Industrial belt longitudinal tearing identification method and system based on high-speed pulse camera

Abstract

The invention relates to the technical field of industrial belt longitudinal tearing recognition, in particular to an industrial belt longitudinal tearing recognition method and system based on a high-speed pulse camera, and the method comprises the steps: carrying out the video collection of the front surface of a target industrial belt based on the high-speed pulse camera, so as to obtain a target industrial belt operation video; performing image acquisition on the side surface of the target industrial belt according to a preset sampling period based on the high-speed pulse camera to obtain a side surface image of the target industrial belt, and performing data acquisition on the target industrial belt based on an acceleration sensor to obtain a vibration signal corresponding to the target industrial belt. According to the method, the dynamic state of the industrial belt is comprehensively known through combination of video and image acquisition, the working state of the industrial belt is comprehensively monitored through front and side multi-angle observation, potential problems possibly neglected by a traditional method can be recognized through the multi-dimensional information collection mode, and the working state of the industrial belt is comprehensively monitored. Therefore, the timeliness and accuracy of fault diagnosis are improved.

Description

Technical Field

[0001] The invention relates to the technical field of industrial belt longitudinal tear recognition, and in particular to an industrial belt longitudinal tear recognition method and system based on a high-speed pulse camera. Background Art

[0002] Longitudinal tearing of industrial belts refers to the tearing or splitting of the belt along its length during operation. This situation is usually caused by overload, friction, material fatigue or improper installation. Longitudinal tearing will not only affect the normal operation of the belt, but may also cause equipment failure and safety hazards.

[0003] Traditional methods usually rely on a single monitoring method, such as vibration monitoring or visual inspection alone, which cannot fully understand the dynamic state of the belt, resulting in potential problems being overlooked; and many traditional detection methods have low sensitivity in identifying instantaneous anomalies, making it difficult to capture tiny, short-term precursors to failures, which may delay fault diagnosis and treatment; and traditional methods often rely on manual analysis and judgment, with slow data processing speeds, and are unable to monitor in real time and respond quickly, which can easily lead to increased downtime; and traditional methods have low accuracy in identifying cracks and damage, and often require problems to be discovered after a failure occurs, which increases the cost of repair and replacement. Summary of the invention

[0004] The purpose of the present invention is to address the problems existing in the background technology and to propose an industrial belt longitudinal tear recognition method and system based on a high-speed pulse camera.

[0005] The technical solution of the present invention is a method for identifying longitudinal tearing of industrial belts based on a high-speed pulse camera, comprising:

[0006] Based on the high-speed pulse camera, the front of the target industrial belt is video-captured to obtain the target industrial belt operation video; based on the high-speed pulse camera, the side of the target industrial belt is image-captured according to a preset sampling period to obtain the side image of the target industrial belt; based on the acceleration sensor, the data of the target industrial belt is collected to obtain the vibration signal corresponding to the target industrial belt;

[0007] Performing mutation frame detection on the target industrial belt operation video to obtain a mutation frame set corresponding to the target industrial belt operation video, and performing anomaly detection on the mutation frame set corresponding to the target industrial belt operation video based on a trained anomaly detection model to obtain a first anomaly recognition parameter;

[0008] Performing image preprocessing on the target industrial belt side image to obtain a target industrial belt side standard image, and performing crack segmentation on the target industrial belt side standard image based on a trained crack segmentation model to obtain a crack image corresponding to the target industrial belt side standard image;

[0009] Extracting characterization features of the crack image corresponding to the standard image of the target industrial belt side surface to obtain a characterization feature set corresponding to the crack image, performing average weighting on the characterization features in the characterization feature set to obtain a crack coefficient corresponding to the crack image, and comparing the crack coefficient with a preset crack coefficient threshold to obtain a second abnormality recognition parameter;

[0010] Performing feature extraction on the vibration signal corresponding to the target industrial belt to obtain a vibration feature set corresponding to the vibration signal, and performing vibration detection on the vibration feature set corresponding to the vibration signal based on a trained vibration classification model to obtain a third abnormality recognition parameter;

[0011] An abnormality alarm is performed based on the first abnormality identification parameter, the second abnormality identification parameter and the third abnormality identification parameter.

