CN118072148A - Precise ball screw pair detection system and method thereof - Google Patents
Precise ball screw pair detection system and method thereof Download PDFInfo
- Publication number
- CN118072148A CN118072148A CN202410503726.0A CN202410503726A CN118072148A CN 118072148 A CN118072148 A CN 118072148A CN 202410503726 A CN202410503726 A CN 202410503726A CN 118072148 A CN118072148 A CN 118072148A
- Authority
- CN
- China
- Prior art keywords
- image
- target
- ball screw
- module
- yolov
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 80
- 238000000034 method Methods 0.000 title claims abstract description 46
- 238000012937 correction Methods 0.000 claims abstract description 63
- 230000002159 abnormal effect Effects 0.000 claims abstract description 58
- 230000005856 abnormality Effects 0.000 claims abstract description 40
- 238000012549 training Methods 0.000 claims abstract description 31
- 230000004927 fusion Effects 0.000 claims abstract description 17
- 238000003702 image correction Methods 0.000 claims abstract description 11
- 238000003384 imaging method Methods 0.000 claims description 22
- 239000013598 vector Substances 0.000 claims description 19
- 238000012545 processing Methods 0.000 claims description 16
- 230000000007 visual effect Effects 0.000 claims description 11
- 238000013526 transfer learning Methods 0.000 claims description 5
- 238000011176 pooling Methods 0.000 claims description 3
- 238000009825 accumulation Methods 0.000 claims description 2
- 238000006243 chemical reaction Methods 0.000 claims description 2
- 238000002372 labelling Methods 0.000 claims description 2
- 230000008569 process Effects 0.000 abstract description 19
- 230000015556 catabolic process Effects 0.000 abstract description 9
- 238000006731 degradation reaction Methods 0.000 abstract description 9
- 238000003745 diagnosis Methods 0.000 abstract description 9
- 230000005540 biological transmission Effects 0.000 abstract description 8
- 238000000605 extraction Methods 0.000 abstract description 8
- 230000003993 interaction Effects 0.000 abstract description 8
- 238000004891 communication Methods 0.000 description 5
- 238000012935 Averaging Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 4
- 230000000694 effects Effects 0.000 description 4
- 238000004590 computer program Methods 0.000 description 3
- 238000007499 fusion processing Methods 0.000 description 3
- 238000004422 calculation algorithm Methods 0.000 description 2
- 238000013527 convolutional neural network Methods 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000008094 contradictory effect Effects 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 238000013508 migration Methods 0.000 description 1
- 230000005012 migration Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000000877 morphologic effect Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000002265 prevention Effects 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
Landscapes
- Image Processing (AREA)
Abstract
The invention relates to the technical field of ball screw pair detection, and provides a precision ball screw pair detection system and a precision ball screw pair detection method, wherein the precision ball screw pair detection system comprises a precision ball screw pair detection middle table, a binocular shooting module, an image correction module, a region determination module, an image fusion module and an abnormality detection module; the anomaly detection module is a target YOLOv model; the target YOLOv model comprises a Swin transducer module and an asymptotic feature pyramid network, and is obtained by training and optimizing based on a nut abnormal data set and a ball screw abnormal data set. According to the method, the Swin transducer module and the target YOLOv model of the asymptotic characteristic pyramid network are combined to perform anomaly detection on the image after distortion correction, so that the global information extraction capability is effectively improved, the loss or degradation of characteristic information in the transmission and interaction processes is avoided, and the accuracy and the robustness of anomaly diagnosis of the precise ball screw pair to be detected are further improved.
Description
Technical Field
The invention relates to the technical field of ball screw pair detection, in particular to a precision ball screw pair detection system and a precision ball screw pair detection method.
Background
In the fields of numerical control machine tools, robots, printing machinery, aerospace and the like, a precise ball screw pair plays a vital role. The precise ball screw assembly usually operates in working environments such as high temperature, low temperature, high humidity, corrosive gas and the like, and equipment is easily worn, corroded and worn after long-term working in such environments, so that abnormality is caused. Therefore, it is important to diagnose the abnormality of the precision ball screw assembly.
The existing precise ball screw pair abnormality diagnosis mainly comprises a YOLO model, the YOLO model is a representative single-stage target detection algorithm, but the YOLO model possibly has the problem of inaccurate detection omission or positioning when processing a small object or a target dense area in a scene, so that the existing precise ball screw pair detection method is not robust enough to process a small target, a low-quality image or a complex scene of the precise ball screw pair, and the problem of easy loss or degradation of semantic information of deep features exists in the detection process.
Disclosure of Invention
The invention provides a precision ball screw pair detection system and a method thereof, which can effectively improve the global information extraction capability, avoid the loss or degradation of characteristic information in the transmission and interaction processes, and improve the accuracy and the robustness of the abnormality diagnosis of the precision ball screw pair to be detected.
The technical scheme for solving the technical problems is as follows:
The precise ball screw pair detection system comprises a precise ball screw pair detection middle table, a binocular shooting module, an image correction module, a region determination module, an image fusion module and an abnormality detection module; the anomaly detection module is a target YOLOv model; the precise ball screw pair detection middle table is respectively connected with the binocular camera module, the image correction module, the image fusion module, the area determination module and the abnormality detection module, and is used for storing and managing data of each module;
the binocular camera module is used for collecting a left original image and a right original image of the precise ball screw pair to be detected;
the image correction module is used for respectively carrying out transverse distortion correction on the left original image and the right original image to obtain a left target image and a right target image;
The binocular shooting module is further used for acquiring a left half image and a right half image of the precise ball screw pair to be detected under the condition that the precise ball screw pair to be detected is located at a preset position of the binocular shooting module;
the area determining module is used for determining an overlapping field area of view of the binocular shooting module based on the left half image, the right half image, the left original image and the right original image;
The image fusion module is used for fusing the left side target image, the right side target image and the overlapped visual field area to obtain an image after distortion correction;
the abnormality detection module is used for inputting the image after distortion correction to the target YOLOv model to obtain an abnormality detection result output by the target YOLOv model;
The target YOLOv model comprises a Swin transducer module and an asymptotic feature pyramid network, and is obtained by training and optimizing based on a nut abnormal data set and a ball screw abnormal data set.
The invention also provides a detection method of the precise ball screw pair, which is characterized by comprising the following steps:
acquiring a left original image and a right original image of a precise ball screw pair to be detected based on binocular imaging equipment;
Respectively carrying out transverse distortion correction on the left original image and the right original image to obtain a left target image and a right target image;
Acquiring a left half image and a right half image of the precise ball screw pair to be detected based on the binocular imaging equipment under the condition that the precise ball screw pair to be detected is positioned at a preset position of the binocular imaging equipment;
Determining an overlapping field of view region of the binocular imaging apparatus based on the left side half image, the right side half image, the left side original image, and the right side original image;
Fusing the left side target image, the right side target image and the overlapped visual field area to obtain an image after distortion correction;
inputting the image after distortion correction to a target YOLOv model to obtain an abnormality detection result output by the target YOLOv model;
The target YOLOv model comprises a Swin transducer module and an asymptotic feature pyramid network, and is obtained by training and optimizing based on a nut abnormal data set and a ball screw abnormal data set.
