Assembly quality detection method and assembly quality detection device

Abstract

The present invention provides an assembly quality inspection method and an assembly quality inspection device. The assembly quality inspection method includes: an assembly quality inspection method, comprising: obtaining and fusing multiple sample images of a sample motor to obtain a fused sample image; using the fused sample image to perform unidirectional training on a constructed generative adversarial network; obtaining and fusing multiple inspection images of a motor to be inspected to obtain a fused inspection image; inputting the fused inspection image into the trained generative adversarial network to generate and output a labeled depth map of a fastening component of the motor to be inspected, so as to obtain installation information of the fastening component based on the labeled depth map; obtaining the maximum gap between the upper cover and the lower cover of the motor to be inspected based on the inspection image and judging whether it is less than or equal to a preset gap, so as to obtain motor fastening information; judging whether the assembly quality of the fastening component is qualified, so as to solve the problem of low efficiency of the assembly quality inspection method of screws on the motor in the prior art.

Description

Technical Field

[0001] The present invention relates to the technical field of air conditioner production, and in particular to an assembly quality detection method and an assembly quality detection device. Background Art

[0002] Currently, when the fan of the air conditioner outdoor unit is assembled and shipped, the assembly quality of the connecting screws of the motor on the fan needs to be tested, that is, the assembly quality of the connecting screws between the upper cover shell and the end cover shell of the motor needs to be tested to ensure that the motor can work normally.

[0003] In the existing technology, the screw detection method mainly relies on technicians to detect the screws. This detection method has the following main defects:

[0004] (1) In order to ensure a tight connection between the upper cover shell and the end cover shell of the motor, the types and quantities of connecting screws used in different models of motors are different. Due to the diversity of the types, lengths and colors of the screws, the nuts and washers that match the different screws are also different. Even for motors of the same model, the screws, nuts and washers used need to meet many consistency requirements. This leads to a more complicated screw detection process, requiring the screw detection technicians to have rich screw detection experience, which increases the threshold for screw detection.

[0005] (2) When the types of motors are different, technicians also need to consult relevant technical information to determine whether the wrong types of screws, nuts and washers are used. The lengths of some screws are small and difficult for the human eye to distinguish. When screws of different lengths are mixed and used, special measuring tools are required for measurement. This screw detection method is generally more dependent on manual experience and is inefficient, requiring a lot of manpower and time.

[0006] (3) Different technicians have different inspection experience and capabilities. Long-term inspection work can easily cause visual fatigue, resulting in uneven inspection quality. The assembly quality of the motor's fasteners cannot be guaranteed. Motors with unqualified assembly quality are prone to shaking, unstable speed, motor overheating, and severe power consumption during operation, and may even cause devastating damage to the motor, posing a serious potential danger to personal safety and affecting the company's quality reputation. At the same time, it increases the company's maintenance costs and wastes warehouse space.

[0007] (4) The screws detected manually cannot quickly and automatically provide the detection information of the screws of each motor, count the assembly qualification rate, error causes, etc. of the screws of multiple motors on the production line and form statistical reports. The coordination efficiency between manual operation and other machines is low, and it is impossible to complete the automatic alarm stop line and motor removal operations, resulting in a low overall degree of automation of the motor production line.

[0008] In response to this, some people have proposed using sensors to measure the displacement of the threaded end of the screw, using 3D cameras to measure the height of the threaded end of the screw, and using traditional image processing technology to detect the assembly position of the screw. However, sensors and 3D cameras are suitable for detecting screws in fixed positions, while the types, lengths, and colors of screws used in different motors are varied, making them difficult to detect using sensors and 3D cameras. Traditional image processing technology has high requirements for light. The light environment on the motor production line is variable, and an additional light-blocking housing is required to completely cover the motor and detection equipment. This causes a serious waste of the motor's production space. In addition, traditional image processing technology mainly uses template matching algorithms, which require the addition of specialized equipment to fix the position of the motor. When producing different motors, the screw image detection template must be manually switched, which greatly affects the motor's production efficiency. Summary of the Invention

[0009] The main purpose of the present invention is to provide an assembly quality detection method and an assembly quality detection device to solve the problem of low detection efficiency of the prior art method for detecting the assembly quality of screws on a motor.

[0010] To achieve the above-mentioned objective, according to one aspect of the present invention, a method for inspecting assembly quality is provided, comprising: obtaining a plurality of sample images of a sample motor, fusing the plurality of sample images to obtain a fused sample image; constructing a generative adversarial network, and performing unidirectional training on the generative adversarial network using the fused sample images to obtain a trained generative adversarial network; obtaining a plurality of inspection images of the motor to be inspected, and fusing the plurality of inspection images to obtain a fused inspection image; inputting the fused inspection image into the trained generative adversarial network to generate and output a labeled depth map of a fastening assembly of the motor to be inspected according to a color labeling rule, and determining whether actual installation information of the fastening assembly on the motor to be inspected meets the installation standard of the fastening assembly of the motor to be inspected based on the generated labeled depth map of the fastening assembly, so as to obtain fastening assembly installation information; obtaining a maximum gap between an upper cover and a lower cover of the motor based on at least one of the plurality of inspection images of the motor to be inspected, and determining whether the maximum gap is less than or equal to a preset gap, so as to obtain motor fastening information; and determining whether the assembly quality of the fastening assembly on the motor to be inspected is qualified based on the fastening assembly installation information and the motor fastening information.

[0011] Furthermore, the generative adversarial network is a CycleGAN network.

[0012] Furthermore, the steps of obtaining multiple sample images of the sample motor and fusing the multiple sample images to obtain a fused sample image specifically include: photographing the sample motor from the first side of the sample motor to obtain a first side sample image of the sample motor; photographing the sample motor from the second side of the sample motor to obtain a second side sample image of the sample motor; photographing the sample motor from above the sample motor to obtain an upper sample image of the sample motor; and fusing multiple sample images including the first side sample image, the second side sample image and the upper sample image in a fusion order to obtain a fused sample image.

[0013] Furthermore, the steps of obtaining multiple detection images of the motor to be detected and fusing the multiple detection images to obtain a fused detection image specifically include: photographing the motor to be detected from the first side of the motor to be detected to obtain a first side detection image of the motor to be detected; photographing the motor to be detected from the second side of the motor to be detected to obtain a second side detection image of the motor to be detected; photographing the motor to be detected from above the motor to be detected to obtain an upper detection image of the motor to be detected; and fusing multiple detection images including the first side detection image, the second side detection image and the upper detection image in a fusion order to obtain a fused detection image.

