CN117218188A - Deep learning positioning optimization method for can body processing - Google Patents
Deep learning positioning optimization method for can body processing Download PDFInfo
- Publication number
- CN117218188A CN117218188A CN202310951295.XA CN202310951295A CN117218188A CN 117218188 A CN117218188 A CN 117218188A CN 202310951295 A CN202310951295 A CN 202310951295A CN 117218188 A CN117218188 A CN 117218188A
- Authority
- CN
- China
- Prior art keywords
- data
- formula
- filtering
- expressed
- ultrasonic
- 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.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 51
- 238000012545 processing Methods 0.000 title claims abstract description 32
- 238000005457 optimization Methods 0.000 title claims abstract description 24
- 238000013135 deep learning Methods 0.000 title claims abstract description 23
- 238000001914 filtration Methods 0.000 claims abstract description 48
- 238000003466 welding Methods 0.000 claims abstract description 32
- 238000001514 detection method Methods 0.000 claims abstract description 28
- 238000010191 image analysis Methods 0.000 claims abstract description 21
- 238000012549 training Methods 0.000 claims abstract description 20
- 230000004927 fusion Effects 0.000 claims abstract description 12
- 238000013480 data collection Methods 0.000 claims abstract description 4
- 238000005070 sampling Methods 0.000 claims description 23
- 238000012544 monitoring process Methods 0.000 claims description 12
- 238000000605 extraction Methods 0.000 claims description 11
- PCHJSUWPFVWCPO-UHFFFAOYSA-N gold Chemical compound [Au] PCHJSUWPFVWCPO-UHFFFAOYSA-N 0.000 claims description 9
- 239000010931 gold Substances 0.000 claims description 9
- 229910052737 gold Inorganic materials 0.000 claims description 9
- 230000003321 amplification Effects 0.000 claims description 8
- 238000003199 nucleic acid amplification method Methods 0.000 claims description 8
- 238000003860 storage Methods 0.000 claims description 5
- 238000012512 characterization method Methods 0.000 claims description 4
- 238000004519 manufacturing process Methods 0.000 claims description 4
- 238000009826 distribution Methods 0.000 claims description 3
- 230000006870 function Effects 0.000 claims description 3
- 238000010438 heat treatment Methods 0.000 claims description 3
- 238000010586 diagram Methods 0.000 description 8
- 238000004458 analytical method Methods 0.000 description 5
- 238000010276 construction Methods 0.000 description 2
- 238000003709 image segmentation Methods 0.000 description 2
- 238000013459 approach Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000013499 data model Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003012 network analysis Methods 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
- 238000001931 thermography Methods 0.000 description 1
- 239000013598 vector Substances 0.000 description 1
Landscapes
- Investigating Or Analyzing Materials By The Use Of Ultrasonic Waves (AREA)
Abstract
The application provides a deep learning positioning optimization method for can body processing, which adopts thermal image analysis and ultrasonic detection means to collect data of a heated can body, and provides a model of multilayer fusion data to fuse two data into a group of Feature maps. In addition, according to the accuracy of the independent training of the two data and the error rate of data collection, the weights of the two data in the Feature Map are distributed, so that a more accurate means obtains higher weights. In addition, the application adopts a large-range filtering and small-range extracting mode to extract the welding seam position, so that the running speed of the whole network is higher, and the error rate is lower.
Description
Technical Field
The application relates to the field of can body processing, in particular to a deep learning positioning optimization method for can body processing.
Background
Positioning of the tank in conventional engineering is a relatively complex and error-prone task, since the tank is often of complex shape and size and may be hot and hot in the engineering field. The most commonly used way is for a worker to manually measure the size and position of the can and then use conventional measuring tools for positioning. However, the method is easily affected by human errors, so that positioning is inaccurate, and the detection mode is slower, so that the requirement of industrialized rapid production is difficult to meet. Thus, a more automated and faster positioning method is currently continued.
Disclosure of Invention
In order to solve the technical problems, the application provides a method for monitoring the approach personnel in a transformer substation monitoring scene, which can realize the rapid positioning of the position and the angle of the tank body.