[0012] Preferably, performing mutation frame detection on the target industrial belt operation video to obtain a set of mutation frames corresponding to the target industrial belt operation video includes:

[0013] Acquire a current target industrial belt operation video frame from the target industrial belt operation video, and extract feature points of the current target industrial belt operation video frame to obtain a feature point set corresponding to the current target industrial belt operation video frame;

[0014] Combining the feature point set corresponding to the current target industrial belt operation video frame with the feature point set corresponding to the previous target industrial belt operation video frame to obtain a matching point set;

[0015] The rotation angle of each matching point in the matching point set is calculated to obtain a rotation angle set corresponding to the matching point set. The calculation formula of the rotation angle is as follows:

[0016]

[0017] Among them, θ i represents the rotation angle corresponding to the i-th matching point in the matching point set, (Q i,y ,P i,y ) represents the ordinate of the feature point corresponding to the previous target industrial belt operation video frame and the current target industrial belt operation video frame in the i-th matching point in the matching point set, (Qi,x ,P i,x ) represents the horizontal coordinate of the feature point corresponding to the previous target industrial belt operation video frame and the current target industrial belt operation video frame in the i-th matching point in the matching point set;

[0018] The rotation angle set corresponding to the matching point set is subjected to histogram conversion to obtain a direction histogram corresponding to the matching point set, wherein the abscissa of the direction histogram represents the rotation angle, and the ordinate of the direction histogram represents the cumulative number of the rotation angle.

[0019] Preferably, performing mutation frame detection on the target industrial belt operation video to obtain a mutation frame set corresponding to the target industrial belt operation video also includes:

[0020] The angle probability histogram corresponding to the matching point set is obtained based on the direction histogram corresponding to the matching point set, wherein the abscissa of the angle probability histogram represents the rotation angle, the ordinate of the angle probability histogram represents the probability accumulation number of the rotation angle, and the angle probability is calculated as follows:

[0021]

[0022] Among them, P(θ i ) represents the occurrence probability of the rotation angle corresponding to the i-th matching point in the matching point set, and N represents the total number of matching points in the matching point set;

[0023] Obtaining the rotation angle corresponding to the maximum probability cumulative number of the rotation angle based on the angle probability histogram corresponding to the matching point set;

[0024] The rotation angle corresponding to the maximum probability cumulative number of the rotation angle is compared with a preset rotation angle threshold. If it is greater than the preset rotation angle threshold, it indicates that the current target industrial belt operation video frame is a mutation frame. Otherwise, it indicates that the current target industrial belt operation video frame is not a mutation frame.

[0025] The above operation is repeated until all video frames in the target industrial belt operation video are traversed to obtain a set of mutation frames corresponding to the target industrial belt operation video.

[0026] Preferably, the target industrial belt side image is subjected to image preprocessing to obtain a target industrial belt side standard image, including:

[0027] Binarizing the target industrial belt side image to obtain a target industrial belt side binary image;

[0028] Traversing each pixel point in the binary image of the side of the target industrial belt, and when traversing to the current pixel point, determining whether the current pixel point is an isolated point;

[0029] If the current pixel point is an isolated point, the current pixel point is deleted; otherwise, the next pixel point of the current pixel point is traversed until each pixel point in the target industrial belt side binary image is traversed to obtain the target industrial belt side standard image.

[0030] Preferably, the crack segmentation model includes a feature extraction module, a segmentation accuracy improvement module and a crack prediction module. The feature extraction module adopts a U-Net network and is used to extract the feature map of the standard image of the industrial belt side. The segmentation accuracy improvement module adopts a dilation operation and a corrosion operation. The dilation operation is used to merge the disconnected feature maps, and the corrosion operation is used to separate the connection areas between different feature maps. The crack prediction module is used to predict the feature map after accuracy improvement based on an upsampling operation to obtain a crack image corresponding to the target industrial belt side standard image.

[0031] Preferably, the formula for the expansion operation is as follows:

[0032]

[0033] Among them, I represents the input feature, W d represents the dilation filter, represents the dilation operation, (i, j, k) represents the size of the input feature, S1, S2, S3 represent the total size of the feature, and S1 = {1, 2, ..., a}, S2 = {1, 2, ..., b}, S3 = {1, 2, ..., c} represents, (a, b, c) represents the size of the dilation filter;

[0034] The formula for the corrosion operation is as follows:

[0035]

[0036] Where W represents the corrosion filter, Represents an erosion operation.

[0037] Preferably, the characterization feature extraction is performed on the crack image corresponding to the standard image of the target industrial belt side surface to obtain a characterization feature set corresponding to the crack image, including:

[0038] Acquire the pixel area of ​​the crack image, and use the pixel area of ​​the crack image as a first characterization feature;

[0039] Acquire an outer frame of the crack image, and use the length and width of the outer frame of the crack image as a second characterization feature;

[0040] Counting edge pixels of the crack image to obtain total edge pixels of the crack image, and using the total edge pixels of the crack image as a third representation feature;

[0041] The first characterizing feature, the second characterizing feature and the third characterizing feature of the crack image are combined to obtain a characterizing feature set corresponding to the crack image.