[0014] Furthermore, a generative adversarial network is constructed, and the generative adversarial network is trained using the fused sample image to obtain the trained generative adversarial network. The steps specifically include: pixel-level segmentation and annotation of the fused sample image according to the color annotation rule to obtain a first sample annotated depth map of the fastening component corresponding to the fused sample image; inputting the fused sample image and the first sample annotated depth map of the fastening component corresponding to the fused sample image into the generative adversarial network; generating a second sample annotated depth map of the fastening component corresponding to the fused sample image according to the color annotation rule by the generator in the generative adversarial network; and comparing the first sample annotated depth map of the corresponding fastening component with the second sample annotated depth map of the fastening component by the discriminator in the generative adversarial network to train the generative adversarial network.

[0015] Furthermore, between generating and outputting a labeled depth map of the fastening components of the motor to be inspected according to color labeling rules and judging whether the actual installation information of the fastening components on the motor to be inspected meets the installation standards of the fastening components of the motor to be inspected based on the generated labeled depth map of the fastening components, the assembly quality inspection method also includes: selecting whether to score the accuracy of the generated labeled depth map of the fastening components; if yes, scoring the accuracy of the generated labeled depth map of the fastening components to provide feedback on the parameters of the generative adversarial network and perform optimization training; if not, obtaining the fastening component installation information based on the labeled depth map of the fastening components.

[0016] Furthermore, the specific steps of performing feedback optimization and adjustment training on the parameters of the generative adversarial network include: scoring the accuracy of the generated annotated depth map of the fastening component to obtain a scoring value, and comparing the scoring value with a preset score; when the scoring value is lower than the preset score, retaining the fused detection image corresponding to the generated annotated depth map of the fastening component; manually annotating the fused detection image to obtain a manually annotated depth map of the fastening component; inputting the annotated depth image of the first fastening component and the generated annotated depth map of the fastening component into the trained generative adversarial network to perform feedback optimization and adjustment training on the parameters of the trained generative adversarial network.

[0017] Furthermore, the fastening assembly includes at least one of a screw, a nut, and a washer that cooperate with each other, and the rules for scoring the accuracy of the generated annotated depth map of the fastening assembly include: a positive correlation between the accuracy of the number of screws in the generated fastening assembly and the scoring value; and / or a positive correlation between the accuracy of the number of nuts in the generated fastening assembly and the scoring value; and / or a positive correlation between the accuracy of the number of washers in the generated fastening assembly and the scoring value; and / or a positive correlation between the accuracy of the color of the screws in the generated fastening assembly and the scoring value; and / or a positive correlation between the accuracy of the color of the nuts in the generated fastening assembly and the scoring value; and / or a positive correlation between the accuracy of the color of the nuts in the generated fastening assembly and the scoring value; and / or a positive correlation between the accuracy of the color of the nuts in the generated fastening assembly and the scoring value; There is a positive correlation between the accuracy of the color of the washer in the fastening assembly and the score value; and / or there is a positive correlation between the accuracy of the position of the screw in the generated fastening assembly and the score value; and / or there is a positive correlation between the accuracy of the position of the nut in the generated fastening assembly and the score value; and / or there is a positive correlation between the accuracy of the position of the washer in the generated fastening assembly and the score value; and / or there is a positive correlation between the accuracy of the shape of the screw in the generated fastening assembly and the score value; and / or there is a positive correlation between the accuracy of the shape of the nut in the generated fastening assembly and the score value; and / or there is a positive correlation between the accuracy of the shape of the washer in the generated fastening assembly and the score value.

[0018] Furthermore, the specific steps of determining whether the actual installation information of the fastening components on the motor to be tested meets the installation standards of the fastening components of the motor to be tested based on the generated annotated depth map of the fastening components to obtain the installation information of the fastening components include: performing image segmentation on the generated annotated depth map of the fastening components using the eight-connected region principle to obtain a segmented image; performing statistics on the annotated depth map of the fastening components according to color annotation rules; and comparing the statistically analyzed annotated information with the installation standards of the fastening components of the motor to be tested to determine whether the actual installation information of multiple fastening components of the motor to be tested are all qualified to obtain the installation information of the fastening components.

[0019] Furthermore, the specific steps of obtaining the motor tightening information include: using a traditional image processing algorithm to process at least one of the multiple detection images; using a Hough line detection algorithm to detect a first straight line corresponding to the surface of the upper cover shell of the motor to be detected on the side close to the end cover shell of the motor to be detected; using the Hough line detection algorithm to detect a second straight line corresponding to the surface of the end cover shell on the side close to the upper cover shell; calculating the maximum gap between the first straight line and the second straight line; judging whether the maximum gap between the first straight line and the second straight line is less than or equal to a preset gap; when the maximum gap is less than or equal to the preset gap, the motor tightening information is that the motor tightening is qualified; when the maximum gap is greater than the preset gap, the motor tightening information is that the motor tightening is unqualified.

[0020] Furthermore, the specific judgment steps for determining whether the assembly quality of the fastening component is qualified include: when at least one of the motor fastening information and the fastening component installation information is unqualified, it is judged that the assembly quality of the fastening component is unqualified, and a rejection information is output to reject the unqualified motor; when both the motor fastening information and the fastening component installation information are qualified, it is judged that the assembly quality of the fastening component is qualified.

[0021] Furthermore, after determining whether the assembly quality of the fastening assembly is qualified based on the fastening assembly installation information and the motor fastening information, the assembly quality detection method includes: outputting the fastening assembly installation information and the motor fastening information to a statistical report system to generate a statistical report.