In order to achieve the above purpose, the technical scheme adopted by the application is as follows:
a deep learning positioning optimization method for can body processing comprises the following steps:
1) Setting a filtering interval;
in the step, two types of data need to be characterized, namely ultrasonic monitoring data and thermal image analysis data;
the method comprises the steps of acquiring a temperature distribution image of the surface of a tank body by using a thermal imager after the heating process of the tank body and before the positioning of a gold stamping process, and acquiring acoustic resistance characteristic data by using an ultrasonic detection device, wherein the collection and characterization of the data are divided into two steps, wherein range filtering data collection is firstly carried out, namely, sampling is carried out in a larger range according to a filtering formula each time, and region overlapping is not generated among acquired pictures;
2) Data image acquisition setting;
the width of the welding line in the middle of the tank body is 3-4 mm, a balance filtering rule is adopted, namely, two filtering methods are adopted, the two filtering methods are collected according to a sampling formula, wherein the first one adopts a large-range filtering rule, and then small-space extraction work is carried out on a region with higher possibility of the welding line after filtering;
3) Setting data fusion;
the multi-layer data fusion model is used for carrying out interval weight fusion on the thermal image analysis data collected at this time and the ultrasonic monitoring data according to a merging formula, wherein the influence of the two data is according to the actual detection quality of the two data and the network correctness during independent training, and the ultrasonic data is directly filled in according to an amplification formula;
4) Setting a network architecture;
the method comprises the steps that the flow of a network architecture is specifically related to, the data output in the step 3) are in the form of fused FeatureMap, wherein the fused FeatureMap comprises thermal image analysis data and ultrasonic detection data, convolution operation is carried out on the fused FeatureMap in the step, and when the data divided after a large-range filtering rule is input into the convolution operation, whether welding seams exist in the data is judged;
5) Outputting a welding line coordinate value;
the detection and storage of all weld coordinate values are completed in the step 4), and the weld coordinate values are processed and output according to a coordinate processing formula in the step;
6) Algorithm deployment and application;
in the above steps, positioning of the welding seam of the can body is completed, deployment and application are carried out in the steps, an algorithm model is deployed into a production line, specific coordinate values of the welding seam of the can body are output, the position of the can body at the moment is judged according to the fed back coordinate values, the angle from the position needing gold stamping is judged, and the can body is rotated to directly gold stamping.
As a further improvement of the present application, the filtering formula in the step 1) is expressed as:
wherein the filtering formula is expressed as follows:
wherein D is the circumference of the round body of the tank body,for the super parameter, d is the sampling width of each time set for sampling in a large range, wherein the areas of each large range do not overlap, and when the rate set by the factory pipeline is higher, the ratio of the sampling width to the sampling width is +.>Setting to a smaller value, so that the number of times of detection of the whole large area is reduced, and when the rate of setting the factory pipeline is smaller, the rate is increased +.>Is a larger value, thereby reducing the first filtering range and improving the detection quality.
As a further improvement of the present application, the sampling formula in the step 2) is expressed as:
wherein, the sampling formula is expressed as follows:
wherein T is expressed as the weld size, k is expressed as a super parameter, S is the size input into the algorithm model, k=1 is taken as a specific intermediate critical point, when the k value is smaller than 1, the large-range filtering rule is indicated at the moment, and when the k value is larger than 1, the small-space extraction work is indicated at the moment.
As a further improvement of the present application, the amplification formula in step 3) is expressed as:
wherein the amplification formula is expressed as follows:
x=Random(x 1 、x 2 、x 3 、x 4 、x 5 )
wherein x is 1 、x 2 、x 3 、x 4 、 x 5 is respectively the reflection amplitude, the echo time, the acoustic impedance, the acoustic energy loss and the sound velocity, wherein Random indicates a Random function, namely a value is randomly extracted from selected elements, x indicates the numerical value of the pixel point, and the expansion of one-dimensional ultrasonic data can be completed through the formula, so that the expansion is consistent with the FeatureMap of thermal image analysis;
the combined formula in the step 3) is expressed as:
wherein the combining formula is expressed as follows:
wherein n represents the number of ultrasonic data to be inserted, ero R ,Erp C Analyzing whether the individual training network training is the error rate of the welding seam or not for the thermal image respectively, and ultrasonically detecting whether the individual training network training is the error rate of the welding seam or not, and arc C In order to detect ultrasonic wave of the tank body, the accuracy of ultrasonic wave is arc R The accuracy of thermal image analysis is realized when thermal image data acquisition is carried out on the tank body.
