Now, the widely used image inspection methods are dominated by pattern matching. In pattern matching, it is impossible to make fuzzy judgments that humans can make, and can only be judged by comparing similar parts with prepared samples, which has certain limitations.
The characteristic of our company's AI deep learning is that it can make fuzzy judgments that are close to humans.

Rule Library
(Pattern matching)

It is necessary to read a considerable number of templates and examine the patterns of objects. (Hundreds to thousands of copies)

Not good at scars, spots, and other areas with unspecified shapes

AI Deep Learning

Sourcehttps://www.pnas.org/doi/10.1073/pnas.1821594116:

Based on the model (teacher) image, the program library (machine learning function) constructs the model by changing parameters on its own. The number of image copies used for pre learning is controlled low. (Dozens to hundreds of copies)

Customization (adjusting threshold) can achieve accuracy similar to that of ordinary people!