The present invention also provides an electronic device including: a memory for storing a computer software program; and the processor is used for reading and executing the computer software program so as to realize the precise ball screw pair detection method.
The invention also provides a non-transitory computer readable storage medium, wherein the storage medium stores a computer software program, and the computer software program realizes the precision ball screw pair detection method when being executed by a processor.
The invention also provides a computer program product comprising a computer program which when executed by a processor implements the precision ball screw pair detection method as described in any one of the above.
The beneficial effects of the invention are as follows: the method has the advantages that the Swin transducer module and the asymptotic characteristic pyramid network are combined, the target YOLOv model is improved and obtained, the target YOLOv model is used for carrying out anomaly detection on the images after distortion correction of the precise ball screw pair to be detected, global information extraction capability is effectively improved, loss or degradation of characteristic information in transmission and interaction processes is avoided, and accuracy and robustness of anomaly diagnosis of the precise ball screw pair to be detected are improved.
Drawings
FIG. 1 is a schematic diagram of a precision ball screw pair detection system provided by the invention;
FIG. 2 is a schematic flow chart of a detection method of a precision ball screw pair provided by the invention;
Fig. 3 is a schematic structural diagram of an electronic device provided by the present invention.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
In the description of the present invention, the terms "first," "second," and the like are used for descriptive purposes only and are not to be construed as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defining "a first" or "a second" may explicitly or implicitly include one or more of the described features. In the description of the present invention, the meaning of "a plurality" is two or more, unless explicitly defined otherwise.
In the description of the present invention, the term "for example" is used to mean "serving as an example, instance, or illustration. Any embodiment described as "for example" in this disclosure is not necessarily to be construed as preferred or advantageous over other embodiments. The following description is presented to enable any person skilled in the art to make and use the invention. In the following description, details are set forth for purposes of explanation. It will be apparent to one of ordinary skill in the art that the present invention may be practiced without these specific details. In other instances, well-known structures and processes have not been described in detail so as not to obscure the description of the invention with unnecessary detail. Thus, the present invention is not intended to be limited to the embodiments shown, but is to be accorded the widest scope consistent with the principles and features disclosed herein.
Referring to fig. 1, fig. 1 is a schematic structural diagram of a precision ball screw pair detection system provided by the invention, which comprises a precision ball screw pair detection middle stage, a binocular shooting module, an image correction module, a region determination module, an image fusion module and an anomaly detection module.
Further, the anomaly detection module is a target YOLOv model, the target YOLOv model comprises a Swin transform module and an asymptotic feature pyramid network, and training and optimizing are performed based on a nut anomaly data set and a ball screw anomaly data set to obtain the model.
Further, the precise ball screw pair detection middle table is respectively connected with the binocular shooting module, the image correction module, the image fusion module, the area determination module and the abnormality detection module, and stores and manages data of the binocular shooting module, the image correction module, the area determination module, the image fusion module and the abnormality detection module.
Further, the binocular camera module comprises a left-eye camera module and a right-eye camera module, so that the binocular camera module can collect a left original image and a right original image of the precise ball screw pair to be detected.
Further, the image correction module respectively carries out transverse distortion correction on the left original image and the right original image to obtain a left target image and a right target image.
Further, under the condition that the precise ball screw pair to be detected is located at a preset position of the binocular camera module, the binocular camera module obtains a left half image and a right half image of the precise ball screw pair to be detected.
Further, the region determining module determines an overlapping field of view region of the binocular camera module based on the left side half image, the right side half image, the left side original image, and the right side original image.
Further, the image fusion module fuses the left target image, the right target image and the overlapped visual field area to obtain an image after distortion correction.
Further, the abnormality detection module inputs the image after distortion correction to the target YOLOv model to obtain an abnormality detection result output by the target YOLOv model.
According to the embodiment of the invention, the target YOLOv model is improved and obtained by combining the Swin transducer module and the asymptotic feature pyramid network, and the target YOLOv model is used for carrying out anomaly detection on the image after distortion correction of the precise ball screw pair to be detected, so that the global information extraction capability is effectively improved, the loss or degradation of feature information in the transmission and interaction processes is avoided, and the accuracy and the robustness of anomaly diagnosis of the precise ball screw pair to be detected are further improved.
In an alternative embodiment, referring to fig. 2, fig. 2 is a flow chart of a method for detecting a precision ball screw assembly according to the present invention, the method includes:
And step 10, acquiring a left original image and a right original image of the precise ball screw pair to be detected based on binocular imaging equipment.
The number of the left original images and the right original images can be at least 2, each left original image is a round image obtained by shooting a precise ball screw pair to be detected by a left camera in the binocular camera, and each right original image is a round image obtained by shooting the precise ball screw pair to be detected by a right camera in the binocular camera.
The number of the precise ball screw pairs to be detected can be 1 or more, and when the number of the precise ball screw pairs to be detected is more than one, the diameters of every two adjacent precise ball screw pairs to be detected are different; when the number of the precise ball screw pairs to be detected is 1, the precise ball screw pairs to be detected are specifically concentric circular calibration plates, and the concentric circular calibration plates contain a plurality of concentric circles with different diameters, such as 4 concentric circles with different diameters, and the diameters of the concentric circles are 200mm, 300mm, 350mm and 400mm.
Specifically, when the precise ball screw pair to be detected reaches the shooting range of the binocular camera through the conveyor belt, the left camera shoots the precise ball screw pair to be detected and obtains a group of left original images, and then the shot group of left original images are further sent to the electronic equipment; meanwhile, the right camera also shoots the precise ball screw pair to be detected, a group of right original images are obtained, and the shot group of right original images are further sent to the electronic equipment; thus, the electronic equipment can acquire the left original image and the right original image acquired by the binocular shooting module.
And step 20, respectively carrying out transverse distortion correction on the left original image and the right original image to obtain a left target image and a right target image.
Under the condition that the precise ball screw pair to be detected enters the shooting range of the binocular shooting module even though the precise ball screw pair rotates through the conveyor belt, the binocular shooting module does not have longitudinal distortion and has transverse distortion, and therefore only the transverse distortion of the binocular shooting module is required to be corrected. Based on the above, the electronic device determines the left correction coefficient and the right correction coefficient corresponding to the binocular camera module by using a preset transverse distortion correction algorithm, and performs transverse distortion correction on the left original image by the left correction coefficient, and performs transverse distortion correction on the right original image by the right correction coefficient, so that a left target image and a right target image after transverse distortion correction can be obtained.
In an embodiment, the number of left target images is the same as and corresponds to the number of left original images, and the number of right target images is the same as and corresponds to the number of right original images.