As a further improvement of the present application, the coordinate processing formula in the step 5) is expressed as:
in the step 4), detection and storage of all weld coordinate values are completed, and in the step, the weld coordinate values are processed and output, wherein a coordinate processing formula is as follows:
wherein j represents the final output weld position, i is the coordinate value when the weld exists in all small ranges, and p is the number value when the weld exists in all small ranges.
Compared with the prior art, the application has the beneficial effects that:
1) The application provides a deep learning positioning optimization method for can body processing, which provides an optimization method for large-range filtering and small-range extraction, and can better and quickly extract key positioning information and timely feed back the key positioning information.
2) The deep learning positioning optimization method for can body processing provided by the application provides a model for fusing two kinds of data into a group of FeatureMap by using a multi-layer fusion data model, and adjusts and optimizes weights of the two kinds of data according to accuracy of the two kinds of data and an error of experimental measurement, so that the expression capacity improvement of a network is finally enhanced
3) The deep learning positioning optimization method for can body processing can realize rapid and accurate positioning of the position of the welding seam of the can body, thereby reducing construction difficulty and construction time for other processes.
Drawings
FIG. 1 is a flow chart of a deep learning positioning optimization method for can body processing, which is provided by an embodiment of the application;
FIG. 2 is a schematic image segmentation diagram of a deep learning positioning optimization method for can body processing according to an embodiment of the present application;
FIG. 3 is a schematic diagram of a large-scale filtering of a deep learning positioning optimization method for can body processing according to an embodiment of the present application;
FIG. 4 is a schematic diagram of the overall network structure of a deep learning positioning optimization method for can body processing according to an embodiment of the present application;
fig. 5 is a schematic diagram of merging data of a deep learning positioning optimization method for can body processing according to an embodiment of the present application.
Detailed Description
The application is described in further detail below with reference to the attached drawings and detailed description:
the application discloses a flow chart of a deep learning positioning optimization method for can body processing, which is shown in fig. 1.
And S1, setting a filtering interval.
FIG. 2 is a schematic view showing image segmentation of a deep learning positioning optimization method for can body processing according to the present application
In this step, two types of data, ultrasonic monitoring data and thermographic analysis data, respectively, need to be characterized. The method is characterized in that after the heating process of the tank body and before the positioning of the gold stamping process, a thermal imager is used for acquiring a temperature distribution image of the surface of the tank body, and an ultrasonic detection device is used for acquiring acoustic resistance characteristic data. The collection and characterization of the data in the application are divided into two steps, wherein range filtering data collection is firstly carried out, namely, large-range sampling is carried out according to a filtering formula each time, and region overlapping is not generated among the collected pictures.
Wherein the filtering formula is expressed as follows:
wherein D is the circumference of the round body of the tank body,and d is the sampling width of each time set for sampling in a large range. Wherein the regions of each large-scale sample do not overlap. When the rate of factory pipeline setting is higher, thenSetting to a smaller value, and further reducing the number of times of detection of the whole large-area. When the rate set by the factory pipeline is smaller, the +.>Is a larger value, thereby reducing the first filtering range and improving the detection quality.
The first data filtering selection of ultrasonic monitoring and thermal image acquisition can be completed by using the above formula 1, as shown in fig. 2, which is a difference chart of the thermal image analysis collected by the filtering formula in the present application, whereinTaking the parameters of 9, namely dividing the original tank body into 9 areas for shooting, thereby achieving one-time performance9 pieces of image data can be acquired, corresponding to different surfaces of the tank body respectively.
In addition, the application adopts an ultrasonic detection device to acquire acoustic resistance data, and the acoustic resistance data comprises five characteristics of reflection amplitude, echo time, acoustic impedance, acoustic energy loss and acoustic velocity, and the specific characterization mode is performed according to a filtering formula.
Step S2: and (5) data image acquisition setting.
The application discloses a large-scale filtering schematic diagram of a deep learning positioning optimization method for can body processing, which is provided by the application as shown in fig. 3.