[0022] According to another aspect of the present invention, an assembly quality inspection device is provided, which is applicable to the above-mentioned assembly quality inspection method; the assembly quality inspection device includes: an image acquisition and fusion unit, which is used to acquire multiple sample images of a sample motor and fuse the multiple sample images to obtain a fused sample image, or to acquire multiple detection images of a motor to be inspected and fuse the multiple detection images to obtain a fused detection image; a training unit, which constructs a generative adversarial network and performs unidirectional training on the generative adversarial network using the fused sample images to obtain a trained generative adversarial network; a first information generation unit, which inputs the fused detection image into the trained generative adversarial network, A marked depth map of the fastening components of the motor to be tested is generated and output according to the color marking rules, so as to judge whether the actual installation information of the fastening components on the motor to be tested meets the installation standards of the fastening components of the motor to be tested based on the generated marked depth map of the fastening components, so as to obtain the fastening component installation information; a second information generation unit obtains the maximum gap between the upper cover shell and the end cover shell of the motor to be tested based on at least one of the multiple detection pictures of the motor to be tested, and judges whether the maximum gap is less than or equal to the preset gap, so as to obtain the motor fastening information; an assembly quality judgment unit judges whether the assembly quality of the fastening components on the motor to be tested is qualified based on the fastening component installation information and the motor fastening information.

[0023] Applying the technical solution of the present invention, the assembly quality inspection method of the present invention includes: obtaining multiple sample images of a sample motor, fusing the multiple sample images to obtain a fused sample image; constructing a generative adversarial network, and performing unidirectional training on the generative adversarial network using the fused sample image to obtain a trained generative adversarial network; obtaining multiple detection images of the motor to be inspected, and fusing the multiple detection images to obtain a fused detection image; inputting the fused detection image into the trained generative adversarial network to generate and output a labeled depth map of the fastening component of the motor to be inspected according to a color labeling rule, so as to judge whether the actual installation information of the fastening component on the motor to be inspected meets the installation standard of the fastening component of the motor to be inspected based on the labeled depth map, so as to obtain fastening component installation information; obtaining the maximum gap between the upper cover and the lower cover of the motor based on at least one of the multiple detection images of the motor to be inspected, and judging whether the maximum gap is less than or equal to the preset gap to obtain motor fastening information; judging whether the assembly quality of the fastening component on the motor to be inspected is qualified based on the fastening component installation information and the motor fastening information. In this way, the assembly quality detection method of the present invention can automatically detect the assembly quality of fastening components on motors of different models to generate a visual screw annotation depth image, and obtain the optimal screw assembly quality detection method through the trained generative adversarial network, avoiding the low efficiency limitation of relying on manual experience to detect fastening components, greatly reducing labor costs and time costs, and does not affect the production process of the motor, realizing a highly automated detection process without human intervention, so as to solve the problem of low detection efficiency of the existing method for detecting the assembly quality of screws on motors. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] The accompanying drawings, which constitute part of this application, are intended to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are intended to explain the present invention and do not constitute an undue limitation of the present invention. In the accompanying drawings:

[0025] Figure 1 A flow chart showing an embodiment of an assembly quality inspection method according to the present invention is shown;

[0026] Figure 2 Shown Figure 1 A schematic structural diagram of an embodiment of the motor shown;

[0027] Figure 3 Shown Figure 2 Schematic diagram of the shooting direction of the motor shown;

[0028] Figure 4 Shown Figure 1 Schematic diagram of the fusion of multiple pictures shown;

[0029] Figure 5 shows a color-coded diagram of a first embodiment of a fastening assembly according to the present invention;

[0030] Figure 6 shows a color-coded diagram of a second embodiment of a fastening assembly according to the present invention;

[0031] Figure 7 A color-coded diagram showing a third embodiment of a fastening assembly according to the present invention is shown.

[0032] The above drawings include the following reference numerals:

[0033] 100. Motor; 101. Upper cover housing; 102. End cover housing; 200. Fastening assembly; 201. Screw; 202. Nut; 203. Washer; 300. First side view; 400. Second side view; 500. Upper view; 600. Conveyor line. DETAILED DESCRIPTION

[0034] It should be noted that, in the absence of conflict, the embodiments and features of the embodiments in this application can be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0035] like Figures 1 to 7 As shown, the present invention provides an assembly quality inspection method, including: obtaining multiple sample images of a sample motor, fusing the multiple sample images to obtain a fused sample image; constructing a generative adversarial network, and performing unidirectional training on the generative adversarial network using the fused sample image to obtain a trained generative adversarial network; obtaining multiple detection images of the motor to be inspected, and fusing the multiple detection images to obtain a fused detection image; inputting the fused detection image into the trained generative adversarial network to generate and output a labeled depth map of the fastening component of the motor to be inspected according to a color labeling rule, so as to judge whether the actual installation information of the fastening component on the motor to be inspected meets the installation standard of the fastening component of the motor to be inspected according to the generated labeled depth map of the fastening component, so as to obtain fastening component installation information; obtaining the maximum gap between the upper cover and the lower cover of the motor according to at least one of the multiple detection images of the motor to be inspected, and judging whether the maximum gap is less than or equal to the preset gap, so as to obtain motor fastening information; judging whether the assembly quality of the fastening component on the motor to be inspected is qualified according to the fastening component installation information and the motor fastening information.

[0036] The assembly quality inspection method of the present invention includes: obtaining multiple sample images of a sample motor, fusing the multiple sample images to obtain a fused sample image; constructing a generative adversarial network, and performing unidirectional training on the generative adversarial network using the fused sample images to obtain a trained generative adversarial network; obtaining multiple detection images of the motor to be inspected, and fusing the multiple detection images to obtain a fused detection image; inputting the fused detection image into the trained generative adversarial network to generate and output a labeled depth map of the fastening component of the motor to be inspected according to a color labeling rule, so as to judge whether the actual installation information of the fastening component on the motor to be inspected meets the installation standard of the fastening component of the motor to be inspected based on the generated labeled depth map of the fastening component, so as to obtain fastening component installation information; obtaining the maximum gap between the upper cover and the lower cover of the motor based on at least one of the multiple detection images of the motor to be inspected, and judging whether the maximum gap is less than or equal to the preset gap, so as to obtain motor fastening information; and judging whether the assembly quality of the fastening component on the motor to be inspected is qualified based on the fastening component installation information and the motor fastening information. In this way, the assembly quality detection method of the present invention can automatically detect the assembly quality of fastening components on motors of different models to generate a visual screw annotation depth image, and obtain the optimal assembly quality detection method through the trained generative adversarial network, avoiding the low efficiency limitation of relying on manual experience to detect fastening components, greatly reducing labor costs and time costs, and does not affect the production process of the motor, realizing a highly automated detection process without human intervention, so as to solve the problem of low detection efficiency of the assembly quality detection method of screws on motors in the prior art.

