Home / Articles / Neural network software AXALIT

Neural network software AXALIT

Категории
Специализированный программное обеспечение AXALIT с большим набором методик по анализу изображений с применением нейронных сетей искусственного интеллекта.
15 June 2018, 12:19

Combining proven technology with the latest developments, "AXALIT Soft" out of the image analysis of micro - and macrostructure of various objects to a new level. 2018 software "AXALIT" works with the application of artificial intelligence. More precisely, the fundamental component of artificial intelligence - neural networks. Today we will tell you what it is and how it works in the software "AXALIT". Go!

Artificial neural network

a Neural network is a mathematical model operating on the principle of networks of nerve cells in the animal body. The structure of neural networks came into the world of programming straight from biology. Due to this structure, the machine acquires the ability to analyze and even memorize different information. Neural network is also able not only to analyze incoming information, but to reproduce it from my memory. In other words, a neural network is a machine interpretation of the human brain, which contains millions of neurons that transmit information in the form of electrical impulses.

the

How it works?

the Main advantage of neural networks over conventional computation algorithms, which were previously used in "AXALIT" is their learning capability. Now when you search for objects, the program is not based on mathematical algorithms, and the experience of learning.

the learning Process FOR neural networks is as follows:

the Program gives your answer; the

  1. Signals from the received data it places into the neural network;
  2. If you received a response not a reality, we need to point out the mistake;
  3. During subsequent processing of the data network will take into account the "experience" and the probability of obtaining a correct answer will be greater than in previous times.

Simple things

Imagine a new neural network with a completely empty memory, which we will train to recognize Arabic numerals.

Upload a picture of the printed digits in the program "Axalit Soft". The program quickly recognized and allocated objects, but does not know what it is (figure 1).

Program

figure 1. Program "AXALIT" allocated numbers and classified them as "Unknown".

Create the database (figure 2) according to the task. In this case, the name of Arabic numerals.

Database

figure 2. Database "Arabic numerals" and their symbology.

Teach the program as a young child to identify the numbers, roughly speaking, by their contour, without going into the complex mechanism of neural connections (figure 3).

neural network Training in the

figure 3. Training neural network in the program "AXALIT".

to Learn the information ON it turns out not at once. But periodic error correction, very soon (in this case we are talking about minutes) the program will start to consistently distinguish between all digits (figure 4).

Neural network "Axalit Soft" trained to distinguish between the Arabic numerals

figure 4. Neural network "AXALIT" trained to distinguish between the Arabic numerals.

To test a full network health, load the image with the digits written by hand (figure 5).

Handwritten Arabic numerals

figure 5. Handwritten Arabic numerals.

To the human eye the difference between printed and written numerals insignificant. But for classical algorithms of object recognition small changes in morphology, such as smoothness of writing of five figures, horizontal line at the group of seven (in the printed version of this features are not present), open the writing the number zero, etc. - play a significant role. However, the program has successfully coped with the task and correctly classified all the figures (figure 6).

Program

figure 6. Program "AXALIT" successfully passed the training and had correctly identified all the numbers.

Level up

let's Complicate the task and teach the program to recognize objects of different shape, color and size. In this case stationery set – button-carnations, paper clips, clips for paper and coins (figure 7).

Stationery set and coin

figure 7. Stationery set and coins.

For a given set database "Arabic numerals" is not suitable. Create a new database applicable to the image (figure 8).

Database for recognition stationery

figure 8. Database to recognize the stationery.

Highlighting the new facilities, the program klassificeret them as unknown (figure 9). Train the neural network to the new "knowledge".

Not learning the software "Axalit Soft"

figure 9. Not learning the software "AXALIT".

After a short learning program with ease features all the objects in the image from each other (figure 10).

Learning to recognize objects IN different shape, color and size

figure 10. Trained AT the detects objects of different shape, color and size.

 

Conclusion

Neural network software AXALIT perfectly analyze and remember incoming information, learn very fast and in many ways superior to the classical algorithms of image analysis.

the Finished product

You ask: how, for example, the technician-metallography useful software that is able to distinguish between "numbers" and "staples"?

the fact that the software "AXALIT" created for solving diverse tasks in image analysis of micro - and macrostructures of objects. Implementation of neural networks has accelerated the process and expanded the range of tasks required by end user. These simple and familiar examples we have dealt only with the principle of the updates. A research assistant metallograph can use the program, for example, for analysis of nonmetallic inclusions in steel now. See example works in the video below!