Of note is the data image size of the input algorithm model. As shown in FIG. 2, the width of the middle welding seam is 3-4 mm, if the data range formed by the ultrasonic or thermal imaging instrument is too small, the overall calculation pressure may be too large, and the processing speed of the tank body can not meet the requirements of an industrial rapid assembly line. If the parameter is set to a larger value, the sampling accuracy may be lower, and the noise data is too much, i.e. the difficulty of subsequent network analysis is increased. Therefore, in this application, a balanced filtering method is proposed, that is, two filtering methods are adopted, both are collected according to a sampling formula, wherein the first one adopts a large-range filtering method, and then a small-space extraction work is performed on a region with a high possibility of a filtered welding seam
Wherein, the sampling formula is expressed as follows:
wherein T is expressed as the weld size, k is expressed as a hyper-parameter, and S is the size input into the algorithm model. Taking k=1 as a specific middle critical point, when the k value is smaller than 1, the filtering rule is large-range at the moment, and when the k value is larger than 1, the filtering rule is small-spacing extraction work at the moment. The large-scale filtering k takes 2, and the small-scale extraction k takes 0.3 as shown in fig. 3.
Step S3: data fusion settings
The application discloses a network overall structure schematic diagram of a deep learning positioning optimization method for can body processing, which is provided by the application as shown in fig. 4.
The application discloses a combined data schematic diagram of a deep learning positioning optimization method for can body processing, which is provided by the application as shown in fig. 5.
In step S1, a method of image acquisition is set, and in step S2, a size of how to process image data with an algorithm model in an image is set. In this step, the data application method needs to be set. The application provides a multi-layer data fusion model, which fuses the thermal image analysis data collected at this time and the ultrasonic monitoring data at intervals, wherein the influence of the two data is based on the actual detection quality of the two data and the network correctness during independent training. When training alone, the thermal image analysis data is analyzed by adopting a common convolution network, the target mapping value is whether a welding line exists in the region, and when training alone, the ultrasonic monitoring data, namely, the five data including reflection amplitude, echo time, acoustic impedance, acoustic energy loss and sound velocity, is mapped by adopting a BP network. When training is performed to a certain degree, if the accuracy of classification of the two data cannot be improved, stopping training, and outputting the accuracy values of classification of the two data. The thermographic analysis data and the ultrasonically detected data are also validated to see the error rate of their acquisition. Fig. 4 is a schematic diagram of data fusion between the two applications. Firstly, dividing all image data and ultrasonic data according to a large-range filtering rule of a formula 2, carrying out each group of ranges according to a sampling formula, and then carrying out merging analysis on thermal image analysis data and ultrasonic monitoring data, wherein the ultrasonic monitoring data need to be processed before merging, namely, the ultrasonic data need to be amplified, and the ultrasonic data are converted into two-dimensional data from one-dimensional data vectors, wherein the application is directly based on the size of an image Featuremap of the thermal image data after large-range filtering, and the ultrasonic data are directly filled in according to an amplification formula.
Wherein the amplification formula is expressed as follows:
x=Random(x 1 ,x 2 ,x 3 ,x 4 ,x 5 ) ⑶
wherein x is 1 、x 2 、x 3 、x 4 、x 5 Random indicates a Random function, i.e. randomly extracting a value from within the selected element, respectively the reflection amplitude, echo time, acoustic impedance, acoustic energy loss and speed of sound. x indicates the value of the pixel. The expansion of the one-dimensional ultrasonic data can be completed through the formula, so that the expansion is consistent with the FeatureMap of thermal image analysis.
When the expansion of the ultrasonic data is completed by using the formula 3, the ultrasonic data needs to be combined as follows, and the application proposes a novel combination mode, wherein the expanded generated wave data and thermal image analysis data are combined to enhance the expression capability of the data characteristics. The merging mode is interval weight merging according to a merging formula, and a sandwich-like structure is formed.
Wherein the combining formula is expressed as follows:
wherein n represents the number of ultrasonic data to be inserted, ero R ,Erp C And analyzing whether the independent training network training is the error rate of the welding seam or not for the thermal image respectively, and detecting whether the independent training network training is the error rate of the welding seam or not by ultrasonic waves. arc (arc) C In order to detect ultrasonic wave of the tank body, the accuracy of ultrasonic wave is arc R The accuracy of thermal image analysis is realized when thermal image data acquisition is carried out on the tank body.