[0037] Specifically, the annotated depth map of the fastening components generated according to the color marking rules contains information such as the number, position, color and shape of the screws, nuts and washers in the fastening components on the motor to be inspected, whether they are used incorrectly, and whether the screws are tightened in place.

[0038] like Figure 1 As shown, the steps of obtaining multiple sample images of the sample motor and fusing the multiple sample images to obtain a fused sample image specifically include: photographing the sample motor from the first side of the sample motor to obtain a first side sample image of the sample motor; photographing the sample motor from the second side of the sample motor to obtain a second side sample image of the sample motor; photographing the sample motor from above the sample motor to obtain an upper sample image of the sample motor; and fusing multiple sample images including the first side sample image, the second side sample image and the upper sample image in a fusion order to obtain a fused sample image.

[0039] like Figure 1As shown, the steps of obtaining multiple detection images of the motor to be detected and fusing the multiple detection images to obtain a fused detection image specifically include: photographing the motor to be detected from the first side of the motor to be detected to obtain a first side detection image of the motor to be detected; photographing the motor to be detected from the second side of the motor to be detected to obtain a second side detection image of the motor to be detected; photographing the motor to be detected from above the motor to be detected to obtain an upper detection image of the motor to be detected; and fusing the multiple detection images including the first side detection image, the second side detection image and the upper detection image in a fusion order to obtain a fused detection image.

[0040] Among them, the fusion is completed through matrix superposition. A color image includes three pages of RGB. The three color images are superimposed together to form a new three-dimensional matrix RGBRGBRGB.

[0041] Specifically, before obtaining multiple sample images of a sample motor or multiple test images of a motor to be tested, it is necessary to install a high-pixel, color first camera and a second camera on opposite sides of the test station of the motor production line. These two cameras are based on the opposite sides of the assembly line to shoot in a mirror-like manner, and their field of view can completely cover the relatively set first side and second side of the corresponding motor.

[0042] Among them, the first side image 300 and the second side image 400 taken by the first camera and the second camera respectively contain information such as the height, width, side shape, the gap between the upper cover and the lower cover of the motor, and the height, width, position, color, side shape, etc. of all screws, nuts and washers.

[0043] like Figure 3 As shown, the shooting positions of the first camera and the second camera are respectively from the left side and the right side of the motor 100 on the conveyor line 600.

[0044] Specifically, when obtaining multiple sample images of a sample motor or multiple test images of a motor to be tested, a third camera needs to be installed directly above the motor. The third camera is a color camera with the same pixels as the first camera and the second camera, and is used to take a bird's-eye view of the upper image 500 of the motor.

[0045] The pictures taken by the camera mainly include the size and shape of the top of the motor, as well as the position, color, size and shape of all screws, nuts and washers from a top-down view.

[0046] like Figure 3 As shown, the shooting position of the third camera is from the top of the motor 100 on the conveyor line 600.

[0047] In this way, by setting up cameras in multiple directions, it can be applied to production inspection environments with different light changes.

[0048] After taking the three pictures, they are fused in a certain order to obtain a fused picture. Such a fused picture generally contains information about the sample motor or the motor to be tested and the screws, nuts, and washers assembled thereon.

[0049] like Figure 4 As shown, the fusion order from left to right is the first side image 300, the second side image 400 and the upper image 500.

[0050] Specifically, the generative adversarial network includes a generator and a discriminator. The generative adversarial network is constructed, and the generative adversarial network is trained using a fusion sample image to obtain the trained generative adversarial network. The steps specifically include: pixel-level segmentation and annotation of the fusion sample image according to a color annotation rule to obtain a first sample annotated depth map of the fastening component corresponding to the fusion sample image; inputting the fusion sample image and the first sample annotated depth map of the fastening component corresponding to the fusion sample image into the generative adversarial network; generating a second sample annotated depth map of the fastening component corresponding to the fusion sample image according to the color annotation rule by the generator in the generative adversarial network; and comparing the first sample annotated depth map of the corresponding fastening component with the second sample annotated depth map of the fastening component by the discriminator in the generative adversarial network to train the generative adversarial network.

[0051] The depth map includes information such as the position, shape, and use errors of the screw 201, nut 202, and washer 203 in the fastening assembly 200 installed on the motor 100, as well as whether the screw 201 in the fastening assembly 200 is tightened properly. The specific color marking rules are as follows:

[0052] When the type and height of the nut 202 meet the installation standards of the motor fastening assembly, it is marked in red; when at least one of the type and height of the nut 202 does not meet the installation standards of the motor fastening assembly, it is marked in orange; when the nut 202 is missing, it is marked in dark orange;

[0053] When the type and height of the washer 203 meet the installation standards of the fastening assembly of the motor, it is marked in yellow. When at least one of the type and height of the washer 203 does not meet the installation standards of the fastening assembly of the motor, it is marked in green. When the washer 203 is missing, it is marked in dark green.

[0054] When the type and height of the screw meet the installation standards of the motor fastening assembly, it is marked in blue; when at least one of the type and height of the screw does not meet the installation standards of the motor fastening assembly, it is marked in gray;

[0055] When the exposed height of the top of the screw 201 is lower than the installation standard of the fastening component of the motor, it means that the screw 201 is not tightened in place and is marked in purple; when the screw 201 is missing, it is marked in dark purple;

[0056] Other areas of non-fastening components are marked in white; in addition, some motors do not require the use of gasket 203, so the gasket 203 part is also marked in white.

[0057] like Figure 5 As shown, the marked depth map indicates that the corresponding motor requires four fastening components, and the types of screws 201, nuts 202 and washers 203 used are all correct, where the blue area represents the screws 201, the red area represents the hexagonal nuts 202, and the yellow area represents the elastic washers 203.

[0058] like Figure 6 As shown, the marked depth diagram indicates that the corresponding motor requires four fastening components, and the types of screws 201 and nuts 202 are used correctly, wherein the blue area represents the screws 201 and the red area represents the hexagonal flange nuts 202.

