PNIPU: Automatic system for recognizing defects in textile factories

The main articles are:

• Video analytics (terms, applications, technologies)
• Computer vision (machine vision)

2024: Announcement of automatic defect recognition system in textile factories

On April 23, 2024, representatives of PNIPU announced the development of an automatic defect recognition system at textile factories.

As reported, in light industry, more than 60% of marketable products are occupied by textiles. In the production of fabrics, various external defects often occur (holes, uneven staining of fabrics), which are difficult to detect in a timely manner. Because of this, most of the material is subsequently discarded or recycled, which is very costly. You can ensure product quality control by computer vision methods, which process images and read defects in the product from photo and video shooting. But existing prototypes of such solutions do not take into account all possible flaws often found in industry.

PNIPU scientists have optimized the computer vision method to quickly and accurately detect defects in production. The article was published in the collection "AIP Conference Proceedings," 2024. The study was carried out as part of the Priority 2030 strategic academic leadership program.

The textile industry is engaged in the processing of plant, animal, artificial and synthetic fibers into yarns, threads and fabrics. Manufacturers of all countries are constantly striving to expand the range and optimize the quality of their products in order to give them valuable consumer properties. This is achieved by automating processes and introducing modern technologies.

The computer vision system allows you to automatically recognize tissue defects by analyzing their appearance. Such a defect often occurs at various stages of production due to poor-quality raw materials, violations in technological processes and equipment errors. As of April 2024, there are practically no complexes in Russia that ensure quality control of the textile industry. And the use of foreign analogues for continuous search for defects is not always available to manufacturers and requires large material costs. The solution may be more flexible and budget systems that use video stream processing algorithms. They are versatile and can fit any stage of production.

The reject recognition system should read images from sensors equipped with a camera, correct them (remove noise, blur and other interference) and reliably identify places with defects. The underlying algorithm can be developed using a variety of techniques to recognize image boundaries. Thus, the fuzzy logic method is widely used, which, when processing photos and videos in accordance with the database, determines the degree of belonging of elements to a particular value (there is a defect or not, and if so, which one). This means that it is useful for detecting defects in textiles.

But the existing prototype of such an algorithm has its drawbacks. It does not take into account the non-sharp color differences of the image, with the help of which you can determine the gaps (roughness of the canvas), as well as the unevenness of the density of the canvas. Therefore, to expand the spectrum of detectable defects Perm , Polytech scientists optimized it.

Our modified processing method involves two phases: fast and more thorough. Different types of fabric during photo and video shooting have their own brightness and contrast. Therefore, in the first phase, the algorithm finds possible defects using color correction, and in the second phase, it checks the reliability of the defect determination, highlights it with color and transmits the result to the screen to a specialist. The algorithm has been tested on images of four types of fabrics and can detect weaving and coloring defects. shared Andrey Zatonsky, Doctor of Technical Sciences, Head of the Department "Automation of Technological Processes" of the Bereznikovsky branch of PNIPU

source = PNIPU

Polytechnics compared the effectiveness of their method with an existing analogue on the example of fabric with a defect in the irregularity of the canvas. As a result, the marriage was discovered only by the development of Perm scientists. This algorithm divides the image into two main parts (common background and defect, if it exists). If you pass an image through the fuzzy output system without any flaws, then the image at the output will be completely colored with the same color or impurities of other colors will be minimal.

For textures of each type of polytechnic material, an average background distribution was determined, that is, a typical fabric-specific texture-background ratio. For example, for denim, it is 72%, for linen - 67%. It is this parameter that is used to move from the first phase of image analysis to the second to confirm or deny the presence of a defect. If in fast processing the algorithm gives a percentage of the background distribution, far from the average, then there is a defect on the tissue. Then, in the long phase, the pixels in the image are marked in red. After that, the specialist receives a signal about the presence of a defect.