Step 30, acquiring a left half image and a right half image of the precise ball screw pair to be detected based on the binocular imaging device under the condition that the precise ball screw pair to be detected is located at the preset position of the binocular imaging device.
Wherein, the number of the left half image and the right half image can be at least 2.
The preset position may be specifically an intermediate position of the binocular camera module.
Specifically, under the condition that the precise ball screw pair to be detected is placed in the middle position of the binocular camera module, incomplete half calibration plates can appear in the respective camera shooting ranges of the left camera and the right camera, at the moment, the left camera shoots the corresponding half calibration plates and obtains a group of left half images, and then the shot group of left half images are further sent to electronic equipment; meanwhile, the right camera also shoots the corresponding half calibration plate and obtains a group of right half images, and then the shot group of right half images are further sent to the electronic equipment; therefore, the electronic equipment can acquire the left half image and the right half image shot by the binocular shooting module.
Step 40, determining an overlapping field of view region of the binocular imaging apparatus based on the left side half image, the right side half image, the left side original image, and the right side original image.
It can be understood that, when the precise ball screw pair to be detected is placed in the middle position of the binocular shooting module to shoot, the respective shooting fields of the binocular shooting modules inevitably overlap and respectively shoot corresponding half calibration plates, so that the electronic equipment can combine left half images and right half images shot when the binocular shooting module overlaps and perform overlapping region calibration on left original images and right original images shot when the binocular shooting module does not overlap, thereby obtaining an overlapping field of view region of the binocular shooting module.
For example, the transverse diameter average values corresponding to the left half images and the right half images may be determined first, then the average value of the transverse diameter average values is calculated, and then the overlapping area calibration is performed based on the average value calculation result, the incomplete transverse diameter average value corresponding to the left half images and the incomplete transverse diameter average value corresponding to the right half images, so as to obtain the overlapping field of view of the binocular camera module.
And step 50, fusing the left target image, the right target image and the overlapped visual field area to obtain an image after distortion correction.
The electronic device can select a pure color subarea which does not contain the precise ball screw pair to be detected from the left target image, subtract the pure color subarea of the left target image from the overlapping visual field area, and then splice the subtracted area image and the right target image to obtain the image after distortion correction.
And step 60, inputting the image after distortion correction to a target YOLOv model to obtain an abnormality detection result output by the target YOLOv model. In the embodiment of the invention, the target YOLOv model comprises a Swin transform module and an asymptotic feature pyramid network, and the target YOLOv model is obtained by training and optimizing based on a nut abnormal data set and a ball screw abnormal data set.
Further, the image after distortion correction is input into a pre-trained target YOLOv model, and an output abnormality detection result can be obtained, wherein the abnormality detection result comprises an abnormality position and an abnormality type of the precise ball screw pair to be detected. It should be emphasized that the target YOLOv model to which the present embodiment is applied is obtained by modifying the original YOLOv model.
Specifically, on the basis of the original YOLOv model, a C2f module in a backbone network of the original YOLOv model is replaced by a Swin transform module so as to improve the attention to a detection target, and an asymptotic feature pyramid network is adopted to strengthen a multi-scale feature fusion process so as to avoid the loss or degradation of feature information in transmission and interaction processes. The process of improving the original YOLOv model is detailed in the examples below and will not be described in detail.
It should be further noted that the target YOLOV model is pre-trained, specifically, the target YOLOv model is obtained by training and optimizing based on the nut abnormal data set and the ball screw abnormal data set.
The nut abnormal data set and the ball screw abnormal data set are composed of nut abnormal images/precise ball screw pair abnormal images to be detected and corresponding label labels.
According to the embodiment of the invention, the image after distortion correction is input to a pre-trained target YOLOv model, and an abnormal detection result output by the target YOLOv model is obtained; the target YOLOv model comprises a Swin transducer module and an asymptotic feature pyramid network, and the target YOLOv model is obtained by training and optimizing based on a nut abnormal data set and a ball screw abnormal data set. According to the method, the Swin transducer module and the asymptotic feature pyramid network are combined, the target YOLOv model is improved, the target YOLOv model is used for carrying out anomaly detection on the images after distortion correction of the precise ball screw pair to be detected, global information extraction capability is effectively improved, loss or degradation of feature information in transmission and interaction processes is avoided, and accuracy and robustness of anomaly diagnosis of the precise ball screw pair to be detected are further improved.
In an alternative embodiment, step 20 performs lateral distortion correction on the left original image and the right original image to obtain a left target image and a right target image, including:
Under the condition that a left camera shoots a precise ball screw pair to be detected to obtain a plurality of left original images and a right camera shoots the precise ball screw pair to be detected to obtain a plurality of right original images, determining left distortion coefficients corresponding to the plurality of left original images and right distortion coefficients corresponding to the plurality of right original images; respectively longitudinally stretching a plurality of left original images based on left distortion coefficients to determine each left target image; and respectively performing longitudinal stretching on the plurality of right original images based on the right distortion coefficients to determine each right target image.
It should be noted that, because the binocular camera modules only have lateral distortion, when the left camera shoots the precise ball screw pair to be detected to obtain a group of left original images, and the right camera shoots the precise ball screw pair to be detected to obtain a group of right original images, the electronic device may calculate left lateral parameter values of each left original image, average each left lateral parameter value to obtain a left lateral parameter average value, calculate left longitudinal parameter values of each left original image, average each left longitudinal parameter value to obtain a left longitudinal parameter average value, and determine a ratio of the left lateral parameter average value to the left longitudinal parameter average value as a left distortion coefficient, for example, record the left distortion coefficient as a1.
Similarly, the electronic device may also calculate a right lateral parameter value of each right original image, average each right lateral parameter value to obtain a right lateral parameter average value, calculate a right longitudinal parameter value of each right original image, average each right longitudinal parameter value to obtain a right longitudinal parameter average value, and determine a ratio of the right lateral parameter average value to the right longitudinal parameter average value as a right distortion coefficient, e.g., record the right distortion coefficient as a2.
At this time, the electronic device may determine a left target image corresponding to each left original image by stretching each left original image longitudinally by a factor of a 1; likewise, the right target image corresponding to each right original image may also be determined by stretching each right original image by a factor of a2 in the longitudinal direction. Therefore, the purpose of respectively carrying out transverse distortion correction on the left original image and the right original image can be achieved, and reliable guarantee is provided for the subsequent image fusion effect.
In an alternative embodiment, determining left distortion coefficients corresponding to the plurality of left original images and right distortion coefficients corresponding to the plurality of right original images includes:
Determining a left lateral diameter and a left longitudinal diameter of each left original image, and determining a right lateral diameter and a right longitudinal diameter of each right original image, respectively; determining a left target transverse diameter corresponding to each left transverse diameter and a left target longitudinal diameter corresponding to each left longitudinal diameter, and determining a right target transverse diameter corresponding to each right transverse diameter and a right target transverse diameter corresponding to each right longitudinal diameter; the ratio of the left target lateral diameter to the left target longitudinal diameter is determined as a left distortion coefficient, and the ratio of the right target lateral diameter to the right target lateral diameter is determined as a right distortion coefficient.