[0042] Preferably, feature extraction is performed on the vibration signal corresponding to the target industrial belt to obtain a vibration feature set corresponding to the vibration signal, including:

[0043] Predefine a sliding window of fixed length, and move the sliding window in the vibration signal to obtain a sub-vibration signal in the sliding window, wherein the vibration signal includes a plurality of sub-vibration signals;

[0044] Calculating the signal-to-noise ratio of the sub-vibration signal, and matching the signal-to-noise ratio corresponding to the sub-vibration signal with a preset signal-to-noise ratio level table to obtain the signal-to-noise ratio level corresponding to the sub-vibration signal;

[0045] Classifying the sub-vibration signals into different processing sets based on the signal-to-noise ratio levels corresponding to the sub-vibration signals, wherein the different processing sets include a primary processing set, a secondary processing set, and a tertiary processing set;

[0046] Performing feature extraction on the sub-vibration signals in the primary processing set based on Hilbert-Huang transform to obtain a first vibration feature set, performing feature extraction on the sub-vibration signals in the secondary processing set based on resonance sparse decomposition to obtain a second vibration feature set, and performing feature extraction on the sub-vibration signals in the secondary processing set based on symmetric point pattern diagram analysis to obtain a third vibration feature set;

[0047] The first vibration feature set, the second vibration feature set, and the third vibration feature set are combined to obtain a vibration feature set corresponding to the vibration signal.

[0048] Preferably, the anomaly detection model adopts Faster R-CNN, and the vibration classification model adopts support vector machine.

[0049] The technical solution of the present invention is an industrial belt longitudinal tear identification system based on a high-speed pulse camera, which is applicable to the industrial belt longitudinal tear identification method based on a high-speed pulse camera, and comprises:

[0050] A data acquisition unit, the data acquisition unit is used to perform video acquisition on the front of the target industrial belt based on a high-speed pulse camera to obtain an operation video of the target industrial belt, perform image acquisition on the side of the target industrial belt based on the high-speed pulse camera according to a preset sampling period to obtain a side image of the target industrial belt, and perform data acquisition on the target industrial belt based on an acceleration sensor to obtain a vibration signal corresponding to the target industrial belt;

[0051] An operation detection unit, the operation detection unit is used to perform mutation frame detection on the target industrial belt operation video to obtain a mutation frame set corresponding to the target industrial belt operation video, and perform anomaly detection on the mutation frame set corresponding to the target industrial belt operation video based on a trained anomaly detection model to obtain a first anomaly recognition parameter;

[0052] A crack segmentation unit, the crack segmentation unit is used to perform image preprocessing on the target industrial belt side image to obtain a target industrial belt side standard image, and perform crack segmentation on the target industrial belt side standard image based on a trained crack segmentation model to obtain a crack image corresponding to the target industrial belt side standard image;

[0053] A crack detection unit, the crack detection unit is used to extract characterization features of the crack image corresponding to the standard image of the target industrial belt side surface to obtain a characterization feature set corresponding to the crack image, averagely weight the characterization features in the characterization feature set to obtain a crack coefficient corresponding to the crack image, and compare the crack coefficient with a preset crack coefficient threshold to obtain a second abnormality recognition parameter;

[0054] A vibration detection unit, the vibration detection unit is used to perform feature extraction on the vibration signal corresponding to the target industrial belt to obtain a vibration feature set corresponding to the vibration signal, and perform vibration detection on the vibration feature set corresponding to the vibration signal based on a trained vibration classification model to obtain a third abnormality recognition parameter;

[0055] An abnormality alarm unit is used to perform an abnormality alarm based on the first abnormality identification parameter, the second abnormality identification parameter and the third abnormality identification parameter.

[0056] Compared with the prior art, the above technical solution of the present invention has the following beneficial technical effects:

[0057] 1. The present invention combines video and image acquisition to provide a comprehensive understanding of the dynamic state of industrial belts. Through multi-angle observation from the front and side, the working state of industrial belts can be fully monitored. This multi-dimensional information collection method helps to identify potential problems that may be overlooked by traditional methods, thereby improving the timeliness and accuracy of fault diagnosis, and can effectively identify instantaneous anomalies occurring during the operation of industrial belts. These mutation frames are often precursors to faults. By analyzing these abnormal frames and combining them with the trained anomaly detection model, the problem can be quickly located. This process not only improves the sensitivity of anomaly detection, but also provides an important basis for subsequent fault analysis.

[0058] 2. The present invention analyzes the standard image of the side of the target industrial belt through a crack segmentation model, and can accurately identify potential crack problems. The characterization feature extraction and weighted processing of the crack image enable the severity of the crack to be quantified. The comparison of the obtained crack coefficient with the preset threshold further enhances the accuracy of fault detection, which can help operators discover structural defects in advance and take preventive measures before the problem develops into a serious fault. The analysis of the vibration feature set provides another layer of protection for health monitoring. Changes in vibration signals are often an important indication of equipment failure. Timely detection can significantly reduce downtime and maintenance costs caused by failures. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 A schematic flow chart of the steps of an overall method in an embodiment of the present invention;

[0060] Figure 2 A schematic diagram of the system architecture of an overall system in an embodiment of the present invention.

