[SOLVED] Machine learning algorithms for identifying electronic components in PCB.

[akerkarprashant]

Printed circuit board - Wikipedia

en.wikipedia.org

Can machine learning algorithms, Computer vision, AI technologies, Image processing identify all the various electronic components in a Printed circuit board?

Input dataset would be images of Different Printed circuit boards housing electronic components, images of all the individual electronic components.

Examples: Resistors, Capacitors, Transistors, Integrated circuits, Light emitting diode, transformer etc

Thanks & Regards,

Prashant S Akerkar


--- Updated May 24, 2021 ---

Thanks.

Attached images.

Thanks & Regards,
Prashant S Akerkar

--- Updated May 24, 2021 ---

Computer vision - Wikipedia

en.wikipedia.org

Thanks & Regards,
Prashant S Akerkar

 

[andre_luis]

Input dataset would be images of Different Printed circuit boards

Surelly not that easy. When working with image classification, prefer to use pictures took from the same camera and from the same perspective once this ensure that at least all devices are on the same scale.

 

[akerkarprashant]

Thank you.

As you mentioned same Camera device taking images of all the electronic components would be useful.

Could this be the Steps?

1 Image classification of a electronic component with n number of images.

Capture all the images with separate folders for each component viz Resistors, Capacitors, Batteries, Transistors, Integrated Circuits, Diode etc.


2 Capture image of Printed Circuit Board and asking the machine to identify a particular component/s in a selected area of the PCB.

The Machine learning (Computer Vision) program compares with all the images fed for the electronic components with the image inputted (selected from PCB board) and should give us the desired output.

Example : Silicon controlled rectifier.

Can this be achieved ?

Thanks & Regards,
Prashant S Akerkar

--- Updated May 25, 2021 ---

Requirements:

Identify all the components in this attached PCB?

Output : The image will be modified and highlighted with all the electronic components identifed with the correct component name.
New image of the PCB as the output. i.e. Image Annotation.

So if it is a Integrated circuit, it should highlight that component area giving us the correct answer as Integrated circuit in the PCB image inputted.

So if it is a Resistor, it should highlight that component area giving us the correct answer as Resistor component in the PCB image inputted.

Is this possible with Machine learning algorithms, Artificial intelligence, Computer vision, Image processing
technologies?

Thanks & Regards,
Prashant S Akerkar

 

[andre_luis]

Capture all the images with separate folders for each component viz Resistors, Capacitors, Batteries, Transistors, Integrated Circuits, Diode etc.

To be honest, I don't believe this would work at all for the purpose you seemgly want to achieve; please note that there are a number of distinct devices that make use of the same package ( e.g: [TO-92] BJT, MOSFET, Linear Regulator, etc... ). In general, Image classification in the PCB scope is mostly used to determine whather components were correctly or not soldered during automated assembly.

 

[akerkarprashant]

Thank you.

There could be major challenges to achieve this.

But i would like to know the approach to achieve this was correct or incorrect?

Image comparison algorithms.

image comparison algorithm - Google Search

I am comparing a component image in the PCB with n number of images in the electronic components database.
If the image matches with any of the existing electronic components images database, that's the electronic component name.

Thanks & Regards,
Prashant S Akerkar

 

[andre_luis]

You will realize that image comparison algorithms are based on obtaining features from samples that consolidate the decision for one or the other pattern being compared. The problem is that the more generic the sample, the greater the presence of the similarity than the difference between the images that could distinguish one from aanother (which in this case would be the letters). As stated, there are a multitude of components with completely different functions and yet using the same type of encapsulation. In the case of the example image that you presented earlier, as they are components of PTH assembly technology, there is still the complicating factor that due to the assembly is made by hand, not pick-and-place machine, not all all of them are positioned "correctly" in the vertical direction, adding an extra degree of difficulty in the image processing. Sorry to say, but I don't think you will be successful in that task