The left target lateral diameter may be a left lateral diameter average value, the left target longitudinal diameter may be a left longitudinal diameter average value, the right target lateral diameter may be a right lateral diameter average value, and the right target lateral diameter may be a right lateral diameter average value.
For example, when the left camera shoots the precision ball screw pair to be detected to obtain at least two left original images, and the right camera shoots the precision ball screw pair to be detected to obtain at least two right original images, the electronic device may calculate a left lateral diameter of each left original image, average the left lateral diameters (d_h1) after averaging the left lateral diameters, calculate a left longitudinal diameter of each left original image, average the left longitudinal diameters (d_z1) after averaging the left longitudinal diameters, and determine a ratio of the left lateral diameter average to the left longitudinal diameter average as a left distortion coefficient a1, a1=d_h1/d_z1.
Similarly, the electronic device may also calculate a right lateral diameter of each right original image, average the right lateral diameters (d_h2) by averaging the right lateral diameters, calculate a right longitudinal diameter of each right original image, average the right longitudinal diameters (d_z2) by averaging the right longitudinal diameters, and determine a ratio of the right lateral diameter average to the right longitudinal diameter average as a right distortion coefficient a2, a2=d_h2/d_z2.
In an alternative embodiment, determining the left lateral diameter and the left longitudinal diameter of each left original image includes:
Determining an outer contour to be processed of a left original image, sequentially carrying out gray level processing, binarization processing and interference point removing processing on an initial contour inner circular image in the outer contour to be processed, and determining a target contour inner circular image based on a processing result; determining a target column pixel and a target row pixel with the smallest pixel value accumulation result in the target contour inner circular image; determining a target column boundary point based on the target column pixels and a target row boundary point based on the target row pixels; the left lateral diameter is determined based on the lateral straight line corresponding to the target column boundary point, and the left longitudinal diameter is determined based on the longitudinal straight line corresponding to the target row boundary point.
Specifically, for any one left original image in the left original images, firstly, selecting a circular contour from at least one circular contour contained in the left original image as an outer contour to be processed, if the left original image contains 3 concentric circular contours, the largest circular contour can be used as the outer contour to be processed, the smallest circular contour can be used as the outer contour to be processed, or the middle circular contour can be used as the outer contour to be processed; then, carrying out graying and binarization on the circular image in the initial outline in the outline to be processed, wherein the white pixel value in the binarized circular image is 1, and the black pixel value in the binarized circular image is 0, so that the purpose of the operation is to accurately extract a white area; and then, performing interference point removal processing on the binarized image by using morphological image processing methods and the like, such as removing black points in a white area, white points in a black area and the like, so as to obtain an internal circular image of the target outline.
Further, the column-by-column pixel value addition and the row-by-row pixel value addition are respectively carried out on the inner circular image of the target outline, and a target column pixel and a target row pixel with the minimum pixel value addition result are determined according to the pixel value addition results of each row and each column.
At this time, the column boundary point in the target column pixel is determined according to the number of white pixel points contained in the target column pixel, and the target row boundary point is determined according to the number of white pixel points contained in the target row pixel. For example, if 1 white pixel point is included in the target column pixel, the white pixel point may be used as the target column boundary point; if the target column pixel contains at least two white pixel points, the white pixel point at the middle position can be taken as the target column boundary point. In this way, the left lateral diameter of the left original image is determined by drawing a lateral straight line along the target column boundary point, and the left longitudinal diameter of the left original image is determined by drawing a longitudinal straight line along the target row boundary point,
For the right lateral diameter and the right longitudinal diameter of each right original image, the determination process may refer to the specific determination process of the left lateral diameter and the left longitudinal diameter of each left original image. And will not be described in detail herein.
Further, in order to ensure that the outer contour to be processed is more accurately selected, the left original image containing a plurality of concentric circular contours can be subjected to inverse color processing first and then the outer contour to be processed is selected. Based on this, the electronic device determines the to-be-processed outer contour of the left original image, and the specific implementation process may include:
under the condition that the precise ball screw pair to be detected is a concentric circle calibration plate with a plurality of different diameters, performing inverse color treatment on the left original image; and determining a target annular region from a plurality of annular regions contained in the calibration plate image obtained after the color reversal treatment, and determining the outer contour of the target annular region as the outer contour to be treated.
When the precise ball screw pair to be detected is a concentric calibration plate, a left original image can be obtained by shooting the precise ball screw pair to be detected through a left camera, and at the moment, the left original image is subjected to inverse color processing, namely, inverse color processing is performed in a manner of subtracting the pixel value of each pixel point on the left original image as the number of the minus number of 255, for example, 255-the pixel value of each pixel point can be obtained, and the calibration plate image obtained after the inverse color processing is obtained; and selecting a minimum circular ring area, a maximum circular ring area or any circular ring area between the minimum circular ring area and the maximum circular ring area from a plurality of circular ring areas contained in the calibration plate image obtained after the back-color treatment as a target circular ring area so as to determine the outer contour of the target circular ring area as the outer contour to be treated. Thus, the subsequent binarization effect can be ensured to be better, and the interference prevention effect is also better.
The precise ball screw pair to be detected is arranged to contain a plurality of concentric circles with different diameters, so that a plurality of calibration plate circular images with different sizes can be obtained simultaneously when one shooting is completed, and the accuracy of the precise ball screw pair detection is ensured by calculating the transverse diameter and the longitudinal diameter of each calibration plate circular image and obtaining an average value.
In an alternative embodiment, step 40 determines an overlapping field of view region of the binocular imaging apparatus based on the left side half image, the right side half image, the left side original image, and the right side original image, comprising:
Under the condition that the number of left half images, right half images, left original images and right original images is multiple, determining a first target axial distance corresponding to the multiple left half images and a second target axial distance corresponding to the multiple right half images; determining a target transverse diameter average value based on the left target transverse diameter corresponding to each left target image and the left target transverse diameter corresponding to each right target image; an overlapping field of view region is determined based on the first target axial distance, the second target axial distance, and the target transverse diameter mean.
Specifically, the precise ball screw pair to be detected is placed in the middle position of the binocular camera module to shoot, incomplete half-sides appear under the lenses of the left camera and the right camera respectively, a group of left half-side images and a group of right half-side images are shot, at the moment, the left incomplete transverse diameter of each left half-side image and the right incomplete transverse diameter of each right half-side image are calculated, the left incomplete transverse diameters are averaged, and the obtained average value is used as a first target axial distance (D_B1) corresponding to a plurality of left half-side images; similarly, each right incomplete transverse diameter is averaged and the resulting average is taken as the second target axial distance (D_B2) for the plurality of right half images.
