A method for detecting bending of bar steel

Abstract

The present invention discloses a method for detecting the bending of bar steel materials. The method uses a high-temperature resistant camera to shoot the bars on a conveyor belt to be sent into a heating furnace in real time, and then uses a segmentation network based on deep learning to segment the image of the steel materials in real time. Then, the segmented steel material image is subjected to an edge detection algorithm to detect the boundary features of the steel materials, and three side edge features are extracted therefrom. It is judged whether the steel materials are bent according to the side edge features, and the bent steel materials are prevented from being sent into the heating furnace, thereby realizing the automatic detection of the bending of bar steel materials and avoiding the problem of steel jamming caused by the bent steel materials entering the heating furnace.

Description

Technical Field

[0001] The invention belongs to the technical field of computer vision and image detection, and in particular relates to a method and device for detecting the bending of a bar steel material. Background Art

[0002] Rolled bar steel will bend and deform during processing or transportation, which will damage the subsequent conveying equipment. When the bent steel is sent to the heating furnace, it may get stuck. Due to the extremely high temperature of the heating furnace, it is difficult to remove the stuck steel, which will seriously waste time and manpower and material resources, and even cause equipment failure and damage. Therefore, before conveying the steel to the heating furnace, the bending of the steel needs to be measured. When it is detected that the bending of the steel exceeds a certain degree, the conveyor belt is stopped from continuing to convey, and an alarm is sent to the front end to allow the monitoring personnel or the robot arm to remove the bent steel. Summary of the invention

[0003] The technical problem to be solved by the present invention is to provide a bar steel material bending detection method in view of the deficiencies of the above-mentioned prior art.

[0004] In order to achieve the above technical objectives, the technical solution adopted by the present invention is:

[0005] A method for detecting the bending of a bar steel material, comprising the following steps:

[0006] Step S1: a high temperature resistant network camera is fixed obliquely above the bar steel material to collect images of the steel material on the conveyor belt in real time;

[0007] Step S2: input the acquired image into a segmentation network based on deep learning to segment the image of the rod in real time;

[0008] Step S3: using an edge detection algorithm to perform edge detection on the segmented steel material image to extract the steel material boundary features and three side edge features;

[0009] Step S4: Detecting whether the steel material is bent based on the three side edge features;

[0010] Step S5: When the bending of the steel material is detected, an alarm is sent to the front end to prevent the bent steel material from being fed into the heating furnace.

[0011] To optimize the above technical solutions, the specific measures taken also include:

[0012] Furthermore, in step S1, the high temperature resistant network camera captures two side surfaces and three side edges of the steel material.

[0013] Furthermore, step S2 is specifically as follows:

[0014] S21: input the image data into a segmentation network based on deep learning to determine whether each pixel belongs to steel material;

[0015] S22: selecting all pixels in the image that are judged to belong to steel materials to form a steel material image;

[0016] S23: Calculate the distance between the two farthest pixels in the steel material image. When the distance is less than a certain threshold, it is determined that the steel material has not completely entered the shooting field of view, and the process goes to S21 to detect the next image.

[0017] Furthermore, the deep learning-based segmentation network in step S21 includes the following steps:

[0018] S211: Use multi-layer perceptron to extract the features of each pixel of the image;

[0019] S212: For each pixel, a 3×3 neighborhood, a 5×5 neighborhood, and a 7×7 neighborhood are maximum pooled to obtain local features of three scales for each pixel, and the three features are linked to form a multi-scale local feature of each pixel;

[0020] S213: extract high-dimensional features from the multi-scale local features of each pixel using a multi-layer perceptron, and then perform maximum pooling of the high-dimensional features of all pixels to obtain global features;

[0021] S214: extracting weights from the high-dimensional features of each pixel, linking the multi-scale local features and global features of each pixel and multiplying the weights to obtain weighted features;

[0022] S215: The weighted features of each pixel are reduced to one dimension through a fully connected layer, the segmentation probability of each pixel is calculated using a sigmoid function, and a cross entropy loss function is calculated for training to determine whether each pixel belongs to a steel pixel.

[0023] Furthermore, step S3 is specifically as follows:

[0024] S31: extracting boundary pixels from the segmented steel material image using an edge detection algorithm, constructing a minimum spanning tree using a Kruskal algorithm, and using boundary pixels as tree nodes;

[0025] S32: Calculate the pixel unit normal vector according to the nearest k neighbors of each boundary pixel;

[0026] S33: Select a pixel from the boundary pixels, and when the absolute value of the inner product of the unit normal vector of any pixel in the k-neighborhood of the pixel and the unit normal vector of the pixel is less than a set threshold, take the pixel as a breakpoint, and traverse all boundary pixels to obtain all breakpoints;

[0027] S34: All breakpoints divide the minimum spanning tree into multiple subtrees, and select the three subtrees with the most nodes as the three side edge features.