The competition relationship and the competition result can be shown by the formula 4 during ultrasonic wave and thermal image analysis. When the value of n is greater than 1, it indicates that the ultrasonic detection value should be weighted more than the thermographic analysis, and therefore its number needs to be greater than the FeatureMap of the thermographic analysis at the time of interval interpolation. And when n is less than 1, it indicates that the influence coefficient of ultrasonic detection should be reduced. As shown in fig. 5, n is 2 at this time.
Step S4: network architecture settings.
In step S3, the setting of data fusion is determined, and specific application is required to the flow of the network architecture in this step. The data output in the step S3 is in the form of fused FeatureMap, wherein the fused FeatureMap comprises thermal image analysis data and ultrasonic detection data, convolution operation is carried out on the fused FeatureMap in the step, the convolution kernel is 3*3, the interval is 1, the filling item is 1, when the data is input into the fused FeatureMap after a large-scale filtering rule, the convolution operation is carried out, whether the fused FeatureMap contains a welding line is judged, and if the fused FeatureMap does not contain the welding line, the other area separated by network operation input is directly interrupted. If a weld is included, the next stage is entered. And (3) performing small-interval extraction according to a formula 2, namely performing small-range extraction on the data which are subjected to large-range filtering and segmentation and are judged to have welding seams, continuously performing the step S3 and convolution classification on the data which are subjected to small-range extraction, judging whether the welding seams exist in a small-range area, recording coordinate values at the moment if the welding seams exist, continuously performing small-range detection, and recording all coordinate values which are judged to be the welding seams until the small-range detection judges that no welding seam termination network exists.
Step S5: and outputting weld coordinate values.
In the step S4, detection and storage of all weld coordinate values are completed, and in this step, the weld coordinate values are processed and output, wherein the coordinate processing formula is as follows:
wherein j represents the final output weld position, i is the coordinate value when the weld exists in all small ranges, and p is the number value when the weld exists in all small ranges.
And the output of the final weld coordinate value can be finished through the formula 5, and the fault tolerance and the accuracy are improved.
Step S6: algorithm deployment and application.
In the steps, positioning of the welding line of the tank body is completed, and deployment and application are carried out in the steps. The algorithm model is deployed into a production line, a specific coordinate value of a welding line of the tank body is output, the position of the tank body at the moment is judged according to the fed back coordinate value, the angle from the position needing gold stamping is judged, and the tank body is rotated to directly gold stamping.
The above description is only of the preferred embodiment of the present application, and is not intended to limit the present application in any other way, but is intended to cover any modifications or equivalent variations according to the technical spirit of the present application, which fall within the scope of the present application as defined by the appended claims.
Claims (5)
1. A deep learning positioning optimization method for can body processing is characterized by comprising the following steps: the method comprises the following steps:
1) Setting a filtering interval;
in the step, two types of data need to be characterized, namely ultrasonic monitoring data and thermal image analysis data;
the method comprises the steps of acquiring a temperature distribution image of the surface of a tank body by using a thermal imager after the heating process of the tank body and before the positioning of a gold stamping process, and acquiring acoustic resistance characteristic data by using an ultrasonic detection device, wherein the collection and characterization of the data are divided into two steps, wherein range filtering data collection is firstly carried out, namely, sampling is carried out in a larger range according to a filtering formula each time, and region overlapping is not generated among acquired pictures;
2) Data image acquisition setting;
the width of the welding line in the middle of the tank body is 3-4 mm, a balance filtering rule is adopted, namely, two filtering methods are adopted, the two filtering methods are collected according to a sampling formula, wherein the first one adopts a large-range filtering rule, and then small-space extraction work is carried out on a region with higher possibility of the welding line after filtering;
3) Setting data fusion;
the multi-layer data fusion model is used for carrying out interval weight fusion on the thermal image analysis data collected at this time and the ultrasonic monitoring data according to a merging formula, wherein the influence of the two data is according to the actual detection quality of the two data and the network correctness during independent training, and the ultrasonic data is directly filled in according to an amplification formula;
4) Setting a network architecture;
the method comprises the steps that the flow of a network architecture is specifically related to, the data output in the step 3) are in the form of fused Feature Map, wherein the fused Feature Map comprises thermal image analysis data and ultrasonic detection data, convolution operation is carried out on the thermal image analysis data and the ultrasonic detection data, and when the data subjected to large-range filtering rule are input into the data for convolution operation, whether welding seams are contained in the data is judged;
5) Outputting a welding line coordinate value;
the detection and storage of all weld coordinate values are completed in the step 4), and the weld coordinate values are processed and output according to a coordinate processing formula in the step;
6) Algorithm deployment and application;
in the above steps, positioning of the welding seam of the can body is completed, deployment and application are carried out in the steps, an algorithm model is deployed into a production line, specific coordinate values of the welding seam of the can body are output, the position of the can body at the moment is judged according to the fed back coordinate values, the angle from the position needing gold stamping is judged, and the can body is rotated to directly gold stamping.
