Looking for breakthrough ideas for innovation challenges? Try Patsnap Eureka!

Welding spot quality detection method based on a deep learning target detection algorithm and an implementation system thereof

A target detection algorithm and solder joint quality technology, applied in the field of image processing, can solve problems such as increased calculation amount and time consumption, complex system model, influence, etc., to achieve better expression of image information, increase positioning accuracy, and increase robustness. Effect

Inactive Publication Date: 2019-04-05
SHANDONG UNIV
View PDF4 Cites 29 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

However, this patent has the following defects or deficiencies: 1. The entire target image and all key areas will pass through an independent convolutional neural network, that is, the parameters of multiple network models need to be adjusted, the system model is complex, and the amount of calculation and time consumption is large Add; 2. The patent uses a sliding window to cut the original image to obtain the key area, which has a certain randomness and cannot guarantee that it will completely match the target IC pin solder joint area, and the result will be affected due to inaccurate positioning

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • Welding spot quality detection method based on a deep learning target detection algorithm and an implementation system thereof
  • Welding spot quality detection method based on a deep learning target detection algorithm and an implementation system thereof
  • Welding spot quality detection method based on a deep learning target detection algorithm and an implementation system thereof

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0078] A solder joint quality detection method based on deep learning target detection algorithm, such as figure 1 shown, including the following steps:

[0079] (1) From the Hisense optical module production line, collect images of the solder joints of the device ROSA or TOSA welded to the circuit board. The data set has a total of 4598 pictures, and all the pictures are marked with data, and each solder joint is framed with a rectangular frame, and the judgment is made manually Finally, record the quality of solder joints and the position of the rectangular frame. Normalize the size of all pictures so that the size of all pictures becomes (600×800), and divide the marked data set into 4098 training sets and 500 test sets. The 4098 marked pictures in the training set are subjected to data enhancement, that is, the pictures are rotated at random angles, scaled, flipped and other operations to achieve the purpose of enhancing the data type.

[0080] (2) feature extraction to ...

Embodiment 2

[0085] According to a kind of solder joint quality detection method based on deep learning target detection algorithm described in embodiment 1, its difference is:

[0086] In the step (1), the data is collected and pre-processed, including steps as follows:

[0087] A. Use a microscope with a light source to take pictures of devices with a resolution of 600-1200 including solder joints;

[0088] B. For each solder joint in each picture, mark its position with a box of appropriate size that completely covers the solder joint. The position of each solder joint is represented by four parameters, namely (x, y, w, h) ;Take the lower left corner of the picture as the origin, the left border of the picture is the x-axis, and the lower border of the picture is the y-axis to establish a coordinate system, (x, y) indicates the coordinate position of the upper left corner of the box in the coordinate system, and w indicates the box The width of h represents the length of the box;

[0...

Embodiment 3

[0092] According to a kind of solder joint quality detection method based on deep learning target detection algorithm described in embodiment 1, its difference is:

[0093] In step (2), extracting features from the preprocessed picture in step (1), including:

[0094] Image features are extracted through the deep neural network model ResNet50. The deep neural network model ResNet50 has a total of 50 layers, including 49 convolutional layers and 1 fully connected layer. The 49-layer convolutional layer includes 5 convolutional parts, and the fourth convolutional part The size of the output feature map is 38×50; the deep neural network model ResNet50 is not only easy to optimize the network, but also can avoid the problem of performance degradation caused by too deep network layers. The definition of the feature map output by the convolutional layer is shown in formula (I):

[0095]

[0096] In formula (Ⅰ), Indicates the j-th feature map of the l-th layer, Represents the...

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

PUM

No PUM Login to View More

Abstract

The invention relates to a welding spot quality detection method based on a deep learning target detection algorithm and an implementation system thereof, and the method comprises the following steps:(1) collecting data and carrying out the pre-processing: collecting the position and category of each welding spot in a picture, carrying out the marking, and enabling the size of the picture to be unified; (2) extracting features; (3) automatically generating a plurality of suggested areas by using an RPN network, and mapping the suggested areas to the feature map of the last layer of the deep convolutional neural network model; (4) screening and generating a feature map with a fixed size; And (5) sending the welding spots into a boundary regression network and a classifier to obtain positions and types of the welding spots so as to prompt detection personnel whether to reprocess the circuit board or not. According to the invention, automation is completely realized, professional knowledge of welding spot detection is not needed, the problem of unstable working efficiency of workers is fundamentally avoided, a large number of quality inspectors do not need to be employed, a large amount of labor force is liberated, and meanwhile, the production cost is saved for enterprises.

Description

technical field [0001] The invention relates to a solder joint quality detection method based on a deep learning target detection algorithm and an implementation system thereof, belonging to the technical field of image processing. Background technique [0002] In today's era, all kinds of electronic equipment have become the necessities of people's production and life. With the progress of society and the improvement of people's living standards, people have a rapidly increasing demand for the quality and quantity of electronic equipment. In the production of electronic equipment, a large number of component welding work is required. At the same time, heavy solder joint quality inspection work also arises as the times require. So far, various manufacturers either hire a large number of quality inspectors to carry out heavy and repetitive quality inspection work, which greatly wastes human and financial resources, or purchase expensive professional testing equipment that lac...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

Application Information

Patent Timeline
no application Login to View More
Patent Type & Authority Applications(China)
IPC IPC(8): G06T7/00G06N3/04
CPCG06T7/0004G06T2207/20104G06T2207/30168G06T2207/30164G06T2207/10061G06N3/045
Inventor 李响吴雨林周洪超
Owner SHANDONG UNIV
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Patsnap Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Patsnap Eureka Blog
Learn More
PatSnap group products

Browse by: Latest US Patents, China's latest patents, Technical Efficacy Thesaurus, Application Domain, Technology Topic, Popular Technical Reports.

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap|About US| Contact US: help@patsnap.com