[0059] like Figure 7 As shown, the marked depth diagram indicates that the corresponding motor 100 requires eight fastening components, and some of the screws 201, nuts 202 and washers 203 do not meet the installation standards of the fastening components of the motor, and the screws 201, nuts 202 and washers are missing; the positions of the eight fastening components are respectively top, bottom, left, right, upper left, lower left, upper right and lower right; among them, only the screws 201, nuts 202 and washers 203 at the lower left and upper right positions are used correctly; the color marking of the fastening component at the top position indicates that the screws 201, nuts 202 and washers 203 are missing at this position; the color marking of the fastening component at the upper left position indicates that the screws 201 and nuts 202 at this position are used correctly, but the washer 203 is missing; the color marking of the fastening component at the left position indicates that the screws 201 and nuts 202 at this position are used correctly, but the washer 203 is missing; the color marking of the fastening component at the left position It indicates that the screw 201 and nut 202 at this position are used correctly, and at least one of the type and height of the washer 203 does not meet the installation standard of the fastening assembly of the motor; the color marking of the fastening assembly at the position directly below indicates that the nut 202 and washer 203 at this position are used correctly, but at least one of the type and height of the screw 201 does not meet the installation standard of the fastening assembly of the motor; the color marking of the fastening assembly at the lower right position indicates that the nut 202 and washer 203 at this position are used correctly, but the height of the screw 201 is not tightened to the right position; the color marking of the fastening assembly at the position directly right indicates that the screw 201 and washer 203 at this position are used correctly, but at least one of the type and height of the nut 202 used does not meet the installation standard of the fastening assembly of the motor.

[0060] Preferably, the generative adversarial network is a CycleGAN network (which can also be replaced by other generative adversarial networks). The CycleGAN network is a convolutional neural network whose main structure includes a generator and a discriminator.

[0061] The present invention uses a combination of multiple fused sample images and labeled first sample annotated depth maps of fastener components corresponding to the fused sample images as a training set to perform unidirectional training on the CycleGAN network, so that the generator in the CycleGAN network can learn the optimal fastener component annotation method, so that the generated annotated depth images of the fastener components are increasingly accurate, and the discriminator can distinguish the generated annotated images from the real annotated images increasingly accurately, and finally the annotated depth maps generated by the generator are difficult for the discriminator to distinguish between true and false, thereby obtaining a trained network.

[0062] In the assembly quality inspection method of the present invention, between generating and outputting a labeled depth map of the fastening components of the motor to be inspected according to color labeling rules and judging whether the actual installation information of the fastening components on the motor to be inspected meets the installation standards of the fastening components of the motor to be inspected based on the generated labeled depth map of the fastening components, the assembly quality inspection method also includes: selecting whether to score the accuracy of the generated labeled depth map of the fastening components; if yes, scoring the accuracy of the generated labeled depth map of the fastening components to provide feedback on the parameters of the generative adversarial network and perform optimization training; if not, obtaining the fastening component installation information based on the labeled depth map of the fastening components.

[0063] Specifically, the specific steps of feedback optimization and adjustment training of the parameters of the generative adversarial network include: scoring the accuracy of the generated labeled depth map of the fastening component to obtain a scoring value, and comparing the scoring value with a preset score; when the scoring value is lower than the preset score, retaining the fused detection image corresponding to the generated labeled depth map of the fastening component; manually annotating the fused detection image to obtain a manually annotated depth map of the fastening component; inputting the manually annotated depth map of the fastening component and the generated annotated depth map of the fastening component into the trained generative adversarial network to feedback optimize and adjust the parameters of the trained generative adversarial network to make the annotated depth map of the fastening component generated by the generative adversarial network more accurate.

[0064] Specifically, the scoring adopts a 100-point system. The higher the score, the better the generative adversarial network is at generating annotated depth maps of fastened components, and the more accurate the detection results. The scoring process can be performed during the detection process or after the detection is completed.

[0065] In the assembly quality inspection method of the present invention, the fastening assembly includes at least one of a screw, a nut and a washer that cooperate with each other, and the rules for scoring the accuracy of the generated annotated depth map of the fastening assembly include: a positive correlation between the accuracy of the number of screws in the generated fastening assembly and the scoring value; and / or a positive correlation between the accuracy of the number of nuts in the generated fastening assembly and the scoring value; and / or a positive correlation between the accuracy of the number of washers in the generated fastening assembly and the scoring value; and / or a positive correlation between the accuracy of the color of the screws in the generated fastening assembly and the scoring value; and / or a positive correlation between the accuracy of the color of the nuts in the generated fastening assembly and the scoring value; and / or there is a positive correlation between the accuracy of the color of the washer in the generated fastening assembly and the score value; and / or there is a positive correlation between the accuracy of the position of the screw in the generated fastening assembly and the score value; and / or there is a positive correlation between the accuracy of the position of the nut in the generated fastening assembly and the score value; and / or there is a positive correlation between the accuracy of the position of the washer in the generated fastening assembly and the score value; and / or there is a positive correlation between the accuracy of the shape of the screw in the generated fastening assembly and the score value; and / or there is a positive correlation between the accuracy of the shape of the nut in the generated fastening assembly and the score value; and / or there is a positive correlation between the accuracy of the shape of the washer in the generated fastening assembly and the score value.

[0066] In the assembly quality inspection method of the present invention, the specific steps of determining whether the actual installation information of the fastening components on the motor to be inspected meets the installation standards of the fastening components of the motor to be inspected based on the generated annotated depth map of the fastening components to obtain the installation information of the fastening components include: performing image segmentation on the generated annotated depth map of the fastening components using the eight-connected region principle to obtain a segmented image; performing statistics on the annotated information on the segmented image according to color annotation rules, and comparing the statistical annotated information with the installation standards of the fastening components of the motor to be inspected to determine whether the actual installation information of multiple fastening components of the motor to be inspected are all qualified to obtain the installation information of the fastening components.

[0067] Among them, the actual installation information of multiple fastening components includes whether the screws, nuts, and washers installed on the motor to be tested meet the installation standards of the fastening components, the number of screws, nuts, and washers that meet the installation standards of the fastening components, the number that do not meet the installation standards of the fastening components, and the number that are missing, etc. Only when the actual installation information of all screws, nuts, and washers meets the installation standards of the fastening components of the motor to be tested and there is no missing, can the fastening component installation information of the motor be considered qualified.

[0068] like Figure 2As shown, the housing of the motor 100 includes an upper cover housing 101 and an end cover housing 102 , and the free end of the rotating shaft of the motor 100 extends to a side of the upper cover housing 101 away from the end cover housing 102 .