2. The deep learning positioning optimization method for can body processing according to claim 1, wherein the method comprises the following steps:
the filtering formula in the step 1) is expressed as follows:
wherein the filtering formula is expressed as follows:
wherein D is the circumference of the round body of the tank body,for the super parameter, d is the sampling width of each time set for sampling in a large range, wherein the areas of each large range do not overlap, and when the rate set by the factory pipeline is higher, the ratio of the sampling width to the sampling width is +.>Setting to a smaller value, so that the number of times of detection of the whole large area is reduced, and when the rate of setting the factory pipeline is smaller, the rate is increased +.>Is a larger value, thereby reducing the first filtering range and improving the detection quality.
3. The deep learning positioning optimization method for can body processing according to claim 1, wherein the method comprises the following steps:
the sampling formula in the step 2) is expressed as follows:
wherein, the sampling formula is expressed as follows:
wherein T is expressed as the weld size, k is expressed as a super parameter, s is the size input into the algorithm model, k=1 is taken as a specific intermediate critical point, when the k value is smaller than 1, the large-range filtering rule is indicated at the moment, and when the k value is larger than 1, the small-space extraction work is indicated at the moment.
4. The deep learning positioning optimization method for can body processing according to claim 1, wherein the method comprises the following steps:
the amplification formula in step 3) is expressed as:
wherein the amplification formula is expressed as follows:
x=Random(x 1 、x 2 、x 3 、x 4 、x 5 )
wherein x is 1 、x 2 、x 3 、x 4 、x 5 Random indicates a Random function, i.e. randomly extracting a value from the selected element, x indicates the image, respectively the reflection amplitude, echo time, acoustic impedance, acoustic energy loss and acoustic velocityThe numerical value of the pixel point can be used for completing the expansion of one-dimensional ultrasonic data through the formula, so that the expansion is consistent with a Feature Map of thermal image analysis;
the combined formula in the step 3) is expressed as:
wherein the combining formula is expressed as follows:
wherein n represents the number of ultrasonic data to be inserted, ero R ,Erp c Analyzing whether the individual training network training is the error rate of the welding seam or not for the thermal image respectively, and ultrasonically detecting whether the individual training network training is the error rate of the welding seam or not, and arc c In order to detect ultrasonic wave of the tank body, the accuracy of ultrasonic wave is arc R The accuracy of thermal image analysis is realized when thermal image data acquisition is carried out on the tank body.
