Introduction to Deep Learning

The whole industry has been very excited over the 2 key terms “deep learning” & “artificial intelligence” and there have been many new applications the industry is using this technology for gaming, security & surveillance.

Some of the applications include face recognition, number plate reading, behaviour characterization, etc.  Deep learning has been used for the industrial applications for reading OCR & defects inspection. It has been proven that deep learning will be able to recognize a defect even if the production parts have some different variation on the threshold or machining marks.

Deep learning has solved problems with image classification, image reconstruction, object detection, and more. With a task like image processing, thousands of images are uploaded to a data set. The whole idea of deep learning is to be able to train sufficient sample size of good parts, and “marking” the defects area on the NG parts. During the production, new group of NG parts can be marked & added to the existing database to improve the inspection capability.

Advantages of Deep Learning

1. Self-Learning Capabilities

The layers in deep learning neural networks allow models to become smarter in learning complex features and perform repetitive tasks within a short period of time as compared to the time would be taken by a human inspector. This is due to deep learning algorithms have the ability of learning from its own errors and determine the accuracy of output from time to time.

2. Computing Power and Data

One of the drawbacks of machine vision’s algorithm is the limitation of analysing unstructured data, and this is where deep learning becomes useful. Deep learning algorithms have the ability to make sense of data formats likes text, images and voice and help business to obtain insights and identify every defect outside of the set tolerance, with no special expertise required.

3. Generation Automation

Deep learning system learns as a human would, but their learning curve is beyond human capabilities to analysing every part or product with defects and providing a solution. In addition to this, deep learning algorithms can generate and perform complex tasks without additional human intervention. For businesses, this means deep learning supports a faster and reliable high-speed application and improving throughput.