At this time, the left lateral diameter average value (d_h1) and the right lateral diameter average value (d_h2) calculated above are called, the left lateral diameter average value (d_h1) and the right lateral diameter average value (d_h2) are averaged, and the resulting average value is determined as the target lateral diameter average value (D); thereby preventing errors. Thus, an overlapping field of view region (gap_d) can be obtained, gap_d=d_b1+d_b2-D.
In an alternative embodiment, step 50 includes fusing the left side target image, the right side target image, and the overlapping field of view region to obtain the distortion corrected image, including:
determining a difference value region between a preset pixel value sub-region and an overlapped visual field region of the left target image; and splicing the difference region and the right target image, and determining the image after distortion correction based on the splicing result.
For example, the preset pixel value sub-region may be a partial region in which pixel values in the left target image are all 0. Specifically, the electronic device may subtract the right semi-pure black area of the left target image from the overlapping field area to remove the overlapping field area for the left target image obtained by correcting the left original image photographed by the left camera by the lateral distortion and the right target image obtained by correcting the right original image photographed by the right camera by the lateral distortion; and then splicing the subtraction result image with the right side target image to obtain the image after distortion correction.
Based on the above embodiments, a detailed description will be given below of the process of obtaining the target YOLOv model by modifying the original YOLOv model.
The original YOLOv model includes Backbone network (Backbone), neck network (Neck), and predictive network (Head). The backbone network comprises 5 convolution modules (ConvModule), 4C 2f modules (CSPLayer _2Conv) and 1 spatial pyramid pooling module (SPPF). The Neck network includes 3 convolution modules, 4C 2f modules, 2 up-sampling modules (Upsample), and 4 connection layers (Concat).
The prediction network comprises 6 convolution modules and 6 two-dimensional convolution layers (Conv 2 d), each of which outputs the current training loss, namely bbox_loss (rectangular box loss) and cls_loss (class loss).
From the above, the original YOLOv model contains multiple CNN (Convolutional Neural Network ) layers, which are only good at capturing local information. Thus, the original YOLOv model includes a backbone network, neck network, and a prediction network, the backbone network including a convolution module, a C2f module, and a spatial pyramid pooling module; based on the Swin transducer module and the asymptotic feature pyramid network, the original YOLOv model is improved, which comprises the following steps: replacing a C2f module in the backbone network by using a Swin Transformer module; and replacing Neck the network with an asymptotic feature pyramid network.
In the modified YOLOv model, the C2f module in the backbone network of the original YOLOv model was replaced with a Swin transducer module; at the same time, the Neck network of the original YOLOv model was also replaced with the asymptotic feature pyramid network. In addition, element-by-element summation is not an efficient method in the overall feature fusion process, since there may be contradictions between different objects at some point between the hierarchies. To solve this problem, the present embodiment utilizes ASFF (ADAPTIVELY SPATIAL Feature Fusion) to assign different spatial weights to different levels of features in the multi-level Feature Fusion process, enhance the importance of key levels, and mitigate the impact of contradictory information from different objects.
In this embodiment, a Swin transducer module is used to replace a C2f module in a backbone network, an asymptotic feature pyramid network is used to replace a Neck network, so as to obtain a target YOLOv model, and then an image after distortion correction is input to a pre-trained target YOLOv model, so that an abnormality detection result output by the target YOLOv model can be obtained. According to the method, the Swin transducer module and the asymptotic feature pyramid network are combined, the target YOLOv model is improved, the target YOLOv model is used for carrying out anomaly detection on the images after distortion correction of the precise ball screw pair to be detected, global information extraction capability is effectively improved, loss or degradation of feature information in transmission and interaction processes is avoided, and accuracy and robustness of anomaly diagnosis of the precise ball screw pair to be detected are further improved.
On the basis of the above-described embodiment, a detailed description will be given below of a process of abnormality detection of an image after distortion correction using the target YOLOv model.
Inputting the image after distortion correction to a pre-trained target YOLOv model to obtain an abnormality detection result output by the target YOLOv model, wherein the method comprises the following steps of: based on the improved backbone network, extracting shallow feature vectors and deep feature vectors in the image after distortion correction; based on the asymptotic feature pyramid network, performing downsampling operation on the shallow feature vectors to obtain shallow features, and performing upsampling operation on the deep feature vectors to obtain deep features; adopting a self-adaptive spatial feature fusion mode to fuse deep features and shallow features layer by layer to obtain multi-scale features; predicting the multi-scale features based on a prediction network to obtain an abnormal detection result corresponding to the image after distortion correction; the abnormality detection result includes an abnormality position and an abnormality category.
Based on the improved backbone network, extracting shallow feature vectors and deep feature vectors in the image after distortion correction, namely sequentially inputting the image after distortion correction into Conv modules, swin transducer modules, conv modules, swin transducer modules and SPPF modules of the backbone network in the target YOLOv model for feature extraction to obtain the shallow feature vectors and the deep feature vectors in the image after distortion correction.
Further, feature vectors are aligned, i.e. downsampling operations are performed on shallow feature vectors to obtain shallow features, while upsampling operations are performed on deep feature vectors to obtain deep features.
And then, adopting a self-adaptive spatial feature fusion mode to fuse deep features and shallow features layer by layer so as to obtain a group of multi-scale features.
And finally, based on a prediction network, predicting the multi-scale features to obtain an abnormal detection result corresponding to the image after distortion correction.
The anomaly detection results include anomaly location and anomaly category including, but not limited to, cracks, fastener breaks, bolt breaks.
In this embodiment, shallow feature vectors and deep feature vectors in the image after distortion correction are extracted based on the improved backbone network, downsampling operation is performed on the shallow feature vectors based on the asymptotic feature pyramid network to obtain shallow features, upsampling operation is performed on the deep feature vectors to obtain deep features, and further the deep features and the shallow features are fused layer by layer in a self-adaptive spatial feature fusion mode to obtain multi-scale features, so that multi-scale features are predicted based on a prediction network, and an abnormal detection result corresponding to the image after distortion correction is obtained. According to the method, the Swin transducer module and the asymptotic feature pyramid network are combined, the target YOLOv model is improved, the target YOLOv model is used for carrying out anomaly detection on the images after distortion correction of the precise ball screw pair to be detected, global information extraction capability is effectively improved, loss or degradation of feature information in transmission and interaction processes is avoided, and accuracy and robustness of anomaly diagnosis of the precise ball screw pair to be detected are further improved.
Based on the above embodiment, the following training procedure for the target YOLOv model is as follows:
Training a target YOLOv model by adopting a transfer learning mode specifically comprises the following steps: constructing a nut abnormal data set and a ball screw abnormal data set; pre-training a target YOLOv model by using a nut abnormal data set to obtain a pre-training YOLOv model with optimal configuration parameters; based on the pre-training YOLOv model, the model parameters of the pre-training YOLOv model are finely tuned by utilizing the abnormal data set of the ball screw, so that a pre-trained target YOLOv model is obtained.