[0069] In the assembly quality inspection method of the present invention, the specific steps of obtaining the motor fastening information include: using a traditional image processing algorithm to process at least one of a plurality of inspection images; using a Hough line detection algorithm to detect a first straight line corresponding to the surface of the upper cover shell of the motor to be inspected on the side close to the end cover shell of the motor to be inspected; using a Hough line detection algorithm to detect a second straight line corresponding to the surface of the end cover shell on the side close to the upper cover shell; calculating the maximum gap between the first straight line and the second straight line; judging whether the maximum gap between the first straight line and the second straight line is less than or equal to a preset gap; when the maximum gap is less than or equal to the preset gap, the motor fastening information is that the motor fastening is qualified; when the maximum gap is greater than the preset gap, the motor fastening information is that the motor fastening is unqualified.

[0070] The detection images used in the specific step of obtaining the motor fastening information are the first side image 300 and the second side image 400 .

[0071] In the assembly quality inspection method of the present invention, the specific judgment steps for determining whether the assembly quality of the fastening component is qualified include: when at least one of the motor fastening information and the fastening component installation information is unqualified, it is judged that the assembly quality of the fastening component is unqualified, and a rejection information is output to reject the unqualified motor; when both the motor fastening information and the fastening component installation information are qualified, it is judged that the assembly quality of the fastening component is qualified.

[0072] Specifically, after determining whether the assembly quality of the fastening assembly is qualified based on the fastening assembly installation information and the motor fastening information, the assembly quality detection method includes: outputting the fastening assembly installation information and the motor fastening information to a statistical report system to generate a statistical report.

[0073] The present invention provides an assembly quality inspection device, which is applicable to the above-mentioned assembly quality inspection method; the assembly quality inspection device comprises: an image acquisition and fusion unit, which is used to acquire multiple sample images of a sample motor and fuse the multiple sample images to obtain a fused sample image, or to acquire multiple detection images of a motor to be inspected and fuse the multiple detection images to obtain a fused detection image; a training unit, which constructs a generative adversarial network and performs unidirectional training on the generative adversarial network using the fused sample images to obtain a trained generative adversarial network; a first information generation unit, which inputs the fused detection image into the trained generative adversarial network, A marked depth map of the fastening components of the motor to be tested is generated and output according to the color marking rules, so as to judge whether the actual installation information of the fastening components on the motor to be tested meets the installation standards of the fastening components of the motor to be tested based on the marked depth map, so as to obtain the fastening component installation information; a second information generation unit obtains the maximum gap between the upper cover shell and the end cover shell of the motor to be tested based on at least one of the multiple detection pictures of the motor to be tested, and judges whether the maximum gap is less than or equal to the preset gap, so as to obtain the motor fastening information; an assembly quality judgment unit judges whether the assembly quality of the fastening components on the motor to be tested is qualified based on the fastening component installation information and the motor fastening information.

[0074] The assembly quality inspection method of the present invention has the following advantages:

[0075] (1) It solves the problem that manual inspection is cumbersome and requires a lot of manpower, material resources and time.

[0076] When switching between different models of motors, manual inspection of screws requires consulting the installation standards of the fastener components in the installation documents of different motors, and a production line needs to be equipped with at least one specially trained inspection technician. The assembly quality inspection method of the present invention does not require human intervention and can accurately inspect the installation information of the fastener components on different motors within seconds.

[0077] (2) It solves the problem that manual testing relies on the experience and judgment of the testers, resulting in uneven testing quality.

[0078] The assembly quality inspection method of the present invention enables the neural network to learn the optimal inspection method through training with a large number of training samples, so as to infer whether the number of screws of different motors is accurate, whether the screws are missing, whether the screws are screwed in place, whether the nuts are used incorrectly, whether the washers are used incorrectly, and other information, so that the inspection standards of the fastening components of motors of the same model are unified.

[0079] (3) The problem of low intelligence and low detection efficiency in the existing detection methods for the assembly quality of fastener components is solved.

[0080] When technicians detect that a fastener component is installed incorrectly, they need to manually send information to instruct the machine to stop the line and remove the component. At the same time, they also need to manually collect statistics on screw detection information to generate statistical reports.

[0081] The assembly quality detection method of the present invention can automatically detect the assembly quality of fastening components on motors of different models, quickly and automatically provide line stop information, alarm information and rejection information, and automatically generate motor screw detection reports.

[0082] (4) The problem of difficulty in detection caused by variable light in the detection environment, variable screw positions, and variable number of screws has been solved.

[0083] The assembly quality inspection method of the present invention is applicable to a production line environment with variable light conditions, so as to inspect fastening components at different positions and in different numbers.

[0084] From the above description, it can be seen that the above embodiments of the present invention achieve the following technical effects:

[0085] The assembly quality inspection method of the present invention includes: obtaining multiple sample images of a sample motor, fusing the multiple sample images to obtain a fused sample image; constructing a generative adversarial network, and performing unidirectional training on the generative adversarial network using the fused sample images to obtain a trained generative adversarial network; obtaining multiple detection images of the motor to be inspected, and fusing the multiple detection images to obtain a fused detection image; inputting the fused detection image into the trained generative adversarial network to generate and output a labeled depth map of the fastening component of the motor to be inspected according to a color labeling rule, so as to judge whether the actual installation information of the fastening component on the motor to be inspected meets the installation standard of the fastening component of the motor to be inspected based on the generated labeled depth map of the fastening component, so as to obtain fastening component installation information; obtaining the maximum gap between the upper cover and the lower cover of the motor based on at least one of the multiple detection images of the motor to be inspected, and judging whether the maximum gap is less than or equal to the preset gap, so as to obtain motor fastening information; and judging whether the assembly quality of the fastening component on the motor to be inspected is qualified based on the fastening component installation information and the motor fastening information. In this way, the assembly quality detection method of the present invention can automatically detect the assembly quality of fastening components on motors of different models to generate a visual screw annotation depth image, and obtain the optimal assembly quality detection method through the trained generative adversarial network, avoiding the low efficiency limitation of relying on manual experience to detect fastening components, greatly reducing labor costs and time costs, and does not affect the production process of the motor, realizing a highly automated detection process without human intervention, so as to solve the problem of low detection efficiency of the assembly quality detection method of screws on motors in the prior art.

[0086] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, they indicate the presence of features, steps, operations, devices, components and / or combinations thereof.

