Belt crack detection method based on image recognition

Abstract

The present invention relates to a belt crack detection method based on image recognition, and the method comprises: step 1: collecting a belt surface image of the belt surface with a first camera; step 2: transmitting the belt surface image obtained in step 1 to a background system for recognition operation, and outputting a tear recognition judgment; if the tear recognition judgment is true, jumping to step 3; step 3: collecting a belt surface image of the belt surface with a second camera; step 4: transmitting the belt surface image obtained in step 3 to a background system for recognition operation, and outputting a tear recognition judgment; if the tear recognition judgment is true again, outputting an alarm. The present invention can ensure monitoring accuracy, reduce false alarms, and realize rapid positioning of the damaged position of the belt during belt operation monitoring.

Description

Technical Field

[0001] The invention relates to a belt crack detection method based on image recognition. Background Art

[0002] Transmission belts are widely used in coal mining, metallurgy and other industrial fields. Since the transported materials are often mixed with various sharp impurities, the transmission belts will have horizontal / vertical tear gaps after long-term use, which seriously reduces the service life of the carrying belts. In the prior art, the tear detection of transmission belts based on image recognition cannot guarantee accuracy when monitoring the belt operation, which is prone to false alarms and affects production. At the same time, it is also impossible to quickly locate the torn part. Therefore, how to develop a more advanced belt crack detection method to overcome the above problems is a direction that technicians in this field need to study. Summary of the invention

[0003] The present invention aims to provide a belt crack detection method based on image recognition to solve the above-mentioned problems existing in the prior art.

[0004] The technical solutions adopted are as follows:

[0005] A belt crack detection method based on image recognition, comprising:

[0006] Step 1: using a first camera to capture a belt surface image;

[0007] Step 2: The surface image obtained in step 1 is transmitted to the background system for recognition operation, and the tear recognition judgment is output; if the tear recognition judgment is true, jump to step 3;

[0008] Step 3: Use the second camera to capture the belt surface image;

[0009] Step 4: The surface image obtained in step 3 is transmitted to the background system for recognition operation, and the tear recognition judgment is output; if the tear recognition judgment is true again, an alarm is output.

[0010] By adopting this technical solution: when the first camera identifies that the current belt is torn, the second camera takes images from different angles for identification detection to verify the first identification, thereby improving the identification accuracy and reducing the occurrence of false alarms. At the same time, based on the relative position relationship and time difference when the first camera and the second camera output detection signals, the specific position of the torn part on the belt is calculated, which helps the staff to quickly locate the damaged part of the belt.

[0011] Preferably, in the above-mentioned belt crack detection method based on image recognition, the recognition operation includes:

[0012] Step A1: Convert the RGB format image into a grayscale image;

[0013] Step A2: performing noise reduction processing on the grayscale image based on a median filtering algorithm;

[0014] Step A3: extracting tearing feature parameters from the grayscale image;

[0015] Step A4: importing the tear feature parameters into the tear recognition model to obtain the tear recognition judgment.

[0016] By adopting this technical solution: the belt surface image is uniformly converted into a grayscale image, resulting in the grayscale value of the flat part of the belt surface being lower than the grayscale value of the cracks on the belt surface, so that the cracks as tear feature vectors can be quickly extracted from the original image, and then put into the model trained by the neural network for identification, which helps to make tear identification intelligent and automated.

[0017] Preferably, in the above-mentioned belt crack detection method based on image recognition, step A3 comprises:

[0018] Step B1: segmenting the grayscale image based on a preset segmentation value, and extracting a linear crack image whose grayscale value is greater than the preset segmentation value from the grayscale image;

[0019] Step B2: respectively obtaining the crack length L and the average crack width A of the linear crack;

[0020] Step B3: Calculate the total area S=L*A of the linear crack image.

[0021] More preferably, in the above-mentioned belt crack detection method based on image recognition: the tear recognition model is trained by importing a pre-stored image containing a torn belt into a ResNet neural network model.

[0022] More preferably, the belt crack detection method based on image recognition further includes steps 5-6;

[0023] Step 5: Stop the belt transmission for maintenance, and take a belt surface image corresponding to the belt surface image described in step 3 for the static belt;

[0024] Step 6: Input the surface image obtained in step 5 into the ResNet neural network model to train and improve the tear recognition model.

[0025] More preferably, in the above-mentioned belt crack detection method based on image recognition: the first camera is a fixed camera, and the second camera is a movable camera.

[0026] More preferably, in the above-mentioned belt crack detection method based on image recognition: the second camera is installed on an inspection robot; the moving trajectory of the inspection robot is parallel to the extension direction of the belt and is located directly above the belt.

[0027] Compared with the prior art, the present invention can quickly locate the torn part when the accuracy cannot be guaranteed during belt operation monitoring, avoids false alarms. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 The following is a flowchart of an embodiment of the present invention. DETAILED DESCRIPTION

[0029] In the description of the present invention, it should be understood that the terms "first" and "second" are only used for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In the description of the present invention, the meaning of "multiple" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined. The terms "installed", "connected", "connected", "fixed" and the like should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral body; it can be a mechanical connection or an electrical connection; it can be directly connected or indirectly connected through an intermediate medium, and the first feature "above" or "below" the second feature can mean that the first and second features are in direct contact, or the first and second features are in indirect contact through an intermediate medium. Moreover, the first feature "above", "above" and "above" the second feature can mean that the first feature is directly above or obliquely above the second feature, or simply means that the first feature is higher in level than the second feature. The first feature "below", "below" and "below" the second feature can mean that the first feature is directly below or obliquely below the second feature, or simply means that the first feature is lower in level than the second feature.

