AI image processing technology-AI image processing

AI image processing technology
Teru Sound Corporation

About This Product

<Improvement of quality testing by AI image recognition>
The mainstream of the quality inspection work for industrial parts manufacturing is visual and pattern -matching image recognition by skilled people (a method in which several standard patterns are registered in the machine and then determined).

However, due to today's business situation, the visibility of skilled people is becoming more difficult to secure human resources due to population decline, and the aging of employees has limited business inheritance and business expansion.

On the other hand, in pattern matching image recognition, machines cannot be recognized unless the defined definition is clear. For this reason, there is a situation that it has not been effective for automobiles and bicycle parts that require subtle color identification.

Based on the above, modern quality inspection work has to be performed by interweaving visual visuals and image recognition, and it is far from realizing "work efficiency", which is required in recent "work style reform".

To solve this issue, "Mabesu Eye" developed in our image processing project has been developed.

With image processing technology with a neural network image processing engine formed by deep learning, all identification from learning after labeling is automatically performed.

By combining this AI technology with its own algorithm, it has reached the phenomenal recognition performance that detects a slight incident such as foreign substances, colors, scratches, and press conditions that cannot be seen in human eyes.

As a result, it is possible to contribute to the improvement of the yield, as well as the automation of the line and the accuracy of defective detection, which is directly linked to labor saving.

<Formation of neural networks by deep learning>
 The neural network for image identification is as shown in the figure below. Each camera pixel is placed with a node that performs an operation and weight, and it is called a neuron.

Considering the connection of one output neuron, the image can be identified by learning so that the output neuron will be ignited when a handwritten character is input.

<Deep learning>
In deep learning, it automatically calculates the characteristics of images by learning the learning data multi -stage, and forms a neural network for identifying objects.

After the neural network is formed, enter the object of the object in the neural network to automatically infer objects and calculate the specific probability of objects.

Labeling the object will allow you to identify the object.
Maebaku Eye allows you to change the learning hierarchy up to 150 steps.

<Characteristics of Maberaku EYE test algorithm>
Maebaru Eye has three different test algorithms.
Inspection algorithm can be selected according to the application, object, and measurement environment.