[0087] Unless otherwise specifically stated, the relative arrangement of the parts and steps, numerical expressions and numerical values ​​set forth in these embodiments do not limit the scope of the present application. At the same time, it should be understood that, for ease of description, the sizes of the various parts shown in the drawings are not drawn according to actual proportional relationships. The techniques, methods and equipment known to those of ordinary skill in the relevant art may not be discussed in detail, but where appropriate, the techniques, methods and equipment should be considered as part of the authorization specification. In all examples shown and discussed here, any specific values ​​should be interpreted as being merely exemplary and not as limitations. Therefore, other examples of the exemplary embodiments may have different values. It should be noted that similar numbers and letters represent similar items in the following figures, and therefore, once an item is defined in one figure, it does not need to be further discussed in subsequent figures.

[0088] In the description of this application, it should be understood that the directions or positional relationships indicated by directional words such as "front, back, up, down, left, right", "horizontal, vertical, vertical, horizontal" and "top, bottom" are usually based on the directions or positional relationships shown in the accompanying drawings. They are only for the convenience of describing this application and simplifying the description. Unless otherwise specified, these directional words do not indicate or imply that the device or element referred to must have a specific direction or be constructed and operated in a specific direction. Therefore, they cannot be understood as limiting the scope of protection of this application; the directional words "inside and outside" refer to the inside and outside relative to the outline of each component itself.

[0089] For ease of description, spatially relative terms such as "above", "above", "on the upper surface of", "above", etc. may be used herein to describe the spatial positional relationship of a device or feature to other devices or features as shown in the figures. It should be understood that spatially relative terms are intended to include different orientations of the device in use or operation in addition to the orientation described in the figures. For example, if the device in the drawings is inverted, the device described as "above other devices or structures" or "above other devices or structures" will be positioned as "below other devices or structures" or "below other devices or structures". Thus, the exemplary term "above" can include both "above" and "below". The device can also be positioned in other different ways (rotated 90 degrees or in other orientations), and the spatially relative descriptions used here are interpreted accordingly.

[0090] In addition, it should be noted that the use of terms such as "first" and "second" to limit components is only for the convenience of distinguishing the corresponding components. Unless otherwise stated, the above terms have no special meaning and therefore cannot be understood as limiting the scope of protection of this application.

[0091] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention are intended to be within the scope of protection of the present invention.

Claims

1. An assembly quality inspection method, characterized in that: include: Acquire multiple sample images of the sample motor, and fuse the multiple sample images to obtain a fused sample image; Constructing a generative adversarial network, and performing unidirectional training on the generative adversarial network using the fused sample image to obtain a trained generative adversarial network; Acquire multiple detection images of the motor to be detected, and fuse the multiple detection images to obtain a fused detection image; Inputting the fused detection image into the trained generative adversarial network to generate and output a labeled depth map of the fastening assembly of the motor to be detected according to a color labeling rule, and determining whether actual installation information of the fastening assembly on the motor to be detected meets the installation standard of the fastening assembly of the motor to be detected based on the generated labeled depth map of the fastening assembly, so as to obtain fastening assembly installation information; Obtaining a maximum gap between an upper cover and a lower cover of the motor according to at least one of the plurality of inspection images of the motor to be inspected, and determining whether the maximum gap is less than or equal to a preset gap to obtain motor fastening information; Determining whether the assembly quality of the fastening assembly on the motor to be inspected is qualified according to the fastening assembly installation information and the motor fastening information; The step of obtaining multiple sample images of the sample motor and fusing the multiple sample images to obtain a fused sample image specifically includes: photographing the sample motor from a first side surface of the sample motor to obtain a sample image of the first side surface of the sample motor; photographing the sample motor from a second side surface of the sample motor to obtain a sample image of the second side surface of the sample motor; photographing the sample motor from above the sample motor to obtain an upper sample picture of the sample motor; fusing the plurality of sample images including the first side sample image, the second side sample image, and the upper sample image in a fusing order to obtain the fused sample image; The step of constructing a generative adversarial network and training the generative adversarial network using the fused sample image to obtain a trained generative adversarial network specifically includes: Performing pixel-level segmentation and annotation on the fused sample image according to the color annotation rule to obtain a first sample annotated depth map of the fastening component corresponding thereto; Inputting the fused sample image and the first sample labeled depth map of the fastening component corresponding to the fused sample image into the generative adversarial network; Generating, by a generator in the generative adversarial network according to the color labeling rule, a second sample labeled depth map of the fastening component corresponding to the fused sample image; The discriminator in the generative adversarial network compares the first sample labeled depth map of the corresponding fastening component with the second sample labeled depth map of the fastening component to train the generative adversarial network.

2. The assembly quality inspection method according to claim 1, characterized in that: The generative adversarial network is a CycleGAN network.

3. The assembly quality inspection method according to claim 1, characterized in that: The step of obtaining multiple detection images of the motor to be detected and fusing the multiple detection images to obtain a fused detection image specifically includes: photographing the motor to be inspected from a first side surface of the motor to be inspected to obtain a first side surface inspection image of the motor to be inspected; photographing the motor to be inspected from a second side surface of the motor to be inspected to obtain a second side surface inspection image of the motor to be inspected; photographing the motor to be inspected from above to obtain an upper inspection picture of the motor to be inspected; The multiple detection pictures including the first side detection picture, the second side detection picture and the upper detection picture are fused in a fusion order to obtain the fused detection picture.

4. The assembly quality inspection method according to claim 1, characterized in that: Between generating and outputting a labeled depth map of the fastening components of the motor to be inspected according to the color labeling rule and judging whether actual installation information of the fastening components on the motor to be inspected meets the installation standard of the fastening components of the motor to be inspected based on the generated labeled depth map of the fastening components, the assembly quality inspection method further includes: selecting whether to score the accuracy of the generated annotated depth map of the fastener component; If yes, scoring the accuracy of the generated annotated depth map of the fastening component to provide feedback to the parameters of the generative adversarial network and perform optimization training; If not, the fastening component installation information is obtained according to the marked depth map of the fastening component.

5. The assembly quality inspection method according to claim 4, characterized in that: The specific steps of performing feedback optimization and adjustment training on the parameters of the generative adversarial network include: Scoring the accuracy of the generated annotated depth map of the fastening component to obtain a scoring value, and comparing the scoring value with a preset scoring value; When the score value is lower than the preset score, retaining the fused detection image corresponding to the generated annotated depth map of the fastening component; Manually annotating the fused detection image to obtain a manually annotated depth map of the fastening component; The manually annotated depth map of the fastening component and the generated annotated depth map of the fastening component are input into the trained generative adversarial network to perform feedback optimization and adjustment training on the parameters of the trained generative adversarial network.