[0030] The following will be combined with the attached Figure 1 , the embodiments of the present invention are further described.

[0031] Example 1

[0032] A belt crack detection method based on image recognition, comprising:

[0033] Step 1: operate a first camera to collect belt surface images at a frequency of one image per second, wherein the first camera is a fixed camera and is vertically fixed above the middle section of the circulating transmission belt to be inspected;

[0034] Step 2: The surface image obtained by the first camera is transmitted to the background system for recognition operation, and the tear recognition judgment is output; if the tear recognition judgment is true, jump to step 3;

[0035] Step 3: Use a second camera to collect a belt surface image, wherein the second camera is installed on an inspection robot; the inspection robot is installed on a slide rail and reciprocates along the extension direction of the slide rail; the slide rail is parallel to the top of the circulating transmission belt.

[0036] Step 4: The surface image obtained by the second camera is transmitted to the background system for recognition operation, and the tear recognition judgment is output; if the tear recognition judgment is true again, an alarm is output;

[0037] Step 5: Stop the belt transmission for maintenance, and take a belt surface image corresponding to the belt surface image described in step 3 for the static belt;

[0038] Step 6: Input the surface image obtained in step 5 into the ResNet neural network model to train and improve the tear recognition model.

[0039] In the above steps, the identification operations in step 2 and step 4 include:

[0040] Step A1: Convert the RGB format image into a grayscale image; the image grayscale conversion formula is:

[0041] h(i,j)=0.1250R(i,j)+0.7154G(i,j)+0.0721B(i,j);

[0042] h(i,j) is the gray value of the i-th row and j-th column in the gray image;

[0043] R(i,j), G(i,j), B(i,j) are the component values ​​of the three basic colors of red, green and blue in the original image respectively;

[0044] Step A2: performing noise reduction processing on the grayscale image based on a median filtering algorithm;

[0045] Step A3: extracting tearing feature parameters from the grayscale image;

[0046] Step A4: importing the tear feature parameters into the tear recognition model to obtain the tear recognition judgment.

[0047] Wherein, the step A3 comprises:

[0048] Step B1: segmenting the grayscale image based on a preset segmentation value, and extracting a linear crack image whose grayscale value is greater than the preset segmentation value from the grayscale image;

[0049] Step B2: respectively obtaining the crack length L and the average crack width A of the linear crack;

[0050] Among them, the skeleton of the linear crack is firstly extracted. After the skeleton is extracted, the linear crack is thinned into a single pixel width, and then the number of white pixels is counted, and the number of pixels is calculated to obtain the length L of the linear crack;

[0051] Step B3: Calculate the total area S=L*A of the linear crack image;

[0052] The tear recognition model is obtained by training by importing pre-stored images containing torn belts into a ResNet neural network model.

[0053] In the above process, since the relative moving speed of the first camera when shooting the belt is completely different from the relative moving speed of the second camera when chasing / facing the belt, it is equivalent to performing secondary detection based on the same operation model at the same position image shot at two different relative speeds. This minimizes the interference of external factors, indirectly ensures the accuracy of image recognition during the belt transmission process, and reduces the probability of false alarms.

[0054] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A belt crack detection method based on image recognition, characterized in that: include: Step 1: using a first camera to capture a belt surface image; Step 2: The surface image obtained in step 1 is transmitted to the background system for recognition operation, and the tear recognition judgment is output; if the tear recognition judgment is true, jump to step 3; Step 3: Use the second camera to capture the belt surface image; Step 4: The surface image obtained in step 3 is transmitted to the background system for recognition operation, and the tear recognition judgment is output; if the tear recognition judgment is true again, an alarm is output; Also includes steps 5-6; Step 5: Stop the belt transmission for maintenance, and take a belt surface image corresponding to the belt surface image described in step 3 for the static belt; Step 6: Input the strip surface image obtained in step 5 into the ResNet neural network model to train and improve the tear recognition model; The first camera is a fixed camera, and the second camera is a movable camera; The second camera is installed on the inspection robot; the movement track of the inspection robot is parallel to the extension direction of the belt and is located directly above the belt; When the first camera identifies that the current belt is torn, the second camera takes images from different angles for identification detection to verify the first identification, thereby improving the identification accuracy and reducing the occurrence of false alarms; at the same time, based on the relative position relationship and time difference when the first camera and the second camera output detection signals, the specific position of the torn part on the belt is calculated; The identification operation includes: Step A1: Convert the RGB format image into a grayscale image; Step A2: performing noise reduction processing on the grayscale image based on a median filtering algorithm; Step A3: extracting tearing feature parameters from the grayscale image; Step A4: importing the tear feature parameters into the tear recognition model to obtain a tear recognition judgment; The step A3 comprises: Step B1: segmenting the grayscale image based on a preset segmentation value, and extracting a linear crack image whose grayscale value is greater than the preset segmentation value from the grayscale image; Step B2: respectively obtaining the crack length L and the average crack width A of the linear crack; By extracting the skeleton of the linear crack, the linear crack is thinned into a single pixel width, and then the number of white pixels is counted to calculate the length L of the linear crack from the number of pixels. Step B3: Calculate the total area S=L*A of the linear crack image; The tear recognition model is obtained by training by importing a pre-stored image containing a torn belt into a ResNet neural network model; The relative movement speed of the two when the first camera shoots the belt is completely different from the relative movement speed of the two when the second camera chases / faces the belt to shoot, which is equivalent to performing secondary detection on the same position picture shot at two different relative speeds based on the same calculation model.

Images