualified is counted, and a first reference range is obtained, the range of the sample correlation of the sample quality that is qualified is counted, and a second reference range is obtained, and the first reference range and the second reference range jointly constitute the surface reference characteristics.

[0037] Further, the sound wave characteristics of thermal paper of known quality are collected and deep learning is performed, and sound wave reference characteristics are obtained, including the following sub-steps:

[0038] Collecting the sound wave features of the sample paper with qualified sample quality, and naming the sound wave features as sound wave sample features, and naming the peak analysis fold line and the trough analysis fold line in the sound wave sample features as the peak sample fold line and the trough sample fold line respectively;

[0039] Naming the leftmost end point of the peak sample fold line as the peak starting point, and naming the leftmost end point of the trough sample fold line as the trough starting point, and naming the turning points and end points in the peak sample fold line and the trough sample fold line as fold line vertices;

[0040] Overlapping the peak starting points of all the peak sample fold lines, and naming the overlapped image as a peak learning image; overlapping the trough starting points of all the trough sample fold lines, and naming the overlapped image as a trough learning image;

[0041] When deep learning is performed on the peak learning image or the trough learning image, naming the peak learning image or the trough learning image currently analyzed as a target analysis image, and naming the X axis and the Y axis corresponding to the target analysis image as a first axis and a second axis respectively;

[0042] For each value of the first axis, obtaining the fold line vertex at the uppermost position and the fold line vertex at the lowermost position, and naming them as an upper limit point and a lower limit point respectively;

[0043] Numbering the upper limit points and the lower limit points in the order from left to right, and representing them by symbols U i and D j respectively, wherein i and j are both non-zero natural numbers, i is the serial number of U, and j is the serial number of D;

[0044] Setting an auxiliary number h, which is initially 1, starting with i = 1, connecting U i and U i+h by a straight line to obtain an upper limit analysis line, and judging whether the upper limit analysis line intersects with other parts of the target analysis image except the upper limit points, if yes, outputting an upper limit error signal, and if no, outputting an upper limit correct signal;

[0045] If the upper limit error signal is outputted, deleting the upper limit analysis line and U i+h at the same time, increasing h by 1 and analyzing the upper limit analysis line again, repeating the judgment until the upper limit correct signal is outputted; if the upper limit correct signal is outputted, retaining the upper limit analysis line and i + 1 at the same time, judging whether U i exists, if U i does not exist, increasing i by 1 again, if U i exists, resetting h to 1 and analyzing another upper limit analysis line again until the maximum value of i is reached, obtaining different upper limit analysis lines, and composing new fold line segments from the upper limit analysis lines, and naming them as upper limit reference lines;

[0046] h is reset to 1, j is started to 1, D j is connected with D j+h , to obtain a lower limit analysis line, and it is judged whether the lower limit analysis line intersects with other parts of the target analysis graph except the lower limit point, if yes, a lower limit error signal is output, if no, a lower limit correct signal is output;

[0047] If the lower limit error signal is output, the lower limit analysis line is deleted, and D j+h is deleted, h is added by one and the lower limit analysis line is analyzed again, the judgment is repeated until the lower limit correct signal is output; if the lower limit correct signal is output, the lower limit analysis line is retained, and j is added by one, it is judged whether D j exists, if D j does not exist, j is added by one again, if D j exists, h is reset to 1 and another lower limit analysis line is analyzed again until the maximum value of j is reached, different lower limit analysis lines are obtained, a new broken line segment is composed of the lower limit analysis lines, and is named as a lower limit reference line;

[0048] The leftmost end points of the upper limit reference line and the lower limit reference line are respectively named as an upper limit left end point and a lower limit left end point, the rightmost end points of the upper limit reference line and the lower limit reference line are respectively named as an upper limit right end point and a lower limit right end point, the upper limit left end point is connected with the lower limit left end point, and the upper limit right end point is connected with the lower limit right end point, to obtain a closed area, which is named as a target reference feature;

[0049] The target reference features obtained by analyzing the peak learning graph and the valley learning graph are respectively named as a peak reference feature and a valley reference feature, and the peak reference feature and the valley reference feature together compose an acoustic wave reference feature.

[0050] Further, the surface feature and the acoustic wave feature are analyzed by the thermal paper tear resistance online detection model to judge whether the tear resistance of the thermal paper is qualified, including the following sub-steps:

[0051] The surface feature and the acoustic wave feature currently required to be analyzed are respectively named as a surface real-time feature and an acoustic wave real-time feature, the contrast and the correlation in the surface real-time feature are respectively named as a real-time contrast and a real-time correlation, and the peak analysis broken line and the valley analysis broken line in the acoustic wave real-time feature are respectively named as a peak real-time broken line and a valley real-time broken line;

[0052] If the real-time contrast is in the first reference range, the real-time correlation is in the second reference range, the peak real-time broken line is all in the peak reference feature, and the valley real-time broken line is all in the valley reference feature, a tear resistance qualified signal is output, otherwise a tear resistance unqualified signal is output.

[0053] In a second aspect, the application provides a machine vision-based online detection system for the tear resistance of thermal paper, comprising an image acquisition module, an audio acquisition module, a surface feature extraction module, an acoustic feature extraction module, a feature analysis module, and a tear resistance judgment module; the image acquisition module, the audio acquisition module, the surface feature extraction module, the acoustic feature extraction module, and the tear resistance judgment module are respectively connected to the feature analysis module.

[0054] The image acquisition module is used to install a light source and a camera on a thermal paper production line to acquire a paper image.

[0055] The audio acquisition module is used to install a paper tapping device on the thermal paper production line to tap the thermal paper and record the waveform of the sound produced when the thermal paper is tapped, which is named as an audio waveform.

[0056] The surface feature extraction module is used to analyze the features of the paper image and extract the surface features of the thermal paper in the paper image.

[0057] The acoustic feature extraction module is used to analyze the features of the audio waveform and obtain the acoustic features of the thermal paper.