Transfer learning is a technique that applies acquired knowledge of known domains to target domains, and can transfer a trained network model from a large data set to a new data set, thereby realizing reuse of network model parameters and weights on the new data set. Aiming at the problem of large-scale lack of the abnormal image sample of the precise ball screw pair to be detected, a migration learning method is introduced to improve the model performance. The abnormal characteristics of the precise ball screw pair to be detected are similar to those of the nut, so that the collected nut abnormal data set is used for pre-training.
Constructing a nut abnormal data set, specifically, sorting a preset number of nut abnormal images collected from high-speed railways and the like, establishing an original data feature table based on the collected nut abnormal images, marking the features of data in the original data feature table, labeling the nut abnormal images according to corresponding categories before training a model, summarizing the feature values of each column in the data feature table to obtain a processed summarized data table, and dividing the processed summarized data table to obtain a training set and a testing set.
The preset number may be set according to actual requirements, and is not specifically limited herein.
In the training stage, the obtained nut abnormal image is subjected to data enhancement operations such as denoising, missing value filling, overturning (left-right overturning of an original picture), zooming (zooming of the original picture), color gamut conversion (changing of brightness, saturation and tone of the original picture), random clipping and the like, so that the effect of expanding training samples is achieved.
It should be noted that, the process of constructing the abnormal ball screw data set is the same as the process of constructing the abnormal nut data set, and will not be described here again. And then, pre-training the target YOLOv model by utilizing the nut abnormal data set after the enhancement processing to obtain the optimal weight of the target YOLOv model so as to obtain a pre-trained YOLOv model with optimal configuration parameters. Further, the pre-training YOLOv model is used as a starting point of transfer learning, the precise ball screw pair abnormality to be detected is selected as a target task, the model parameters of the pre-training YOLOv model are finely tuned by utilizing the constructed ball screw abnormality data set through limited times of training so as to adapt to the characteristics of a target domain, and the optimal model weight suitable for the target task is obtained, so that a trained target YOLOv model is obtained. The trained target YOLOv model can be directly used for detecting the abnormality of the image after distortion correction.
In this embodiment, a nut abnormal data set and a ball screw abnormal data set are constructed, a nut abnormal data set is utilized to pretrain a target YOLOv model, a pretrained YOLOv model with optimal configuration parameters is obtained, further, based on a pretrained YOLOv model, model parameters of the pretrained YOLOv model are finely tuned by utilizing the ball screw abnormal data set, a pretrained target YOLOv model is obtained, and therefore, an image after distortion correction is input to the pretrained target YOLOv model, and an abnormal detection result output by the target YOLOv model can be obtained. According to the method, the optimization target YOLOv model is trained by adopting a transfer learning mode, so that the model convergence speed is further increased, and the accuracy and the robustness of the abnormality diagnosis of the precise ball screw pair to be detected under the condition of a small sample are improved.
Fig. 3 illustrates a physical schematic diagram of an electronic device, as shown in fig. 3, where the electronic device may include: processor 310, communication interface (Communications Interface) 320, memory 330 and communication bus 340, wherein processor 310, communication interface 320 and memory 330 communicate with each other via communication bus 340. The processor 310 may invoke logic instructions in the memory 330 to perform a precision ball screw pair detection method comprising:
acquiring a left original image and a right original image of a precise ball screw pair to be detected based on binocular imaging equipment;
Respectively carrying out transverse distortion correction on the left original image and the right original image to obtain a left target image and a right target image;
Acquiring a left half image and a right half image of the precise ball screw pair to be detected based on the binocular imaging equipment under the condition that the precise ball screw pair to be detected is positioned at a preset position of the binocular imaging equipment;
determining an overlapping field of view region of the binocular imaging apparatus based on the left side half image, the right side half image, the left side original image, and the right side original image;
Fusing the left side target image, the right side target image and the overlapped visual field area to obtain an image after distortion correction;
inputting the image after distortion correction to a target YOLOv model to obtain an abnormality detection result output by the target YOLOv model;
The target YOLOv model comprises a Swin transducer module and an asymptotic feature pyramid network, and is obtained by training and optimizing based on a nut abnormal data set and a ball screw abnormal data set.
Further, the logic instructions in the memory 330 described above may be implemented in the form of software functional units and may be stored in a computer-readable storage medium when sold or used as a stand-alone product. Based on this understanding, the technical solution of the present invention may be embodied essentially or in a part contributing to the prior art or in a part of the technical solution in the form of a software product stored in a storage medium, comprising several instructions for causing a computer device (which may be a personal computer, a server, a network device, etc.) to perform all or part of the steps of the method of the embodiments of the present invention. And the aforementioned storage medium includes: a usb disk, a removable hard disk, a Read-Only Memory (ROM), a random access Memory (RAM, random Access Memory), a magnetic disk, or an optical disk, or other various media capable of storing program codes.
In another aspect, the present invention also provides a non-transitory computer readable storage medium having stored thereon a computer program which, when executed by a processor, is implemented to perform the precision ball screw pair detection method provided by the above methods, the method comprising:
acquiring a left original image and a right original image of a precise ball screw pair to be detected based on binocular imaging equipment;
Respectively carrying out transverse distortion correction on the left original image and the right original image to obtain a left target image and a right target image;
Acquiring a left half image and a right half image of the precise ball screw pair to be detected based on the binocular imaging equipment under the condition that the precise ball screw pair to be detected is positioned at a preset position of the binocular imaging equipment;
determining an overlapping field of view region of the binocular imaging apparatus based on the left side half image, the right side half image, the left side original image, and the right side original image;
Fusing the left side target image, the right side target image and the overlapped visual field area to obtain an image after distortion correction;
inputting the image after distortion correction to a target YOLOv model to obtain an abnormality detection result output by the target YOLOv model;
The target YOLOv model comprises a Swin transducer module and an asymptotic feature pyramid network, and is obtained by training and optimizing based on a nut abnormal data set and a ball screw abnormal data set.
The system embodiments described above are merely illustrative, in which elements illustrated as separate elements may or may not be physically separate, and elements shown as elements may or may not be physical elements, may be located in one place, or may be distributed over a plurality of network elements. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art will understand and implement the present invention without undue burden.
From the above description of the embodiments, it will be apparent to those skilled in the art that the embodiments may be implemented by means of software plus necessary general hardware platforms, or of course may be implemented by means of hardware. Based on such understanding, the foregoing technical solutions may be embodied essentially or in part in the form of a software product, which may be stored in a computer-readable storage medium, such as a ROM/RAM, a magnetic disk, an optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, a server, or a network device, etc.) to perform the various embodiments or methods of some parts of the embodiments.