5. The deep learning positioning optimization method for can body processing according to claim 1, wherein the method comprises the following steps:
the coordinate processing formula in the step 5) is expressed as follows:
in the step 4), detection and storage of all weld coordinate values are completed, and in the step, the weld coordinate values are processed and output, wherein a coordinate processing formula is as follows:
wherein j represents the final output weld position, i is the coordinate value when the weld exists in all small ranges, and p is the number value when the weld exists in all small ranges.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310951295.XA CN117218188A (en) | 2023-07-31 | 2023-07-31 | Deep learning positioning optimization method for can body processing |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310951295.XA CN117218188A (en) | 2023-07-31 | 2023-07-31 | Deep learning positioning optimization method for can body processing |
Publications (1)
Publication Number | Publication Date |
---|---|
CN117218188A true CN117218188A (en) | 2023-12-12 |
Family
ID=89045013
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202310951295.XA Pending CN117218188A (en) | 2023-07-31 | 2023-07-31 | Deep learning positioning optimization method for can body processing |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN117218188A (en) |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109175608A (en) * | 2018-09-30 | 2019-01-11 | 华南理工大学 | Weld bead feature points position On-line Measuring Method and seam track automatic measurement system |
CN111299763A (en) * | 2020-02-28 | 2020-06-19 | 华南理工大学 | Anti-noise-interference laser visual welding seam automatic tracking method and system |
CN114453708A (en) * | 2022-02-22 | 2022-05-10 | 中国计量大学 | Aluminum alloy welding defect online monitoring method |
CN115239632A (en) * | 2022-06-23 | 2022-10-25 | 西南交通大学 | Steel rail surface damage detection method fusing inspection image and ultrasonic image |
CN116503417A (en) * | 2023-06-29 | 2023-07-28 | 武汉纺织大学 | Automatic recognition, positioning and size calculation method for ultra-long weld joint and typical defect |
-
2023
- 2023-07-31 CN CN202310951295.XA patent/CN117218188A/en active Pending
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109175608A (en) * | 2018-09-30 | 2019-01-11 | 华南理工大学 | Weld bead feature points position On-line Measuring Method and seam track automatic measurement system |
CN111299763A (en) * | 2020-02-28 | 2020-06-19 | 华南理工大学 | Anti-noise-interference laser visual welding seam automatic tracking method and system |
CN114453708A (en) * | 2022-02-22 | 2022-05-10 | 中国计量大学 | Aluminum alloy welding defect online monitoring method |
CN115239632A (en) * | 2022-06-23 | 2022-10-25 | 西南交通大学 | Steel rail surface damage detection method fusing inspection image and ultrasonic image |
CN116503417A (en) * | 2023-06-29 | 2023-07-28 | 武汉纺织大学 | Automatic recognition, positioning and size calculation method for ultra-long weld joint and typical defect |
Non-Patent Citations (2)
Title |
---|
王增勇 等: ""长罐体环焊缝缺陷定量检测技术"", 《焊接质量控制与管理》, 31 December 2014 (2014-12-31), pages 59 - 62 * |
程柏松 等: ""图文预变形补偿技术在覆膜铁制罐中的应用"", 《学术天地》, 31 August 2015 (2015-08-31), pages 38 - 39 * |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Zhang et al. | Weld image deep learning-based on-line defects detection using convolutional neural networks for Al alloy in robotic arc welding | |
CN110047073B (en) | X-ray weld image defect grading method and system | |
CN109886298B (en) | Weld quality detection method based on convolutional neural network | |
CN112232400A (en) | Stainless steel weld ultrasonic defect detection method based on depth feature fusion | |
CN105389814B (en) | A kind of bubble detecting method for air-tight test | |
CN101501487B (en) | Non-destructive testing by ultrasound of foundry products | |
CN105938563A (en) | Weld surface defect identification method based on image texture | |
CN102162577B (en) | Pipeline defect surface integrity detection device and detection method | |
CN105913415A (en) | Image sub-pixel edge extraction method having extensive adaptability | |
CN110441329B (en) | Laser welding defect identification method, device and equipment based on deep learning | |
CN112946081B (en) | Ultrasonic imaging method based on intelligent extraction and fusion of defect multiple features | |
Dubey | Fruit defect detection based on speeded up robust feature technique | |
Bastomi et al. | Object detection and distance estimation tool for blind people using convolutional methods with stereovision | |
Ekambaram et al. | Identification of defects in casting products by using a convolutional neural network | |
CN117218188A (en) | Deep learning positioning optimization method for can body processing | |
CN116721080A (en) | Small-diameter pipe welding defect detection method based on shape and position sensing and balanced sampling | |
CN106556606A (en) | Line Quality Control method based on the total member of big machinery product of machine vision | |
CN106022365A (en) | Surface defect depth estimating method based on data fusion and RBF neural network | |
CN116452936B (en) | Rotation target detection method integrating optics and SAR image multi-mode information | |
CN115994893B (en) | Sealing element weak and small leakage target detection method based on low-resolution infrared image | |
CN115115578B (en) | Defect detection method and system in additive manufacturing process | |
CN114418929A (en) | Weld defect identification method based on consistency multi-scale metric learning | |
CN111882535B (en) | Resistance welding shear strength identification method based on improved Unet network | |
Naddaf-Sh et al. | Next-generation of weld quality assessment using deep learning and digital radiography | |
CN111753876B (en) | Product quality detection method based on deep neural network |
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 |