[0058] The feature analysis module is used to construct an online detection model for the tear resistance of thermal paper, collect the surface features and acoustic features of thermal paper of known quality, and perform deep learning.

[0059] The tear resistance judgment module is used to analyze the surface features and acoustic features through the online detection model for the tear resistance of thermal paper to determine whether the tear resistance of the thermal paper is qualified.

[0060] The application has the following advantages: the application installs a light source and a camera on a thermal paper production line to acquire a paper image, then installs a paper tapping device on the thermal paper production line to tap the thermal paper and record the waveform of the sound produced when the thermal paper is tapped, which is named as an audio waveform, then analyzes the features of the paper image and extracts the surface features of the thermal paper in the paper image, and analyzes the features of the audio waveform to obtain the acoustic features of the thermal paper, which has the advantage that the surface of the thermal paper has microscopic textures, the surface features are the embodiment of the microscopic textures, and the sound produced when papers with different tear resistances are tapped is also different, so the acoustic features are obtained, and the accuracy and effectiveness of the detection of the tear resistance of the thermal paper are improved.

[0061] The application constructs a tear resistance strength online detection model of thermal paper, collects surface features of thermal paper of known quality and performs deep learning to obtain surface reference features, simultaneously collects audio features of thermal paper of known quality and performs deep learning to obtain audio reference features, and finally analyzes the surface features and the audio features through the tear resistance strength online detection model of thermal paper to determine whether the tear resistance strength of the thermal paper is qualified, the advantage being that for the same material paper, the sound difference is small, but the computer can still extract the tiny difference, and the machine vision and voiceprint recognition are combined through the deep learning mode, and the accuracy and effectiveness of the tear resistance strength online detection of the thermal paper are further improved. BRIEF DESCRIPTION OF DRAWINGS

[0062] Figure 1 It is a principle block diagram of the system of the application;

[0063] Figure 2 It is a schematic diagram of the paper beating device of the application;

[0064] Figure 3 It is a schematic diagram of the audio waveform of the application;

[0065] Figure 4 It is a schematic diagram of the L n of the application;

[0066] Figure 5 It is a schematic diagram of the peak analysis fold line of the application;

[0067] Figure 6 It is a schematic diagram of the peak-valley analysis diagram of the application;

[0068] Figure 7 It is a schematic diagram of the peak learning diagram of the application;

[0069] Figure 8 It is a schematic diagram of the upper limit point and the lower limit point of the application;

[0070] Figure 9 It is a schematic diagram of the peak reference feature of the application;

[0071] Figure 10 It is a step flowchart of the method of the application. DETAILED DESCRIPTION

[0072] The technical solutions in the embodiments of the application will be clearly and completely described below with reference to the drawings in the embodiments of the application, obviously, the described embodiments are only part of the embodiments of the application, rather than all the embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor fall within the protection scope of the application.

[0073] Embodiment 1, please refer to Figure 1 As shown in the figure, the present application provides a machine vision-based thermal paper tear resistance strength online detection system, which comprises an image acquisition module, an audio acquisition module, a surface feature extraction module, an acoustic feature extraction module, a feature analysis module and a tear resistance strength judgment module; the image acquisition module, the audio acquisition module, the surface feature extraction module, the acoustic feature extraction module and the tear resistance strength judgment module are respectively connected with the feature analysis module in data;

[0074] The image acquisition module is used to install a light source and a camera on the thermal paper production line for collecting paper images;

[0075] The image acquisition module is configured with an image acquisition strategy, which includes:

[0076] The light source uses a non-flickering LED linear light source, and the camera uses a linear array camera;

[0077] The LED linear light source irradiates the thermal paper at a first angle, and the linear array camera is located directly above the thermal paper to take pictures of the thermal paper, obtaining the paper image;

[0078] In practical application, the linear array camera uses ultra-high resolution, and the present embodiment uses an 8k resolution linear array camera. The linear array camera only collects one row of pixel points, and there are 8192 pixel points in this row. The texture information of the thermal paper surface is collected, and under the irradiation of the LED linear light source, the texture information presents a certain degree of light and dark contrast. The non-flickering LED linear light source and the linear array camera are both existing devices, and the present embodiment will not be described in detail; the first angle is adjusted by the test personnel to ensure that there is light and dark contrast in the collected paper image.

[0079] The audio acquisition module is used to install a paper slapping device on the thermal paper production line to slap the thermal paper and record the waveform of the sound produced when the thermal paper is slapped, which is named as audio waveform;

[0080] The audio acquisition module is configured with an audio acquisition strategy, which includes:

[0081] Please refer to Figure 2 As shown in the figure, the paper slapping device comprises a front conveying belt, a rear conveying belt, a front fixing device, a rear fixing device and a slapping device;

[0082] The front fixing device is installed above the front conveying belt, and the rear fixing device is installed above the rear conveying belt;

[0083] Before the thermal paper is tapped, the front conveyor belt stops running, the front fixing device presses down to fix the front end of the thermal paper, the rear conveyor belt stops running, the rear fixing device presses down to fix the rear end of the thermal paper, and then the rear conveyor belt is started. The rear fixing device can move synchronously with the rear conveyor belt. At this time, a protrusion will be generated on the thermal paper between the front fixing device and the rear fixing device. The tapping device taps the protrusion and collects the waveform of the sound generated by the thermal paper during the tapping to obtain the audio waveform.

[0084] In practical applications, the front conveyor belt, rear conveyor belt, front fixing device, and rear fixing device in the paper tapping device are as follows: Figure 2 As shown, the tapping device is located between the front and rear fixing devices. The tapping device is a robotic arm, and the axis of rotation of the robotic arm is far from the thermal paper. The long robotic arm taps the thermal paper, ensuring that the axis of rotation of the robotic arm does not affect the recording of the audio waveform. The different tear resistance of the same type of paper when tapped will also cause slight differences in the audio waveform. The difference in tear resistance is caused by the different structures or materials of the paper during production. The audio waveform generated by the paper being tapped is also affected by the material and structure of the paper. Therefore, when the tear resistance of the paper is different, its audio waveform will also produce slight changes under controlled variables.