Finally, it should be noted that: the above embodiments are only for illustrating the technical solution of the present invention, and are not limiting; although the invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims (10)
1. The precise ball screw pair detection system is characterized by comprising a precise ball screw pair detection middle stage, a binocular shooting module, an image correction module, a region determination module, an image fusion module and an abnormality detection module; the anomaly detection module is a target YOLOv model; the precise ball screw pair detection middle table is respectively connected with the binocular camera module, the image correction module, the image fusion module, the area determination module and the abnormality detection module, and is used for storing and managing data of each module;
the binocular camera module is used for collecting a left original image and a right original image of the precise ball screw pair to be detected;
the image correction module is used for respectively carrying out transverse distortion correction on the left original image and the right original image to obtain a left target image and a right target image;
The binocular shooting module is further used for acquiring a left half image and a right half image of the precise ball screw pair to be detected under the condition that the precise ball screw pair to be detected is located at a preset position of the binocular shooting module;
the area determining module is used for determining an overlapping field area of view of the binocular shooting module based on the left half image, the right half image, the left original image and the right original image;
The image fusion module is used for fusing the left side target image, the right side target image and the overlapped visual field area to obtain an image after distortion correction;
the abnormality detection module is used for inputting the image after distortion correction to the target YOLOv model to obtain an abnormality detection result output by the target YOLOv model;
The target YOLOv model comprises a Swin transducer module and an asymptotic feature pyramid network, and is obtained by training and optimizing based on a nut abnormal data set and a ball screw abnormal data set.
2. The method for detecting the precise ball screw pair is characterized by comprising the following steps of:
acquiring a left original image and a right original image of a precise ball screw pair to be detected based on binocular imaging equipment;
Respectively carrying out transverse distortion correction on the left original image and the right original image to obtain a left target image and a right target image;
Acquiring a left half image and a right half image of the precise ball screw pair to be detected based on the binocular imaging equipment under the condition that the precise ball screw pair to be detected is positioned at a preset position of the binocular imaging equipment;
Determining an overlapping field of view region of the binocular imaging apparatus based on the left side half image, the right side half image, the left side original image, and the right side original image;
Fusing the left side target image, the right side target image and the overlapped visual field area to obtain an image after distortion correction;
inputting the image after distortion correction to a target YOLOv model to obtain an abnormality detection result output by the target YOLOv model;
The target YOLOv model comprises a Swin transducer module and an asymptotic feature pyramid network, and is obtained by training and optimizing based on a nut abnormal data set and a ball screw abnormal data set.
3. The method of claim 2, wherein the target YOLOv model is derived from an original YOLOv model based on the Swin transform module and the asymptotic feature pyramid network; the original YOLOv model comprises a backbone network, a Neck network and a prediction network, wherein the backbone network comprises a convolution module, a C2f module and a spatial pyramid pooling module;
Accordingly, the original YOLOv model is improved based on the Swin transform module and the asymptotic feature pyramid network to obtain the target YOLOv8 model as follows: replacing a C2f module in the backbone network with the Swin Transformer module; and replacing the Neck network with the asymptotic feature pyramid network;
The step of inputting the image after distortion correction to a pre-trained target YOLOv model to obtain an abnormality detection result output by the target YOLOv model includes:
Based on the improved backbone network, extracting shallow feature vectors and deep feature vectors in the image after distortion correction;
based on the asymptotic feature pyramid network, performing downsampling operation on the shallow feature vector to obtain shallow features, and performing upsampling operation on the deep feature vector to obtain deep features;
Adopting a self-adaptive spatial feature fusion mode to fuse the deep features and the shallow features layer by layer to obtain multi-scale features;
predicting the multi-scale features based on the prediction network to obtain an abnormal detection result corresponding to the image after distortion correction;
the abnormality detection result comprises an abnormality position and an abnormality category.
4. The method for detecting a precision ball screw assembly according to claim 3, wherein the training of the target YOLOv model by means of transfer learning specifically comprises:
constructing the nut abnormal data set and the ball screw abnormal data set;
pre-training the target YOLOv model by using the nut abnormal data set to obtain a pre-trained YOLOv model with optimal configuration parameters;
Based on the pre-training YOLOv model, utilizing the abnormal data set of the ball screw to fine tune model parameters of the pre-training YOLOv model to obtain a pre-trained target YOLOv model;
wherein the step of constructing a nut anomaly dataset comprises:
Collecting abnormal images of nuts in a preset number;
labeling the nut abnormal image according to the abnormal category, and executing enhancement operation on the nut abnormal image to obtain the nut abnormal data set; the enhancement operations include one or more combinations of denoising, missing value filling, flipping, scaling, gamut conversion, and random clipping.
5. The method for detecting a precision ball screw assembly according to claim 2, wherein the step of performing the lateral distortion correction on the left side original image and the right side original image to obtain a left side target image and a right side target image, respectively, includes:
Under the condition that a left camera shoots the precise ball screw pair to be detected to obtain a plurality of left original images and a right camera shoots the precise ball screw pair to be detected to obtain a plurality of right original images, determining left distortion coefficients corresponding to the left original images and right distortion coefficients corresponding to the right original images;
Respectively longitudinally stretching a plurality of left original images based on the left distortion coefficients to determine each left target image;
And respectively carrying out longitudinal stretching on the plurality of right original images based on the right distortion coefficients to determine each right target image.
6. The method of claim 5, wherein determining left distortion coefficients corresponding to the plurality of left original images and right distortion coefficients corresponding to the plurality of right original images comprises:
determining a left lateral diameter and a left longitudinal diameter of each of the left original images, and determining a right lateral diameter and a right longitudinal diameter of each of the right original images, respectively;
Determining a left target transverse diameter corresponding to each left transverse diameter and a left target longitudinal diameter corresponding to each left longitudinal diameter, and determining a right target transverse diameter corresponding to each right transverse diameter and a right target transverse diameter corresponding to each right longitudinal diameter, respectively;
The ratio of the left target lateral diameter and the left target longitudinal diameter is determined as the left distortion coefficient, and the ratio of the right target lateral diameter and the right target lateral diameter is determined as the right distortion coefficient.
7. The precision ball screw assembly detection method of claim 6, wherein said determining a left lateral diameter and a left longitudinal diameter of each of said left raw images comprises:
Determining an outer contour to be processed of the left original image, sequentially performing gray level processing, binarization processing and interference point removing processing on an initial contour inner circular image in the outer contour to be processed, and determining an inner circular image of a target contour based on a processing result;
Determining a target column pixel and a target row pixel with the smallest pixel value accumulation result in the target outline inner circular image;
determining a target column boundary point based on the target column pixels and a target row boundary point based on the target row pixels;
the left lateral diameter is determined based on a lateral straight line corresponding to the target column boundary point, and the left longitudinal diameter is determined based on a longitudinal straight line corresponding to the target row boundary point.
8. The method of claim 7, wherein determining the to-be-processed outer contour of the left-side raw image comprises:
Under the condition that the precise ball screw pair to be detected is a concentric circle calibration plate with a plurality of different diameters, performing inverse color processing on the left original image;
And determining a target annular region from a plurality of annular regions contained in the calibration plate image obtained after the color reversal treatment, and determining the outer contour of the target annular region as the outer contour to be treated.