[0061] Figure numerals: 1. Data acquisition unit; 2. Operation detection unit; 3. Crack segmentation unit; 4. Crack detection unit; 5. Vibration detection unit; 6. Abnormal alarm unit. DETAILED DESCRIPTION

[0062] Embodiment 1, as Figure 1 As shown, the method for identifying longitudinal tearing of industrial belts based on a high-speed pulse camera proposed by the present invention comprises:

[0063] S1. Based on a high-speed pulse camera, a video of the front of the target industrial belt is collected to obtain an operation video of the target industrial belt. Based on a high-speed pulse camera, an image of the side of the target industrial belt is collected according to a preset sampling period to obtain a side image of the target industrial belt. Based on an acceleration sensor, data of the target industrial belt is collected to obtain a vibration signal corresponding to the target industrial belt.

[0064] S2. Perform mutation frame detection on the target industrial belt operation video to obtain a mutation frame set corresponding to the target industrial belt operation video, and perform anomaly detection on the mutation frame set corresponding to the target industrial belt operation video based on the trained anomaly detection model to obtain a first anomaly recognition parameter;

[0065] S3, performing image preprocessing on the target industrial belt side image to obtain a target industrial belt side standard image, and performing crack segmentation on the target industrial belt side standard image based on the trained crack segmentation model to obtain a crack image corresponding to the target industrial belt side standard image;

[0066] S4, extracting the characterization features of the crack image corresponding to the standard image of the side of the target industrial belt to obtain a characterization feature set corresponding to the crack image, performing average weighting on the characterization features in the characterization feature set to obtain a crack coefficient corresponding to the crack image, and comparing the crack coefficient with a preset crack coefficient threshold to obtain a second abnormality recognition parameter;

[0067] S5. Extract features of the vibration signal corresponding to the target industrial belt to obtain a vibration feature set corresponding to the vibration signal, and perform vibration detection on the vibration feature set corresponding to the vibration signal based on a trained vibration classification model to obtain a third abnormality recognition parameter;

[0068] S6. Perform an abnormality alarm based on the first abnormality identification parameter, the second abnormality identification parameter, and the third abnormality identification parameter.

[0069] In the present invention, a high-speed pulse camera is a camera that can capture a large number of images in a very short time, and is often used for analyzing high-speed moving objects; mutation frame detection refers to analyzing frames with significant changes in the video to identify key moments or abnormal events; vibration signals refer to dynamic signals generated by the operation of the equipment, and are often used to monitor the state of the machine; abnormal alarm refers to triggering an alarm for prompting based on the identified abnormal parameters.

[0070] Embodiment 2, the method for identifying longitudinal tearing of industrial belts based on a high-speed pulse camera proposed by the present invention, compared with embodiment 1, this embodiment further includes: performing mutation frame detection on the target industrial belt operation video to obtain a mutation frame set corresponding to the target industrial belt operation video, including:

[0071] Acquire a current target industrial belt operation video frame from the target industrial belt operation video, and extract feature points of the current target industrial belt operation video frame to obtain a feature point set corresponding to the current target industrial belt operation video frame;

[0072] Combining the feature point set corresponding to the current target industrial belt operation video frame with the feature point set corresponding to the previous target industrial belt operation video frame to obtain a matching point set;

[0073] The rotation angle of each matching point in the matching point set is calculated to obtain the rotation angle set corresponding to the matching point set. The calculation formula of the rotation angle is as follows:

[0074]

[0075] Among them, θ i represents the rotation angle corresponding to the i-th matching point in the matching point set, (Q i,y ,P i,y ) represents the ordinate of the feature point corresponding to the previous target industrial belt operation video frame and the current target industrial belt operation video frame in the i-th matching point in the matching point set, (Q i,x ,P i,x ) represents the horizontal coordinate of the feature point corresponding to the previous target industrial belt operation video frame and the current target industrial belt operation video frame in the i-th matching point in the matching point set;

[0076] The rotation angle set corresponding to the matching point set is converted into a histogram to obtain a direction histogram corresponding to the matching point set, wherein the abscissa of the direction histogram represents the rotation angle, and the ordinate of the direction histogram represents the cumulative number of the rotation angle.

[0077] In this embodiment, the cumulative number of rotation angles refers to the cumulative sum of the number of corresponding feature points in the matching point set at a specific rotation angle. Specifically, it is a statistic that is used to show how many matching points have rotation angles within a certain angle range. When constructing a direction histogram, the rotation angles are usually grouped by angle intervals, and the number of matching points in each interval is calculated. The cumulative number is the sum of these numbers to more intuitively represent the distribution at different angles, which can help analyze the rotation trend and direction of the object during movement.

[0078] In an optional embodiment, performing mutation frame detection on the target industrial belt operation video to obtain a mutation frame set corresponding to the target industrial belt operation video further includes:

[0079] The angle probability histogram corresponding to the matching point set is obtained based on the direction histogram corresponding to the matching point set, where the horizontal axis of the angle probability histogram represents the rotation angle, and the vertical axis of the angle probability histogram represents the probability accumulation number of the rotation angle. The angle probability is calculated as follows:

[0080]

[0081] Among them, P(θ i ) represents the occurrence probability of the rotation angle corresponding to the i-th matching point in the matching point set, and N represents the total number of matching points in the matching point set;

[0082] Obtaining the rotation angle corresponding to the maximum probability cumulative number of the rotation angle based on the angle probability histogram corresponding to the matching point set;

[0083] The rotation angle corresponding to the maximum probability cumulative number of the rotation angle is compared with the preset rotation angle threshold. If it is greater than the preset rotation angle threshold, it indicates that the current target industrial belt operation video frame is a mutation frame. Otherwise, it indicates that the current target industrial belt operation video frame is not a mutation frame.

[0084] Repeat the above operation until all video frames in the target industrial belt operation video are traversed to obtain a set of mutation frames corresponding to the target industrial belt operation video.

[0085] In an optional embodiment, image preprocessing is performed on the target industrial belt side image to obtain the target industrial belt side standard image, including:

[0086] Binarizing the target industrial belt side image to obtain a target industrial belt side binary image;

[0087] Traverse each pixel point in the binary image of the side of the target industrial belt, and when traversing to the current pixel point, determine whether the current pixel point is an isolated point;

[0088] If the current pixel is an isolated point, the current pixel is deleted; otherwise, the next pixel of the current pixel is traversed until each pixel in the target industrial belt side binary image is traversed to obtain the target industrial belt side standard image.

[0089] It should be noted that binarization refers to the process of converting an image into only two colors (usually black and white). By setting a threshold, pixels above the threshold are set to white, and pixels below the threshold are set to black. Isolated points refer to those points in the binary image that have no adjacent connection with other pixels. Usually there are no similar pixels around them. Isolated points may be noise or unnecessary details.

[0090] In an optional embodiment, the crack segmentation model includes a feature extraction module, a segmentation accuracy improvement module and a crack prediction module. The feature extraction module adopts a U-Net network and is used to extract a feature map of the standard image of the side of the industrial belt. The segmentation accuracy improvement module adopts a dilation operation and a corrosion operation. The dilation operation is used to merge disconnected feature maps, and the corrosion operation is used to separate the connection areas between different feature maps. The crack prediction module is used to predict the feature map after accuracy improvement based on an upsampling operation to obtain a crack image corresponding to the target industrial belt side standard image.

[0091] It should be noted that the U-Net network is a convolutional neural network (CNN) architecture commonly used in medical image segmentation. It consists of two parts: an encoder and a decoder, and can effectively extract and restore image features.

[0092] In an optional embodiment, the formula for the expansion operation is as follows:

[0093]

[0094] Among them, I represents the input feature, W d represents the dilation filter, represents the dilation operation, (i, j, k) represents the size of the input feature, S1, S2, S3 represent the total size of the feature, and S1 = {1, 2, ..., a}, S2 = {1, 2, ..., b}, S3 = {1, 2, ..., c} represents, (a, b, c) represents the size of the dilation filter;

[0095] The formula for the corrosion operation is as follows:

[0096]

[0097] Where W represents the corrosion filter, Represents an erosion operation.

[0098] In an optional embodiment, characterization features are extracted from the crack image corresponding to the standard image of the target industrial belt side surface to obtain a characterization feature set corresponding to the crack image, including:

[0099] Acquire the pixel area of ​​the crack image, and use the pixel area of ​​the crack image as the first characterization feature;

[0100] Obtaining an outer frame of the crack image, and using the length and width of the outer frame of the crack image as a second characterization feature;

[0101] Counting the edge pixels of the crack image to obtain the total edge pixels of the crack image, and using the total edge pixels of the crack image as the third representation feature;

[0102] The first characterization feature, the second characterization feature, and the third characterization feature of the crack image are combined to obtain a characterization feature set corresponding to the crack image.

[0103] In an optional embodiment, feature extraction is performed on the vibration signal corresponding to the target industrial belt to obtain a vibration feature set corresponding to the vibration signal, including:

[0104] A sliding window of fixed length is predefined, and the sliding window is moved in the vibration signal to obtain a sub-vibration signal in the sliding window, wherein the vibration signal includes a plurality of sub-vibration signals;

[0105] Calculating the signal-to-noise ratio of the sub-vibration signal, and matching the signal-to-noise ratio corresponding to the sub-vibration signal with a preset signal-to-noise ratio level table to obtain the signal-to-noise ratio level corresponding to the sub-vibration signal;

[0106] Classifying the sub-vibration signals into different processing sets based on the signal-to-noise ratio levels corresponding to the sub-vibration signals, wherein the different processing sets include a primary processing set, a secondary processing set, and a tertiary processing set;

[0107] Performing feature extraction on the sub-vibration signals in the primary processing set based on Hilbert-Huang transform to obtain a first vibration feature set, performing feature extraction on the sub-vibration signals in the secondary processing set based on resonance sparse decomposition to obtain a second vibration feature set, and performing feature extraction on the sub-vibration signals in the secondary processing set based on symmetric point pattern diagram analysis to obtain a third vibration feature set;

[0108] The first vibration feature set, the second vibration feature set, and the third vibration feature set are combined to obtain a vibration feature set corresponding to the vibration signal.

[0109] It should be noted that the signal-to-noise ratio is used to quantify the ratio of signal strength to background noise strength, reflecting the clarity and availability of the signal; the Hilbert-Huang transform is an analysis method used to extract features from nonlinear and non-stationary signals; resonance sparse decomposition is a signal processing technique that extracts features by extracting sparse features from signals; and symmetry point pattern analysis is an analysis method that focuses on the patterns of symmetric points in the signal to help extract specific features.

[0110] In an optional embodiment, the anomaly detection model uses Faster R-CNN and the vibration classification model uses a support vector machine.

[0111] It should be noted that Faster R-CNN is an object detection algorithm that combines the region proposal network (RPN) and the convolutional neural network (CNN). It can efficiently identify and locate objects in images and is suitable for real-time detection. Support vector machine is a supervised learning model used for classification and regression analysis. It distinguishes data points of different categories by finding the best hyperplane and is suitable for processing high-dimensional data.

[0112] Embodiment three, as Figure 2 As shown, the industrial belt longitudinal tear recognition system based on high-speed pulse camera is applicable to the above-mentioned industrial belt longitudinal tear recognition method based on high-speed pulse camera, including:

[0113] The data acquisition unit 1 is used to collect video of the front of the target industrial belt based on a high-speed pulse camera to obtain the target industrial belt operation video, collect images of the side of the target industrial belt based on the high-speed pulse camera according to a preset sampling period to obtain the side image of the target industrial belt, and collect data of the target industrial belt based on the acceleration sensor to obtain the vibration signal corresponding to the target industrial belt;

[0114] The operation detection unit 2 is used to perform mutation frame detection on the target industrial belt operation video to obtain a mutation frame set corresponding to the target industrial belt operation video, and perform anomaly detection on the mutation frame set corresponding to the target industrial belt operation video based on the trained anomaly detection model to obtain a first anomaly recognition parameter;

[0115] The crack segmentation unit 3 is used to perform image preprocessing on the target industrial belt side image to obtain the target industrial belt side standard image, and perform crack segmentation on the target industrial belt side standard image based on the trained crack segmentation model to obtain a crack image corresponding to the target industrial belt side standard image;

[0116] The crack detection unit 4 is used to extract the characterization features of the crack image corresponding to the standard image of the side of the target industrial belt to obtain a characterization feature set corresponding to the crack image, averagely weight the characterization features in the characterization feature set to obtain a crack coefficient corresponding to the crack image, and compare the crack coefficient with a preset crack coefficient threshold to obtain a second abnormality recognition parameter;

[0117] A vibration detection unit 5, the vibration detection unit 5 is used to extract features of the vibration signal corresponding to the target industrial belt to obtain a vibration feature set corresponding to the vibration signal, and perform vibration detection on the vibration feature set corresponding to the vibration signal based on a trained vibration classification model to obtain a third abnormality recognition parameter;

[0118] The abnormality alarm unit 6 is used to make an abnormality alarm based on the first abnormality identification parameter, the second abnormality identification parameter and the third abnormality identification parameter.

[0119] The embodiments of the present invention are described in detail above with reference to the accompanying drawings, but the present invention is not limited thereto, and various changes can be made within the knowledge scope of technicians in the relevant technical field without departing from the purpose of the present invention.

Claims

1. The method for identifying longitudinal tearing of industrial belts based on a high-speed pulse camera is characterized in that: include: Based on the high-speed pulse camera, the front of the target industrial belt is video-captured to obtain the target industrial belt operation video; based on the high-speed pulse camera, the side of the target industrial belt is image-captured according to a preset sampling period to obtain the side image of the target industrial belt; based on the acceleration sensor, the data of the target industrial belt is collected to obtain the vibration signal corresponding to the target industrial belt; Performing mutation frame detection on the target industrial belt operation video to obtain a mutation frame set corresponding to the target industrial belt operation video, and performing anomaly detection on the mutation frame set corresponding to the target industrial belt operation video based on a trained anomaly detection model to obtain a first anomaly recognition parameter; Performing image preprocessing on the target industrial belt side image to obtain a target industrial belt side standard image, and performing crack segmentation on the target industrial belt side standard image based on a trained crack segmentation model to obtain a crack image corresponding to the target industrial belt side standard image; Extracting characterization features of the crack image corresponding to the standard image of the target industrial belt side surface to obtain a characterization feature set corresponding to the crack image, performing average weighting on the characterization features in the characterization feature set to obtain a crack coefficient corresponding to the crack image, and comparing the crack coefficient with a preset crack coefficient threshold to obtain a second abnormality recognition parameter; Performing feature extraction on the vibration signal corresponding to the target industrial belt to obtain a vibration feature set corresponding to the vibration signal, and performing vibration detection on the vibration feature set corresponding to the vibration signal based on a trained vibration classification model to obtain a third abnormality recognition parameter; An abnormality alarm is performed based on the first abnormality identification parameter, the second abnormality identification parameter and the third abnormality identification parameter.

2. The method for identifying longitudinal tearing of industrial belts based on a high-speed pulse camera according to claim 1, characterized in that: Performing mutation frame detection on the target industrial belt operation video to obtain a mutation frame set corresponding to the target industrial belt operation video includes: Acquire a current target industrial belt operation video frame from the target industrial belt operation video, and extract feature points of the current target industrial belt operation video frame to obtain a feature point set corresponding to the current target industrial belt operation video frame; Combining the feature point set corresponding to the current target industrial belt operation video frame with the feature point set corresponding to the previous target industrial belt operation video frame to obtain a matching point set; The rotation angle of each matching point in the matching point set is calculated to obtain a rotation angle set corresponding to the matching point set. The calculation formula of the rotation angle is as follows: Among them, θ i represents the rotation angle corresponding to the i-th matching point in the matching point set, (Q i,y ,P i,y ) represents the ordinate of the feature point corresponding to the previous target industrial belt operation video frame and the current target industrial belt operation video frame in the i-th matching point in the matching point set, (Q i,x ,P i,x ) represents the horizontal coordinate of the feature point corresponding to the previous target industrial belt operation video frame and the current target industrial belt operation video frame in the i-th matching point in the matching point set; The rotation angle set corresponding to the matching point set is subjected to histogram conversion to obtain a direction histogram corresponding to the matching point set, wherein the abscissa of the direction histogram represents the rotation angle, and the ordinate of the direction histogram represents the cumulative number of the rotation angle.

3. The method for identifying longitudinal tearing of industrial belts based on a high-speed pulse camera according to claim 2, characterized in that: Performing mutation frame detection on the target industrial belt operation video to obtain a mutation frame set corresponding to the target industrial belt operation video also includes: The angle probability histogram corresponding to the matching point set is obtained based on the direction histogram corresponding to the matching point set, wherein the abscissa of the angle probability histogram represents the rotation angle, the ordinate of the angle probability histogram represents the probability accumulation number of the rotation angle, and the angle probability is calculated as follows: Among them, P(θ i ) represents the occurrence probability of the rotation angle corresponding to the i-th matching point in the matching point set, and N represents the total number of matching points in the matching point set; Obtaining the rotation angle corresponding to the maximum probability cumulative number of the rotation angle based on the angle probability histogram corresponding to the matching point set; The rotation angle corresponding to the maximum probability cumulative number of the rotation angle is compared with a preset rotation angle threshold. If it is greater than the preset rotation angle threshold, it indicates that the current target industrial belt operation video frame is a mutation frame. Otherwise, it indicates that the current target industrial belt operation video frame is not a mutation frame. The above operation is repeated until all video frames in the target industrial belt operation video are traversed to obtain a set of mutation frames corresponding to the target industrial belt operation video.

4. The method for identifying longitudinal tearing of industrial belts based on a high-speed pulse camera according to claim 1, characterized in that: Performing image preprocessing on the target industrial belt side image to obtain a target industrial belt side standard image, including: Binarizing the target industrial belt side image to obtain a target industrial belt side binary image; Traversing each pixel point in the binary image of the side of the target industrial belt, and when traversing to the current pixel point, determining whether the current pixel point is an isolated point; If the current pixel point is an isolated point, the current pixel point is deleted; otherwise, the next pixel point of the current pixel point is traversed until each pixel point in the target industrial belt side binary image is traversed to obtain the target industrial belt side standard image.

5. The method for identifying longitudinal tearing of industrial belts based on a high-speed pulse camera according to claim 1, characterized in that: The crack segmentation model includes a feature extraction module, a segmentation accuracy improvement module and a crack prediction module. The feature extraction module adopts a U-Net network and is used to extract the feature map of the standard image of the industrial belt side. The segmentation accuracy improvement module adopts an expansion operation and a corrosion operation. The expansion operation is used to merge the disconnected feature maps, and the corrosion operation is used to separate the connection areas between different feature maps. The crack prediction module is used to predict the feature map after accuracy improvement based on an upsampling operation to obtain a crack image corresponding to the target industrial belt side standard image.

6. The method for identifying longitudinal tearing of industrial belts based on a high-speed pulse camera according to claim 5, characterized in that: The formula for the dilation operation is as follows: Among them, I represents the input feature, W d represents the dilation filter, represents the dilation operation, (i, j, k) represents the size of the input feature, S1, S2, S3 represent the total size of the feature, and S1 = {1, 2, ..., a}, S2 = {1, 2, ..., b}, S3 = {1, 2, ..., c} represents, (a, b, c) represents the size of the dilation filter; The formula for the corrosion operation is as follows: Where W represents the corrosion filter, Represents an erosion operation.

7. The method for identifying longitudinal tearing of industrial belts based on a high-speed pulse camera according to claim 1, characterized in that: Extracting the characterization features of the crack image corresponding to the standard image of the target industrial belt side surface to obtain a characterization feature set corresponding to the crack image, including: Acquire the pixel area of ​​the crack image, and use the pixel area of ​​the crack image as a first characterization feature; Acquire an outer frame of the crack image, and use the length and width of the outer frame of the crack image as a second characterization feature; Counting edge pixels of the crack image to obtain total edge pixels of the crack image, and using the total edge pixels of the crack image as a third representation feature; The first characterizing feature, the second characterizing feature and the third characterizing feature of the crack image are combined to obtain a characterizing feature set corresponding to the crack image.

8. The method for identifying longitudinal tearing of industrial belts based on a high-speed pulse camera according to claim 1, characterized in that: Feature extraction is performed on the vibration signal corresponding to the target industrial belt to obtain a vibration feature set corresponding to the vibration signal, including: Predefine a sliding window of fixed length, and move the sliding window in the vibration signal to obtain a sub-vibration signal in the sliding window, wherein the vibration signal includes a plurality of sub-vibration signals; Calculating the signal-to-noise ratio of the sub-vibration signal, and matching the signal-to-noise ratio corresponding to the sub-vibration signal with a preset signal-to-noise ratio level table to obtain the signal-to-noise ratio level corresponding to the sub-vibration signal; Classifying the sub-vibration signals into different processing sets based on the signal-to-noise ratio levels corresponding to the sub-vibration signals, wherein the different processing sets include a primary processing set, a secondary processing set, and a tertiary processing set; Performing feature extraction on the sub-vibration signals in the primary processing set based on Hilbert-Huang transform to obtain a first vibration feature set, performing feature extraction on the sub-vibration signals in the secondary processing set based on resonance sparse decomposition to obtain a second vibration feature set, and performing feature extraction on the sub-vibration signals in the secondary processing set based on symmetric point pattern diagram analysis to obtain a third vibration feature set; The first vibration feature set, the second vibration feature set, and the third vibration feature set are combined to obtain a vibration feature set corresponding to the vibration signal.

9. The method for identifying longitudinal tearing of industrial belts based on a high-speed pulse camera according to claim 1, characterized in that: The anomaly detection model adopts Faster R-CNN, and the vibration classification model adopts support vector machine.

10. An industrial belt longitudinal tear identification system based on a high-speed pulse camera, which is applicable to an industrial belt longitudinal tear identification method based on a high-speed pulse camera as claimed in any one of claims 1 to 9, characterized in that: include: A data acquisition unit (1), the data acquisition unit (1) is used to perform video acquisition on the front of a target industrial belt based on a high-speed pulse camera to obtain an operation video of the target industrial belt, perform image acquisition on the side of the target industrial belt based on the high-speed pulse camera according to a preset sampling period to obtain a side image of the target industrial belt, and perform data acquisition on the target industrial belt based on an acceleration sensor to obtain a vibration signal corresponding to the target industrial belt; An operation detection unit (2), the operation detection unit (2) is used to perform mutation frame detection on the target industrial belt operation video to obtain a mutation frame set corresponding to the target industrial belt operation video, and perform anomaly detection on the mutation frame set corresponding to the target industrial belt operation video based on a trained anomaly detection model to obtain a first anomaly recognition parameter; A crack segmentation unit (3), the crack segmentation unit (3) is used to perform image preprocessing on the target industrial belt side image to obtain a target industrial belt side standard image, and perform crack segmentation on the target industrial belt side standard image based on a trained crack segmentation model to obtain a crack image corresponding to the target industrial belt side standard image; A crack detection unit (4), the crack detection unit (4) being used to extract characterization features from a crack image corresponding to the standard image of the target industrial belt side surface to obtain a characterization feature set corresponding to the crack image, averagely weight the characterization features in the characterization feature set to obtain a crack coefficient corresponding to the crack image, and compare the crack coefficient with a preset crack coefficient threshold to obtain a second abnormality recognition parameter; A vibration detection unit (5), the vibration detection unit (5) being used to perform feature extraction on the vibration signal corresponding to the target industrial belt to obtain a vibration feature set corresponding to the vibration signal, and to perform vibration detection on the vibration feature set corresponding to the vibration signal based on a trained vibration classification model to obtain a third abnormality recognition parameter; An abnormality alarm unit (6), the abnormality alarm unit (6) is used to perform an abnormality alarm based on the first abnormality identification parameter, the second abnormality identification parameter and the third abnormality identification parameter.

Images