[0085] The surface feature extraction module is used to perform feature analysis on paper images and extract the surface features of thermal paper from the paper images;

[0086] The surface feature extraction module is configured with surface feature extraction strategies, which include:

[0087] The paper image is converted to grayscale to obtain a grayscale image;

[0088] Obtain the gray-level co-occurrence matrix of the gray-level image, and extract the contrast and correlation in the gray-level co-occurrence matrix as surface features of the thermal paper;

[0089] In practical applications, the acquisition and construction of gray-level co-occurrence matrix are existing technologies, and will not be described in detail in this embodiment. The analysis of gray-level co-occurrence matrix is ​​based on machine vision methods. However, in the actual analysis process, machine vision still has certain drawbacks. Therefore, this embodiment integrates voiceprint recognition on the basis of machine vision, and performs a deeper analysis of the tear resistance of thermal paper by extracting subtle changes in audio waveforms.

[0090] The sound wave feature extraction module is used to perform feature analysis on audio waveforms to obtain the sound wave features of thermal paper;

[0091] The sound wave feature extraction module is configured with a sound wave feature extraction strategy, which includes:

[0092] Please refer to Figure 3 As shown in the figure, the audio waveform is stored in the audio graph, the X axis of the audio graph represents time, and the Y axis represents amplitude. The coordinates of the peak of the audio waveform are obtained, which are named as peak points. The coordinates of the trough of the audio waveform are obtained, which are named as trough points.

[0093] The peak points and the trough points are numbered in chronological order, and are represented by symbols F n and G n respectively, where n is a non-zero natural number and n is the serial number of F and G. In general, the number of peak points and trough points is the same, and G n is the first trough point after F n .

[0094] Please refer to Figure 4 As shown in the figure, F n and G n are connected by a straight line, and the obtained auxiliary line is marked as L n . The slope of L n is obtained and marked as K n .

[0095] Please refer to Figure 5 As shown in the figure, the adjacent two F n are connected by a straight line, and finally a polyline segment is obtained, which is named as the peak analysis polyline.

[0096] In practical applications, due to the fact that the audio waveform in the original audio graph is too dense to be observed by the naked eye, only a small number of waveforms in this embodiment are taken as examples to illustrate the subsequent analysis process. The audio waveform in this embodiment is shown in Figure 3 , wherein the peak points and the trough points are marked, and F n and G n are obtained by numbering, wherein 1≤n≤4. L n is obtained by connection, as shown in Figure 4 , Figure 4 The dashed line in n is L n , and K n is obtained. Then the peak analysis polyline is obtained by connection, as shown in Figure 5 .

[0097] Please refer to Figure 6 As shown in the figure, the time corresponding to F n is marked as T n , and K n is associated with T n . A plane rectangular coordinate system is established with T n as the horizontal axis and K n as the vertical axis, which is named as the peak valley analysis graph. K nThe coordinate points in the peak-valley analysis graph are named as peak-valley analysis points;

[0098] K n and T n are connected by a straight line, and finally a polyline segment is obtained, which is named as a peak-valley analysis polyline; n n The peak-valley analysis polyline is obtained.

[0099] The peak-valley analysis polyline is obtained.

[0100] In actual application, the peak-valley analysis graph is constructed as shown in Figure 6 , and the peak-valley analysis polyline is also shown in Figure 6 .

[0101] The characteristic analysis module is used to construct the thermal paper tear resistance strength online detection model, collect the surface characteristics and the audio characteristics of the thermal paper with known quality and perform deep learning; the characteristic analysis module includes a surface analysis unit and an audio analysis unit;

[0102] The surface analysis unit is used to construct the thermal paper tear resistance strength online detection model, collect the surface characteristics of the thermal paper with known quality and perform deep learning to obtain surface reference characteristics;

[0103] The surface analysis unit is configured with a surface analysis strategy, and the surface analysis strategy includes:

[0104] The thermal paper tear resistance strength online detection model is constructed, the thermal paper with known quality is taken as a sample paper, the quality of the sample paper is named as a sample quality, the surface characteristics of the sample paper are extracted and named as surface sample characteristics, and the sample quality and the surface sample characteristics correspond to each other;

[0105] The contrast and the correlation in the surface sample characteristics are respectively named as a sample contrast and a sample correlation;

[0106] The range of the sample contrast of the sample quality is qualified to obtain a first reference range, the range of the sample correlation of the sample quality is qualified to obtain a second reference range, and the first reference range and the second reference range jointly constitute the surface reference characteristics;

[0107] ​In practical applications, since the contrast reflects the sharpness and the texture of the image, and the correlation reflects the similarity of the gray level of the image in the row or column direction, they both intuitively reflect the characteristics of the surface of the thermal paper, when the tear resistance strength is unqualified, it represents that there is a defect on the surface of the thermal paper, and this part of the defect will cause a large change in the correlation and the contrast, so the range of the correlation and the contrast is directly counted, thereby obtaining the first reference range and the second reference range, if the contrast is in the first reference range and the correlation is in the second reference range, it represents that the surface of the thermal paper meets the characteristics of the thermal paper with normal tear resistance strength, but in special cases, the machine vision cannot detect the change of the tear resistance strength, so the scheme of the machine vision has high requirements for the light source and the shooting equipment, and a slight deviation will cause an error in the detection result, therefore, the recognition of the audio waveform is added in the embodiment to verify whether the tear resistance strength of the thermal paper is normal.