9. The method according to claim 2, wherein the fusing based on the left side target image, the right side target image, and the overlapping field of view region to obtain the distortion corrected image includes:
determining a difference value region between a preset pixel value sub-region of the left target image and the overlapped visual field region;
And splicing the difference value region and the right target image, and determining the image after distortion correction based on a splicing result.
10. The precision ball screw assembly detection method according to claim 2, wherein the determining the overlapping field of view region of the binocular imaging apparatus based on the left side half image, the right side half image, the left side original image, and the right side original image includes:
Determining a first target axial distance corresponding to a plurality of left half images and a second target axial distance corresponding to a plurality of right half images when the number of the left half images, the number of the right half images, the number of the left original images and the number of the right original images are respectively a plurality of;
Determining a target transverse diameter average value based on the left target transverse diameter corresponding to each left target image and the left target transverse diameter corresponding to each right target image;
The overlapping field of view region is determined based on the first target axial distance, the second target axial distance, and the target transverse diameter mean.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410503726.0A CN118072148B (en) | 2024-04-25 | 2024-04-25 | Precise ball screw pair detection system and method thereof |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410503726.0A CN118072148B (en) | 2024-04-25 | 2024-04-25 | Precise ball screw pair detection system and method thereof |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN118072148A true CN118072148A (en) | 2024-05-24 |
| CN118072148B CN118072148B (en) | 2024-06-25 |
Family
ID=91108123
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202410503726.0A Active CN118072148B (en) | 2024-04-25 | 2024-04-25 | Precise ball screw pair detection system and method thereof |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN118072148B (en) |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2018086348A1 (en) * | 2016-11-09 | 2018-05-17 | 人加智能机器人技术（北京）有限公司 | Binocular stereo vision system and depth measurement method |
| CN114565900A (en) * | 2022-01-18 | 2022-05-31 | 广州软件应用技术研究院 | Target detection method based on improved YOLOv5 and binocular stereo vision |
| CN114965489A (en) * | 2022-04-20 | 2022-08-30 | 江苏品德机电科技有限公司 | Conveying line with surface defect detection function for machining ball screw pair |
| CN115272694A (en) * | 2022-07-06 | 2022-11-01 | 南京理工大学 | Method for predicting remaining service life of ball screw pair based on raceway surface profile |
| CN116467575A (en) * | 2023-03-03 | 2023-07-21 | 青岛理工大学 | Real-time data monitoring system and method design of ball screw pair system |
| CN116934685A (en) * | 2023-06-07 | 2023-10-24 | 重庆理工大学 | Steel surface defect detection algorithm based on Focal module and deformable convolution |
| US20230394806A1 (en) * | 2022-06-07 | 2023-12-07 | Hefei University Of Technology | Quickly extraction of morphology characterization parameters of recycled concrete sand particles based on deep learning technology |
-
2024
- 2024-04-25 CN CN202410503726.0A patent/CN118072148B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2018086348A1 (en) * | 2016-11-09 | 2018-05-17 | 人加智能机器人技术（北京）有限公司 | Binocular stereo vision system and depth measurement method |
| CN114565900A (en) * | 2022-01-18 | 2022-05-31 | 广州软件应用技术研究院 | Target detection method based on improved YOLOv5 and binocular stereo vision |
| CN114965489A (en) * | 2022-04-20 | 2022-08-30 | 江苏品德机电科技有限公司 | Conveying line with surface defect detection function for machining ball screw pair |
| US20230394806A1 (en) * | 2022-06-07 | 2023-12-07 | Hefei University Of Technology | Quickly extraction of morphology characterization parameters of recycled concrete sand particles based on deep learning technology |
| CN115272694A (en) * | 2022-07-06 | 2022-11-01 | 南京理工大学 | Method for predicting remaining service life of ball screw pair based on raceway surface profile |
| CN116467575A (en) * | 2023-03-03 | 2023-07-21 | 青岛理工大学 | Real-time data monitoring system and method design of ball screw pair system |
| CN116934685A (en) * | 2023-06-07 | 2023-10-24 | 重庆理工大学 | Steel surface defect detection algorithm based on Focal module and deformable convolution |
Non-Patent Citations (3)
| Title |
|---|
| 上海慧腾: "滚珠丝杠的故障原因及维修方法", pages 1 - 2, Retrieved from the Internet <URL:https://zhuanlan.zhihu.com/p/595093427> * |
| 姜媛媛等: "轻量化的印刷电路板缺陷检测网络Multi-CR YOLO", 电子测量与仪器学报, no. 11, 19 March 2024 (2024-03-19), pages 217 - 224 * |
| 鲁伟俊等: "基于最小二乘法的非接触缺陷检测方法研究", 中国仪器仪表, no. 8, 31 August 2020 (2020-08-31), pages 78 - 82 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN118072148B (en) | 2024-06-25 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108961217B (en) | Surface defect detection method based on regular training | |
| CN111383209B (en) | Unsupervised flaw detection method based on full convolution self-encoder network | |
| CN107833238B (en) | Maximum connected domain marking method, target tracking method and augmented reality/virtual reality device | |
| CN111652085B (en) | Object identification method based on combination of 2D and 3D features | |
| CN109685097B (en) | Image detection method and device based on GAN | |
| CN115063430B (en) | Electric pipeline crack detection method based on image processing | |
| CN113516619B (en) | Product surface flaw identification method based on image processing technology | |
| JP2018506046A (en) | Method for detecting defects on the tire surface | |
| CN112200790B (en) | Cloth defect detection method, device and medium | |
| CN115830004A (en) | Surface defect detection method, device, computer equipment and storage medium | |
| CN111489337A (en) | Method and system for removing false defects through automatic optical detection | |
| CN114495098B (en) | Diaxing algae cell statistical method and system based on microscope image | |
| CN113469971A (en) | Image matching method, image matching detection device and storage medium | |
| CN114565552A (en) | Method and device for detecting surface defects of optical element and electronic equipment | |
| CN110472640B (en) | Target detection model prediction frame processing method and device | |
| CN117252815A (en) | Industrial part defect detection method, system, equipment and storage medium based on 2D-3D multi-mode image | |
| CN116205907A (en) | Decorative plate defect detection method based on machine vision | |
| CN114187247A (en) | Ampoule bottle printing character defect detection method based on image registration | |
| CN112614094B (en) | Insulator string abnormity positioning and identifying method based on sequence state coding | |
| CN108021914B (en) | Method for extracting character area of printed matter based on convolutional neural network | |
| CN118072148B (en) | Precise ball screw pair detection system and method thereof | |
| CN117132503A (en) | Method, system, equipment and storage medium for repairing local highlight region of image | |
| CN115797314B (en) | Method, system, equipment and storage medium for detecting surface defects of parts | |
| CN114202544B (en) | Complex workpiece defect detection method based on self-encoder | |
| CN115330688A (en) | Image anomaly detection method considering tag uncertainty |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant |