Hello
I have to realize a deep learning model (with tensorflow and open cv) and connect it to a surveillance camera to detect non conform products in a company.
It's my first project and I'm lost, I don't know where to start and where to get a good database (knowing that I've been asked to make a standard model that can be used for any kind of product).
I need your advice to guide me, or if you can recommend tutorials that can help me.
Thanks in advance
The best way is to follow some related videos/lectures and then text literature review via googling your queries.
Some Video/Lectures e.g., : https://www.youtube.com/watch?v=ORrStCArmP4
https://www.youtube.com/watch?v=a9K563V52VQ and https://www.youtube.com/watch?v=5tvmMX8r_OM
If I was given this problem I would take each camera image and divide it into equal squares of 100 rows and 100 columns. Then you record the predominant colour in each square giving you 10,000 colour points derived from your image.
After that you can start scanning products and tell the algorithm which patterns are good and which are bad. You might have to do some adjustment in the processing to make sure the product is correctly centred in the image.
I hope this works for you. I am not an expert in this field.
Richard
https://www.kaggle.com/yoshida146/defect-detection-train-and-predict-by-class
I believe all these kinds of neural networks and deep learning algorithms require the prior knowledge of populations with and without the response variable you want to analyze, that's the problem.
That's why collecting data is so important.
Another type of deep learning model that you could look into are Autoencoders, that only require training and prior knowledge of normal products. When using the model on anomalies or non-coforming products these will be detected as the features extracted of these product will be outside the range or classification of normal features.
Hi Dr Zakia Mahmoudi . See the following link: https://www.mdpi.com/1996-1944/13/24/5755/pdf
See the following link: https://www.semanticscholar.org/paper/Using-Deep-Learning-to-Detect-Defects-in-A-Survey-Yang-Li/c7e6f85aa4fabb6756cdb00efe472ae5730c283c
See also the following link: https://towardsdatascience.com/quality-inspection-in-manufacturing-using-deep-learning-based-computer-vision-daa3f8f74f45
Check https://www.researchgate.net/publication/347821384_Using_Deep_Learning_to_Detect_Defects_in_Manufacturing_A_Comprehensive_Survey_and_Current_Challenges
You can also use this reference
Article A Machine Learning Method for Defect Detection and Visualiza...
I advice you before staring to download some of the recent surveys related to the subject, through that you can select some related studies and the type of datasets that are used.
The idea for products quality monitoring, it’s be through the parallel image segmentations, you need to define the negative and positive patterns and samples for defects detection.
Good luck
- Badges
- Science topic
- Similar topics
- Mathematics