[0108] The audio analysis unit is used to collect the audio features of the thermal paper with known quality and perform deep learning to obtain audio reference features;

[0109] The audio analysis unit is configured with an audio analysis strategy, and the audio analysis strategy includes:

[0110] The audio features of the sample paper with qualified sample quality are collected and named as audio sample features, and the peak analysis fold line and the valley analysis fold line in the audio sample features are respectively named as the peak sample fold line and the valley sample fold line;

[0111] The leftmost endpoint of the peak sample fold line is named as the peak starting point, the leftmost endpoint of the valley sample fold line is named as the valley starting point, and the turning points and endpoints in the peak sample fold line and the valley sample fold line are named as fold line vertices;

[0112] Please refer to Figure 7 The peak starting points of all the peak sample fold lines are overlapped, and the overlapped image is named as a peak learning graph; the valley starting points of all the valley sample fold lines are overlapped, and the overlapped image is named as a valley learning graph;

[0113] When performing deep learning on the peak learning graph or the valley learning graph, the peak learning graph or the valley learning graph being analyzed is named as a target analysis graph, and the X axis and the Y axis corresponding to the target analysis graph are respectively named as a first axis and a second axis;

[0114] In practical applications, since the analysis processes of the peak sample fold line and the valley sample fold line are completely the same, the analysis process of the peak sample fold line is taken as an example to illustrate the subsequent analysis process in the embodiment; since the data amount is too large to be observed in the embodiment, only a small amount of data is taken as an example to construct a peak learning graph as shown in Figure 7 .Figure 7 That is, the target analysis diagram. Figure 7 The X-axis is the first axis, and the Y-axis is the second axis.

[0115] Please see Figure 8 As shown, for each value of the first axis, the topmost and bottommost polyline vertices are obtained and named the upper limit point and lower limit point, respectively.

[0116] The upper and lower limits are numbered from left to right, and are respectively represented by the symbol U. i and D j This indicates that i and j are both non-zero natural numbers, i is the index of U, and j is the index of D;

[0117] Set an auxiliary number h, initially 1, starting with i=1, and pass through U via a straight line. i with U i+h Connect the points to obtain the upper limit analysis line. Determine whether the upper limit analysis line intersects with any part of the target analysis diagram except for the upper limit point. If yes, output an upper limit error signal; otherwise, output an upper limit correct signal.

[0118] If an upper limit error signal is output, the upper limit analysis line will be deleted, and U will be... i+h Delete, increment h and reanalyze the upper limit analysis line, repeat the judgment until the upper limit is correctly output; if the upper limit is correctly output, retain the upper limit analysis line, and increment i by 1, then judge U. i Does it exist, if U i If it does not exist, increment i again; if U i If it exists, reset h to 1 and analyze another upper limit analysis line again until the maximum value of i is reached, to obtain different upper limit analysis lines. The upper limit analysis lines form a new polyline segment, which is named the upper limit reference line.

[0119] In practical applications, the upper and lower limits are obtained as follows: Figure 8 As shown, U is obtained through numbering. i and D j In this context, i is actually equal to j, and 1 ≤ i = j ≤ 4. An auxiliary number h is set, initially set to 1. When i = 1, U1 and U2 are connected to obtain the upper limit analysis line. Within the range of the first axis where the upper limit analysis line is located, if all values ​​in the peak learning graph are less than or equal to the upper and lower analysis lines, then the upper limit correct signal is output. In fact, it means that all broken lines within the range of the first axis where the upper limit analysis line is located are not above the upper limit analysis line. Under normal circumstances, the upper limit error signal will not be output. Setting the upper limit error signal is only to prevent errors from occurring during the analysis process, so that errors can be detected in time and the analysis can be re-performed.

[0120] Reset h to 1, starting with j=1, and move D through a straight line. j With D j+h Connect the points to obtain the lower limit analysis line. Determine whether the lower limit analysis line intersects with any part of the target analysis diagram except for the lower limit point. If yes, output the lower limit error signal; otherwise, output the lower limit correct signal.

[0121] If a lower limit error signal is output, the lower limit analysis line will be deleted, and D will be... j+h Delete, increment h and re-analyze the lower limit analysis line, repeat the judgment until the lower limit correct signal is output; if the lower limit correct signal is output, retain the lower limit analysis line, and increment j by 1, then judge D. j Does it exist if D j If it does not exist, increment j again. If D j If it exists, reset h to 1 and analyze another lower limit analysis line again until the maximum value of j is reached, to obtain different lower limit analysis lines. The lower limit analysis lines form a new line segment, which is named the lower limit reference line.

[0122] Please see Figure 9 As shown, the leftmost endpoints of the upper limit reference line and the lower limit reference line are named the upper limit left endpoint and the lower limit left endpoint, respectively. The rightmost endpoints of the upper limit reference line and the lower limit reference line are named the upper limit right endpoint and the lower limit right endpoint, respectively. The upper limit left endpoint and the lower limit left endpoint are connected, and the upper limit right endpoint and the lower limit right endpoint are connected to obtain a closed region, which is named the target reference feature.

[0123] The target reference features obtained from the analysis of the peak learning map and the peak-valley learning map are named peak reference features and peak-valley reference features, respectively. The peak reference features and peak-valley reference features together constitute the sound wave reference features.

[0124] In practical applications, the process of analyzing the lower limit reference line is exactly the same as that of analyzing the upper limit reference line. This embodiment will not provide further details. The final analysis yields the peak reference characteristics as follows: Figure 9 As shown, Figure 9 The gray area in the image represents the peak reference feature.

[0125] The tear resistance strength assessment module is used to analyze surface features and acoustic features through an online tear resistance strength detection model for thermal paper to determine whether the tear resistance strength of the thermal paper is qualified.

[0126] The tear strength assessment module is configured with a tear strength assessment strategy, which includes:

[0127] The surface feature and the acoustic feature currently required to be analyzed are respectively named as surface real-time feature and acoustic real-time feature, the contrast and the correlation in the surface real-time feature are respectively named as real-time contrast and real-time correlation, and the wave peak analysis broken line and the peak valley analysis broken line in the acoustic real-time feature are respectively named as wave peak real-time broken line and peak valley real-time broken line;

[0128] If the real-time contrast is in the first reference range, the real-time correlation is in the second reference range, the wave peak real-time broken line is all in the wave peak reference feature, and the peak valley real-time broken line is all in the peak valley reference feature, an anti-tear strength qualified signal is output, otherwise an anti-tear strength unqualified signal is output;

[0129] In actual application, the process of judging the anti-tear strength is clearly stated in the anti-tear strength judgment strategy, and only when all conditions are met can the anti-tear strength be judged to be qualified, which will not be specifically described in this embodiment.

[0130] Embodiment 2, please refer to Figure 10 As shown in the figure, the application provides a machine vision-based thermal paper anti-tear strength online detection method, which includes the following steps:

[0131] Step S1, install a light source and a camera on the thermal paper production line for collecting paper images; step S1 includes the following sub-steps:

[0132] Step S101, the light source uses a non-flickering LED linear light source, and the camera uses a linear array camera;

[0133] Step S102, the LED linear light source irradiates the thermal paper at a first angle, and the linear array camera is located directly above the thermal paper to take pictures of the thermal paper, obtaining a paper image;

[0134] Step S2, install a paper beating device on the thermal paper production line for beating the thermal paper and recording the waveform of the sound generated when the thermal paper is beaten, which is named as audio waveform; step S2 includes the following sub-steps:

[0135] Step S201, the paper beating device includes a front conveying belt, a rear conveying belt, a front fixing device, a rear fixing device, and a beating device;

[0136] Step S202, the front fixing device is installed above the front conveying belt, and the rear fixing device is installed above the rear conveying belt;

[0137] In step S203, before the thermal paper is tapped, the front conveyor belt stops running, the front fixing device presses down to fix the front end of the thermal paper, the rear conveyor belt stops running, the rear fixing device presses down to fix the rear end of the thermal paper, and then the rear conveyor belt is started. The rear fixing device can move synchronously with the rear conveyor belt. At this time, a protrusion will be generated on the thermal paper between the front fixing device and the rear fixing device. The tapping device taps the protrusion and collects the waveform of the sound generated by the thermal paper during the tapping to obtain the audio waveform.

[0138] Step S3 involves performing feature analysis on the paper image to extract the surface features of the thermal paper. Step S3 includes the following sub-steps:

[0139] Step S301: Perform grayscale processing on the paper image to obtain a grayscale image;

[0140] Step S302: Obtain the gray-level co-occurrence matrix of the gray-level image, and extract the contrast and correlation in the gray-level co-occurrence matrix as the surface features of the thermal paper.

[0141] Step S4 involves performing feature analysis on the audio waveform to obtain the acoustic characteristics of the thermal paper. Step S4 includes the following sub-steps:

[0142] Step S401: The audio waveform is stored in the audio graph. The X-axis of the audio graph represents time, and the Y-axis represents amplitude. The coordinates of the peak of the audio waveform are obtained and named as peak points. The coordinates of the trough of the audio waveform are obtained and named as trough points.

[0143] Step S402: Number the peaks and valleys according to the chronological order, and use the symbol F respectively. n and G n This represents the expression, where n is a non-zero natural number and n is the index of F and G. Typically, the number of peaks and valleys is the same, and G... n For located at F n The first valley after that;

[0144] Step S403, pass F through a straight line n With G n Connect the lines and mark the resulting auxiliary lines as L. n , obtain L n The slope, denoted as K n ;

[0145] Step S404, pass through two adjacent Fs by a straight line n Connecting the segments yields a single line segment, which is named the peak analysis line.

[0146] Step S405, F n The corresponding time stamp is T n , will Kn With T n Related, with T n K is the horizontal axis. n Establish a Cartesian coordinate system for the vertical axis, named the Peak-Valley Analysis Chart, and set K... n According to T n Enter the peak-valley analysis chart and name the coordinate points in the chart as peak-valley analysis points;

[0147] Step S406, K n With T n The coordinate point formed is labeled P. n Through a straight line to two adjacent P n Connecting the segments yields a broken line segment, which is named the Peak-Valley Analysis Broken Line.

[0148] Step S407, the peak analysis line and the peak-valley analysis line are the sound wave characteristics;

[0149] Step S5: Construct an online detection model for the tear resistance of thermal paper, collect surface features and acoustic features of thermal paper of known quality, and perform deep learning; Step S5 includes the following sub-steps:

[0150] Step S501: Construct an online detection model for the tear resistance of thermal paper, collect surface features of thermal paper of known quality and perform deep learning to obtain surface reference features;

[0151] Step S501 includes the following sub-steps:

[0152] Step S501.01: Construct an online detection model for the tear resistance of thermal paper, obtain thermal paper of known quality as sample paper, name the quality of the sample paper as sample quality, extract the surface features of the sample paper and name them as surface sample features, and the sample quality corresponds to the surface sample features.

[0153] Step S501.02: Name the contrast and correlation in the surface sample features as sample contrast and sample correlation, respectively.

[0154] Step S501.03: Calculate the range of contrast of samples with acceptable sample quality to obtain the first reference range; calculate the range of correlation of samples with acceptable sample quality to obtain the second reference range; the first reference range and the second reference range together constitute the surface reference feature.

[0155] Step S502: Collect the acoustic wave characteristics of thermal paper of known quality and perform deep learning to obtain acoustic wave reference characteristics;

[0156] Step S502 includes the following sub-steps:

[0157] Step S502.01, collect the sound wave features of the sample paper with qualified sample quality, name it as sound wave sample feature, name the peak analysis fold line and the trough analysis fold line in the sound wave sample feature as peak sample fold line and trough sample fold line respectively;

[0158] Step S502.02, name the leftmost end point of the peak sample fold line as peak starting point, name the leftmost end point of the trough sample fold line as trough starting point, and name the turning points and end points in the peak sample fold line and the trough sample fold line as fold line vertexes;

[0159] Step S502.03, overlap the peak starting points of all peak sample fold lines, name the overlapped image as peak learning image; overlap the trough starting points of all trough sample fold lines, name the overlapped image as trough learning image;

[0160] Step S502.04, when deep learning is performed on the peak learning image or the trough learning image, name the peak learning image or the trough learning image currently analyzed as target analysis image, name the X axis and the Y axis corresponding to the target analysis image as first axis and second axis respectively;

[0161] Step S502.05, for each value of the first axis, obtain the fold line vertexes at the uppermost and the lowermost, name them as upper limit point and lower limit point respectively;

[0162] Step S502.06, number the upper limit points and the lower limit points in the order from left to right, represent them by symbols U i and D j respectively, where i and j are both non-zero natural numbers, i is the serial number of U, and j is the serial number of D;

[0163] Step S502.07, set an auxiliary number h, h is initially 1, start with i = 1, connect U i and U i+h by a straight line to obtain upper limit analysis line, judge whether the upper limit analysis line intersects with other parts of the target analysis image except the upper limit points, if yes, output upper limit error signal, if no, output upper limit correct signal;

[0164] Step S502.08, if the upper limit error signal is output, delete the upper limit analysis line, delete U i+h at the same time, increase h by 1 and analyze the upper limit analysis line again, repeat the judgment until the upper limit correct signal is output; if the upper limit correct signal is output, keep the upper limit analysis line, increase i by 1 at the same time, judge whether U i exists, if U i does not exist, increase i by 1 again, if U iIf D exists, h is reset to 1 and another upper limit analysis line is analyzed again until the maximum value of i is reached, different upper limit analysis lines are obtained, a new polyline segment is composed of the upper limit analysis lines, and is named as an upper limit reference line;

[0165] Step S502.09, h is reset to 1, j is started as 1, and D is connected by a straight line to obtain a lower limit analysis line. It is judged whether the lower limit analysis line intersects with other parts of the target analysis graph except the lower limit point. If yes, a lower limit error signal is output. If no, a lower limit correct signal is output. j is connected with D j+h to obtain a lower limit analysis line. It is judged whether the lower limit analysis line intersects with other parts of the target analysis graph except the lower limit point. If yes, a lower limit error signal is output. If no, a lower limit correct signal is output.

[0166] Step S502.10, if the lower limit error signal is output, the lower limit analysis line is deleted, D j+h is deleted, h is added by 1 and the lower limit analysis line is analyzed again. The judgment is repeated until the lower limit correct signal is output. If the lower limit correct signal is output, the lower limit analysis line is kept, j is added by 1, and it is judged whether D j exists. If D j does not exist, j is added by 1 again. If D j exists, h is reset to 1 and another lower limit analysis line is analyzed again until the maximum value of j is reached. Different lower limit analysis lines are obtained, a new polyline segment is composed of the lower limit analysis lines, and is named as a lower limit reference line.

[0167] Step S502.11, the leftmost end point of the upper limit reference line and the lower limit reference line is respectively named as an upper limit left end point and a lower limit left end point. The rightmost end point of the upper limit reference line and the lower limit reference line is respectively named as an upper limit right end point and a lower limit right end point. The upper limit left end point is connected with the lower limit left end point, and the upper limit right end point is connected with the lower limit right end point to obtain a closed area, which is named as a target reference feature.

[0168] Step S502.12, the target reference features obtained by analyzing the peak learning graph and the peak-valley learning graph are respectively named as a peak reference feature and a peak-valley reference feature. The peak reference feature and the peak-valley reference feature together compose an acoustic wave reference feature.

[0169] Step S6, the surface feature and the acoustic wave feature are analyzed by the thermal paper tear resistance strength online detection model to judge whether the tear resistance strength of the thermal paper is qualified. Step S6 includes the following sub-steps:

[0170] Step S601, the surface feature and the acoustic wave feature currently needed to be analyzed are respectively named as a surface real-time feature and an acoustic wave real-time feature. The contrast and the correlation in the surface real-time feature are respectively named as a real-time contrast and a real-time correlation. The peak analysis polyline and the peak-valley analysis polyline in the acoustic wave real-time feature are respectively named as a peak real-time polyline and a peak-valley real-time polyline.

[0171] In step S602, if the real-time contrast is within the first reference range, the real-time correlation is within the second reference range, all the real-time peak fold lines are within the peak reference feature, and all the real-time trough fold lines are within the trough reference feature, an anti-tearing strength qualified signal is output, otherwise an anti-tearing strength unqualified signal is output.

[0172] In embodiment 3, the electronic device provided by the application can include a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory can communicate with each other through the communication bus. The memory stores computer readable instructions, and the processor can call the instructions in the memory. When the computer readable instructions are executed by the processor, the steps in the method for online detection of the anti-tearing strength of thermal paper based on machine vision are run to realize the following functions: collecting a paper image; recording an audio waveform generated when the thermal paper is hit; extracting surface features of the thermal paper in the paper image; obtaining audio features of the thermal paper; constructing an online detection model for the anti-tearing strength of thermal paper, collecting surface features and audio features of thermal paper of known quality and performing deep learning; analyzing the surface features and the audio features through the online detection model for the anti-tearing strength of thermal paper to determine whether the anti-tearing strength of the thermal paper is qualified.

[0173] In addition, the logical instructions in the memory described above can be implemented in the form of a software functional unit and sold or used as an independent product, and can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the application essentially or the part that contributes to the prior art or part of the technical solutions can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in the various embodiments of the application. The foregoing storage medium includes a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various media that can store program codes.

[0174] In embodiment 4, the application further provides a computer readable storage medium, and the application provides a storage medium having a computer program stored thereon, the computer program being executed by a processor to run the steps of the above machine vision-based online detection method for tear resistance of thermal paper to realize the following functions: collecting a paper image; recording an audio waveform generated when the thermal paper is hit; extracting surface features of the thermal paper in the paper image; obtaining audio features of the thermal paper; constructing an online detection model for tear resistance of thermal paper, collecting surface features and audio features of thermal paper of known quality and performing deep learning; and analyzing the surface features and the audio features by the online detection model for tear resistance of thermal paper to determine whether the tear resistance of the thermal paper is qualified.

[0175] Through the description of the above embodiments, the embodiments of the application can be provided as a method, a system or a computer program product. Based on such understanding, the above technical solutions can be embodied in the form of a software product, which can be stored in a computer readable storage medium, such as a ROM / RAM, a magnetic disk, an optical disk, etc., and includes a plurality of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute the method described in each embodiment or some parts of the embodiment.

[0176] In the embodiments provided in the present application, it should be understood that the disclosed system or method can be implemented in other ways. The above described embodiments are only illustrative, for example, the division of the modules or units is only a logical function division, and in actual implementation, another division mode can be used, for example, a plurality of modules or units can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the displayed or discussed each other can be through some communication interface, indirect coupling or communication connection between the systems, modules and units can be electrical, mechanical or other forms.

[0177] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for some technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for online detection of tear resistance of thermal paper based on machine vision, characterized in that, The method comprises the following steps: installing a light source and a camera on a thermal paper production line for collecting a paper image; installing a paper beating device on the thermal paper production line for beating the thermal paper and recording a waveform of a sound generated when the thermal paper is beaten, named as an audio waveform; performing feature analysis on the paper image to extract surface features of the thermal paper in the paper image; performing feature analysis on the audio waveform to obtain sound wave features of the thermal paper; constructing a thermal paper tear resistance strength online detection model, collecting surface features and sound wave features of thermal paper of known quality and performing deep learning; analyzing the surface features and the sound wave features through the thermal paper tear resistance strength online detection model to determine whether the tear resistance strength of the thermal paper is qualified; installing a light source and a camera on a thermal paper production line for collecting a paper image comprises the following sub-steps: the light source is a non-flickering LED linear light source, and the camera is a linear array camera; the LED linear light source irradiates the thermal paper at a first angle, and the linear array camera is located directly above the thermal paper to take a picture of the thermal paper, thereby obtaining a paper image; installing a paper beating device on the thermal paper production line for beating the thermal paper and recording a waveform of a sound generated when the thermal paper is beaten comprises the following sub-steps: the paper beating device comprises a front conveying belt, a rear conveying belt, a front fixing device, a rear fixing device, and a beating device; the front fixing device is installed above the front conveying belt, and the rear fixing device is installed above the rear conveying belt; before beating the thermal paper, the front conveying belt stops running, the front fixing device presses down to fix the front end of the thermal paper, the rear conveying belt stops running, the rear fixing device presses down to fix the rear end of the thermal paper, then the rear conveying belt is started, the rear fixing device can move synchronously with the rear conveying belt, at this time, the thermal paper between the front fixing device and the rear fixing device will have a protrusion, the protrusion is beaten by the beating device, and a waveform of a sound generated during beating of the thermal paper is collected, thereby obtaining an audio waveform; performing feature analysis on the paper image to extract surface features of the thermal paper in the paper image comprises the following sub-steps: performing grayscale processing on the paper image to obtain a grayscale image; obtaining a gray level co-occurrence matrix of the grayscale image, and extracting contrast and correlation in the gray level co-occurrence matrix as surface features of the thermal paper; performing feature analysis on the audio waveform to obtain sound wave features of the thermal paper comprises the following sub-steps: the audio waveform is stored in an audio graph, the X-axis of the audio graph represents time, and the Y-axis represents amplitude, coordinates of a wave peak of the audio waveform are obtained, named as a peak point, and coordinates of a wave trough of the audio waveform are obtained, named as a valley point; The peak points and the valley points are numbered in chronological order, and are represented by symbols F n and G n respectively, wherein n is a non-zero natural number and n is the serial number of F and G, usually, the number of peak points and valley points is the same, and G n is the first valley point after F n . F through the straight line n With G n Connect the lines and mark the resulting auxiliary lines as L. n , obtain L n The slope, denoted as K n ; Two F n The connection is made, and finally a polyline segment is obtained, which is named as the peak analysis polyline. F n The corresponding time mark is T n K n is associated with T n , and a plane rectangular coordinate system is established with T n as the horizontal axis and K n as the vertical axis, named as a peak-valley analysis diagram, K n is recorded in the peak-valley analysis diagram according to T n , and the coordinate points in the peak-valley analysis diagram are named as peak-valley analysis points; K n with T n The coordinate point formed by the intersection of the straight line and the straight line is marked as P n , and the two adjacent P n are connected by a straight line, and finally a polyline segment is obtained, named peak-valley analysis polyline. the wave peak analysis fold line and the peak-valley analysis fold line are sound wave features. 2.The machine vision-based online detection method of tear resistance strength of thermal paper according to claim 1, characterized in that, constructing a thermal paper tear resistance strength online detection model, collecting surface features and sound wave features of thermal paper of known quality and performing deep learning comprises the following sub-steps: constructing a thermal paper tear resistance strength online detection model, collecting surface features of thermal paper of known quality and performing deep learning to obtain surface reference features; collecting sound wave features of thermal paper of known quality and performing deep learning to obtain sound wave reference features. 3.The machine vision-based online detection method of tear resistance strength of thermal paper according to claim 2, characterized in that, The surface reference feature of the thermal paper is obtained by collecting the surface feature of the thermal paper with known quality and deep learning, and the surface reference feature includes the following sub-steps: The surface reference feature of the thermal paper is obtained by collecting the surface feature of the thermal paper with known quality and deep learning, and the surface reference feature includes the following sub-steps: The contrast and correlation in the surface sample feature are named as sample contrast and sample correlation respectively. The range of the sample contrast of the sample quality is calculated, and the first reference range is obtained; the range of the sample correlation of the sample quality is calculated, and the second reference range is obtained; and the first reference range and the second reference range jointly constitute the surface reference feature. 4.The machine vision-based online detection method of tear resistance strength of thermal paper according to claim 3, characterized in that, The sound wave reference feature is obtained by collecting the sound wave feature of the thermal paper with known quality and deep learning, and the sound wave reference feature includes the following sub-steps: The sound wave feature of the sample paper with qualified sample quality is collected and named as sound wave sample feature; the wave peak analysis fold line and the peak valley analysis fold line in the sound wave sample feature are named as wave peak sample fold line and peak valley sample fold line respectively; The leftmost endpoint of the wave peak sample fold line is named as wave peak starting point, and the leftmost endpoint of the peak valley sample fold line is named as peak valley starting point; the turning points and endpoints in the wave peak sample fold line and the peak valley sample fold line are named as fold line vertexes; The wave peak starting points of all wave peak sample fold lines are overlapped, and the overlapped image is named as wave peak learning image; The peak valley starting points of all peak valley sample fold lines are overlapped, and the overlapped image is named as peak valley learning image; When deep learning is performed on the wave peak learning image or the peak valley learning image, the currently analyzed wave peak learning image or peak valley learning image is named as target analysis image, and the X axis and Y axis corresponding to the target analysis image are named as first axis and second axis respectively; For each value of the first axis, the uppermost fold line vertex and the lowermost fold line vertex are obtained and named as upper limit point and lower limit point respectively; The upper and lower points are numbered in the order from left to right, respectively by the symbols U i and D j , where i and j are both non-zero natural numbers and i is the ordinal number of U and j is the ordinal number of D; An auxiliary number h is set, which is initially 1, starting with i = 1, a straight line is drawn through U i connected with U i+h to obtain an upper limit analysis line, it is determined whether the upper limit analysis line intersects other parts of the target analysis graph except the upper limit point, if yes, an upper limit error signal is output, and if no, an upper limit correct signal is output. If an upper limit error signal is output, the upper limit analysis line will be deleted, and U will be... i+h Delete, increment h and reanalyze the upper limit analysis line, repeat the judgment until the upper limit is correctly output; if the upper limit is correctly output, retain the upper limit analysis line, and increment i by 1, then judge U. i Does it exist, if U i If it does not exist, increment i again; if U i If it exists, reset h to 1 and analyze another upper limit analysis line again until the maximum value of i is reached, to obtain different upper limit analysis lines. The upper limit analysis lines form a new polyline segment, which is named the upper limit reference line. h is reset to 1, and j is set to 1, a straight line is drawn through D j connected with D j+h , to obtain a lower limit analysis line, and it is determined whether the lower limit analysis line intersects with other parts of the target analysis graph except the lower limit point. If yes, a lower limit error signal is output, and if no, a lower limit correct signal is output. If the lower limit error signal is output, the lower limit analysis line is deleted, and D j+h is added to h and the lower limit analysis line is analyzed again, and the judgment is repeated until the lower limit correct signal is output; if the lower limit correct signal is output, the lower limit analysis line is retained, and j is added to 1, and D j is judged; if D j exists, j is added to 1 again, and D j is judged; if D j exists, h is reset to 1 and another lower limit analysis line is analyzed again until the maximum value of j is reached, different lower limit analysis lines are obtained, a new broken line segment is composed of the lower limit analysis lines, and is named as a lower limit reference line; The leftmost endpoints of the upper limit reference line and the lower limit reference line are named as upper limit left endpoint and lower limit left endpoint respectively, and the rightmost endpoints of the upper limit reference line and the lower limit reference line are named as upper limit right endpoint and lower limit right endpoint respectively; the upper limit left endpoint is connected with the lower limit left endpoint, and the upper limit right endpoint is connected with the lower limit right endpoint, to obtain a closed area, which is named as target reference feature; The target reference features obtained by analyzing the wave peak learning image and the peak valley learning image are named as wave peak reference feature and peak valley reference feature respectively, and the wave peak reference feature and the peak valley reference feature jointly constitute the sound wave reference feature. 5.The machine vision-based online detection method of tear resistance strength of thermal paper according to claim 4, characterized in that, The surface feature and the sound wave feature are analyzed by the thermal paper tear resistance strength online detection model to determine whether the tear resistance strength of the thermal paper is qualified, and the method includes the following sub-steps: The surface feature and the sound wave feature to be analyzed at present are respectively named as surface real-time feature and sound wave real-time feature, the contrast and the correlation in the surface real-time feature are respectively named as real-time contrast and real-time correlation, and the wave peak analysis fold line and the peak valley analysis fold line in the sound wave real-time feature are respectively named as wave peak real-time fold line and peak valley real-time fold line; If the real-time contrast is in the first reference range, the real-time correlation is in the second reference range, the wave peak real-time fold line is all in the wave peak reference feature, and the peak valley real-time fold line is all in the peak valley reference feature, a tear resistance strength qualified signal is output, otherwise a tear resistance strength unqualified signal is output.