[0028] Further, step S4 is specifically as follows:

[0029] Step S41: Use the RANSAC algorithm to perform straight line detection on the three side edge features respectively;

[0030] Step S42: Calculate the distances from all pixels in the three side edge features to the corresponding straight lines. When there is a pixel whose distance is greater than a set threshold, determine that the boundary line corresponding to the pixel is curved.

[0031] Step S43: When any one of the three side edge features is bent, the steel material is bent.

[0032] Furthermore, step S5 is specifically as follows: when the bending of the steel material is detected, an alarm is sent to the front-end controller to interrupt the transmission of the steel material conveyor belt; after removing the bent steel material, the conveyor belt continues to transport the steel material and the next steel material is detected.

[0033] Beneficial effects of the present invention:

[0034] The present invention proposes a method and device for detecting the bending of bar steel based on deep learning image segmentation algorithm and image edge detection technology. The method uses a high temperature resistant camera to collect images of conveyed steel, uses a deep learning image segmentation algorithm to segment the steel, and uses an edge detection algorithm to extract the boundary features of the steel to detect the bending degree of the steel, thereby solving the problem of detecting the bending of steel before feeding it into a heating furnace and avoiding subsequent accidents caused by feeding bent steel into a heating furnace. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 A diagram of a bar steel bending detection method and device according to the present invention;

[0036] Figure 2 This is a schematic diagram of the shooting position of a high temperature resistant camera in an embodiment of the present invention;

[0037] Figure 3 This is a schematic diagram of a segmentation network structure based on deep learning in an embodiment of the present invention;

[0038] Figure 4 Schematic diagram of the boundary features and three side edge features of steel material in an embodiment of the present invention. DETAILED DESCRIPTION

[0039] The embodiments of the present invention are further described in detail below in conjunction with the accompanying drawings.

[0040] like Figure 1 As shown, the present invention is a bar steel bending detection method, comprising the following steps:

[0041] Step S1: a high temperature resistant network camera is fixed obliquely above the bar steel material to collect images of the steel material on the conveyor belt in real time;

[0042] Since the detection environment is located next to the heating furnace and the ambient temperature is relatively high, in order not to affect the image acquisition effect, the DS-NXCN3A204 CMOS starlight-level high-temperature resistant air-cooled cylindrical network camera is used to collect images. When the steel is bent, at least one surface (two side edges) will be bent, so only three of the four side edges need to be detected to determine whether the steel is bent. The camera is fixed obliquely above the conveyor belt of the bar steel material so that the two side surfaces (three side edges) of the steel material can be completely located within the camera's field of view. After the image is collected, due to the limited wireless transmission speed and the high ambient temperature, a dedicated high-temperature resistant cable is required to transmit the image data. Figure 2 This is a schematic diagram of the shooting position of the high-temperature resistant camera.

[0043] Step S2: input the acquired image into a segmentation network based on deep learning to segment the image of the rod in real time;

[0044] The image data is input into the segmentation network based on deep learning to determine whether each pixel belongs to steel. Then all pixels that are judged to belong to steel are selected to form a steel image. The distance between the two farthest pixels in the steel image is calculated. When the distance is less than a certain threshold, it is determined that the steel has not completely entered the shooting field of view, and the steel image is judged to be incomplete, and the next image will be detected. Figure 3 Schematic diagram of the segmentation network structure based on deep learning;

[0045] The deep learning-based segmentation network includes the following steps:

[0046] Step (1) extracting the features of each pixel from an image using a multi-layer perceptron;

[0047] Step (2) For each pixel, the 3×3 neighborhood, 5×5 area, and 7×7 area are max-pooled to obtain the local features of each pixel at three scales, and the three features are linked to form the multi-scale local features of each pixel;

[0048] Step (3) extracting high-dimensional features from the multi-scale local features of each pixel using a multi-layer perceptron, and then maximally pooling the high-dimensional features of all pixels to obtain global features;

[0049] Step (4) extracts weights from the high-dimensional features of each pixel, concatenates the multi-scale local features and global features of each pixel, and multiplies the weights to obtain weighted features;

[0050] Step (5) reduces the weighted features of each pixel to one dimension through a fully connected layer, uses the sigmoid function to obtain the segmentation probability of each pixel, calculates the cross entropy loss function for training, and determines whether each pixel belongs to a steel pixel.

[0051] Step S3: using an edge detection algorithm to perform edge detection on the segmented steel material image to extract the steel material boundary features and three side edge features;

[0052] The edge detection algorithm is used to extract boundary pixels from the segmented steel image, and the Kruskal algorithm is used to construct a minimum spanning tree with boundary pixels as tree nodes; the unit normal vector of each boundary pixel is calculated according to the nearest k neighbors of the pixel; a pixel among the boundary pixels is selected, and when the absolute value of the inner product of the unit normal vector of any pixel in the k neighbors of the pixel and the unit normal vector of the pixel is less than the set threshold, the pixel is taken as a breakpoint, and all boundary pixels are traversed to obtain all breakpoints; all breakpoints divide the minimum spanning tree into multiple subtrees, and the three subtrees with the most nodes are selected as three side edge features.