6. The assembly quality inspection method according to claim 5, characterized in that: The fastening assembly includes at least one of a screw, a nut, and a washer that cooperate with each other. The rules for scoring the accuracy of the generated annotated depth map of the fastening assembly include: There is a positive correlation between the accuracy of the number of screws in the generated fastener assembly and the score; and / or There is a positive correlation between the accuracy of the number of nuts in the generated fastener assembly and the score; and / or There is a positive correlation between the accuracy of the number of washers in the generated fastener assembly and the score; and / or There is a positive correlation between the accuracy of the color of the screws in the generated fastener assembly and the score value; and / or There is a positive correlation between the accuracy of the color of the nut in the generated fastener assembly and the score value; and / or There is a positive correlation between the accuracy of the color of the washers in the generated fastener assembly and the score value; and / or There is a positive correlation between the accuracy of the position of the screws in the generated fastener assembly and the score value; and / or There is a positive correlation between the accuracy of the position of the nut in the generated fastener assembly and the score; and / or There is a positive correlation between the accuracy of the position of the washers in the generated fastener assembly and the score; and / or There is a positive correlation between the accuracy of the shape of the screw in the generated fastener assembly and the score; and / or There is a positive correlation between the accuracy of the shape of the nut in the generated fastener assembly and the score; and / or There is a positive correlation between the accuracy of the shape of the washers in the generated fastener assembly and the score value.

7. The assembly quality inspection method according to any one of claims 1 to 6, characterized in that: The specific steps of determining whether the actual installation information of the fastening assembly on the motor to be inspected meets the installation standard of the fastening assembly of the motor to be inspected based on the generated annotated depth map of the fastening assembly to obtain the fastening assembly installation information include: Performing image segmentation on the generated annotated depth map of the fastening component using the eight-connected region principle to obtain a segmented image; Collecting statistics on the marking information on the marking depth map of the fastening component according to the color marking rule; The statistically labeled information is compared with the installation standard of the fastening components of the motor to be tested to determine whether the actual installation information of the multiple fastening components of the motor to be tested are all qualified, so as to obtain the fastening component installation information.

8. The assembly quality inspection method according to any one of claims 1 to 6, characterized in that: The specific steps to obtain motor fastening information include: Processing at least one of the plurality of detection images using a traditional image processing algorithm; Using a Hough line detection algorithm to detect a first straight line corresponding to a surface of the upper cover housing of the motor to be detected that is close to the end cover housing of the motor to be detected; Using a Hough line detection algorithm to detect a second straight line corresponding to a surface of the end cover housing on a side close to the upper cover housing; calculating a maximum gap between the first straight line and the second straight line; determining whether a maximum gap between the first straight line and the second straight line is less than or equal to a preset gap; When the maximum gap is less than or equal to the preset gap, the motor fastening information indicates that the motor fastening is qualified; When the maximum gap is greater than the preset gap, the motor fastening information indicates that the motor fastening is unqualified.

9. The assembly quality inspection method according to claim 1, characterized in that: The specific steps of judging whether the assembly quality of the fastening component is qualified include: When at least one of the motor fastening information and the fastening assembly installation information is unqualified, it is determined that the assembly quality of the fastening assembly is unqualified, and a rejection information is output to reject the unqualified motor; When both the motor fastening information and the fastening assembly installation information are qualified, it is determined that the assembly quality of the fastening assembly is qualified.

10. The assembly quality inspection method according to claim 1 or 9, characterized in that: After determining whether the assembly quality of the fastening assembly is qualified according to the fastening assembly installation information and the motor fastening information, the assembly quality detection method includes: The fastening assembly installation information and the motor fastening information are output to a statistical report system to generate a statistical report.

11. An assembly quality inspection device, characterized in that: The assembly quality inspection device is applicable to the assembly quality inspection method described in any one of claims 1 to 10; The assembly quality detection device comprises: An image acquisition and fusion unit, configured to acquire multiple sample images of a sample motor and fuse the multiple sample images to obtain a fused sample image, or to acquire multiple detection images of a motor to be detected and fuse the multiple detection images to obtain a fused detection image; A training unit, configured to construct a generative adversarial network and perform unidirectional training on the generative adversarial network using the fused sample image to obtain a trained generative adversarial network; a first information generation unit, configured to input the fused detection image into the trained generative adversarial network to generate and output a labeled depth map of the fastening assembly of the detected motor according to a color labeling rule, and to determine, based on the labeled depth map, whether actual installation information of the fastening assembly on the motor to be detected complies with installation standards for the fastening assembly of the motor to be detected, thereby obtaining fastening assembly installation information; a second information generating unit, configured to obtain a maximum gap between an upper cover housing and an end cover housing of the motor to be inspected based on at least one of the plurality of inspection images of the motor to be inspected, and determine whether the maximum gap is less than or equal to a preset gap, so as to obtain motor fastening information; an assembly quality judgment unit, configured to judge whether the assembly quality of the fastening assembly on the motor to be inspected is qualified based on the fastening assembly installation information and the motor fastening information; The image acquisition and fusion unit is specifically used for: photographing the sample motor from a first side surface of the sample motor to obtain a sample image of the first side surface of the sample motor; photographing the sample motor from a second side surface of the sample motor to obtain a sample image of the second side surface of the sample motor; photographing the sample motor from above the sample motor to obtain an upper sample picture of the sample motor; fusing the plurality of sample images including the first side sample image, the second side sample image, and the upper sample image in a fusing order to obtain the fused sample image; The training unit is specifically used for: Performing pixel-level segmentation and annotation on the fused sample image according to the color annotation rule to obtain a first sample annotated depth map of the fastening component corresponding thereto; Inputting the fused sample image and the first sample labeled depth map of the fastening component corresponding to the fused sample image into the generative adversarial network; Generating, by a generator in the generative adversarial network according to the color labeling rule, a second sample labeled depth map of the fastening component corresponding to the fused sample image; The discriminator in the generative adversarial network compares the first sample labeled depth map of the corresponding fastening component with the second sample labeled depth map of the fastening component to train the generative adversarial network.

Images