6. The machine vision-based online detection system for tear resistance of thermal paper according to any one of claims 1-5, characterized in that, The image acquisition module, the audio acquisition module, the surface feature extraction module, the sound wave feature extraction module, the feature analysis module and the tear resistance strength judgment module are included; the image acquisition module, the audio acquisition module, the surface feature extraction module, the sound wave feature extraction module and the tear resistance strength judgment module are respectively connected with the feature analysis module in data; The image acquisition module is used for installing a light source and a camera on a thermal paper production line to collect a paper image; The audio acquisition module is used for installing a paper slapping device on the thermal paper production line to slap the thermal paper and record the waveform of the sound generated when the thermal paper is slapped, which is named as an audio waveform; The surface feature extraction module is used for analyzing the features of the paper image and extracting the surface features of the thermal paper in the paper image; The sound wave feature extraction module is used for analyzing the features of the audio waveform and obtaining the sound wave features of the thermal paper; The feature analysis module is used for constructing a thermal paper tear resistance strength online detection model, collecting the surface features and the sound wave features of the thermal paper with known quality and performing deep learning; The tear resistance strength judgment module is used for analyzing the surface features and the sound wave features through the thermal paper tear resistance strength online detection model to determine whether the tear resistance strength of the thermal paper is qualified.

Images