[0053] Step S4: Detecting whether the steel material is bent based on the three side edge features;

[0054] The RANSAC algorithm is used to perform straight line detection on the three side edge features respectively; the distances from all pixel points in the three side edge features to the corresponding straight lines are calculated, and when there is a pixel whose distance is greater than the set threshold, it is determined that the boundary line corresponding to the pixel is bent; when any of the three side edge features is bent, the steel material is bent.

[0055] Step S5: When the bending of the steel material is detected, an alarm is sent to the front-end controller to interrupt the transmission of the steel material conveyor belt; after removing the bent steel material, the conveyor belt continues to transport the steel material and the next steel material is tested.

[0056] The above are only preferred embodiments of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technicians in this technical field, some improvements and modifications without departing from the principle of the present invention should be regarded as the protection scope of the present invention.

Claims

1. A method for detecting bending of a bar steel material, characterized in that: The steps include: Step S1: a high temperature resistant network camera is fixed obliquely above the bar steel material to collect images of the steel material on the conveyor belt in real time; Step S2: input the acquired image into a segmentation network based on deep learning to segment the image of the rod in real time; Step S3: using an edge detection algorithm to perform edge detection on the segmented steel material image to extract the steel material boundary features and three side edge features; Step S4: Detecting whether the steel material is bent based on the three side edge features; Step S5: When the steel material is detected to be bent, an alarm is sent to the front end to prevent the bent steel material from being fed into the heating furnace; The step S3 is specifically as follows: Step S31: extracting boundary pixels from the segmented steel material image using an edge detection algorithm, and constructing a minimum spanning tree using a Kruskal algorithm, with the boundary pixels serving as tree nodes; Step S32: Calculate the pixel unit normal vector according to the nearest k neighbors of each boundary pixel; Step S33: select a pixel from the boundary pixels, and when the absolute value of the inner product of the unit normal vector of any pixel in the k-neighborhood of the pixel and the unit normal vector of the pixel is less than a set threshold, take the pixel as a breakpoint, and traverse all boundary pixels to obtain all breakpoints; Step S34: All breakpoints divide the minimum spanning tree into multiple subtrees, and select three subtrees with the most nodes as three side edge features; The step S4 is specifically as follows: Step S41: Use the RANSAC algorithm to perform straight line detection on the three side edge features respectively; Step S42: Calculate the distances from all pixels in the three side edge features to the corresponding straight lines. When there is a pixel whose distance is greater than a set threshold, determine that the boundary line corresponding to the pixel is curved. Step S43: When any one of the three side edge features is bent, the steel material is bent.

2. A bar steel material bending detection method according to claim 1, characterized in that: In step S1, the high temperature resistant network camera captures two side surfaces and three side edges of the steel material.

3. A bar steel material bending detection method according to claim 2, characterized in that: The step S2 is specifically as follows: Step S21: input the image data into a segmentation network based on deep learning to determine whether each pixel belongs to steel material; Step S22: selecting all pixels in the image that are judged to belong to steel materials to form a steel material image; Step S23: Calculate the distance between the two farthest pixels in the steel material image. When the distance is less than a certain threshold, it is determined that the steel material has not completely entered the shooting field of view, and go to S21 to detect the next image.

4. A bar steel material bending detection method according to claim 3, characterized in that: The deep learning-based segmentation network in step S21 comprises the following steps: Step S211: extracting features of each pixel of the image using a multi-layer perceptron; Step S212: For each pixel, the 3×3 neighborhood, 5×5 neighborhood, and 7×7 neighborhood are max-pooled to obtain local features of three scales for each pixel, and the three features are linked to form a multi-scale local feature of each pixel; Step S213: extract high-dimensional features from the multi-scale local features of each pixel using a multi-layer perceptron, and then perform maximum pooling of the high-dimensional features of all pixels to obtain global features; Step S214: extracting weights from the high-dimensional features of each pixel, linking the multi-scale local features and global features of each pixel and multiplying the weights to obtain weighted features; Step S215: The weighted features of each pixel are reduced to one dimension through a fully connected layer, the segmentation probability of each pixel is calculated using a sigmoid function, and a cross entropy loss function is calculated for training to determine whether each pixel belongs to a steel pixel.

5. A bar steel material bending detection method according to claim 4, characterized in that: The step S5 specifically includes: when the bending of the steel material is detected, an alarm is sent to the front-end controller to interrupt the transmission of the steel material conveyor belt; after the bent steel material is removed, the conveyor belt continues to transport the steel material and the next steel material is detected.

Images