+7 8202 20-21-17
Mon-Fri 9:00AM – 6:00PM (MSK)

Classification of complicated objects

For 20 years, Mallenom Systems specialists have developed and successfully implemented dozens of machine vision systems. In our solutions, we use both our own unique developments in the field of image analysis, as well as proven algorithms from the world leader - Cognex.

In recent years, many new, interesting and at the same time complicated tasks have appeared, for which deterministic algorithms do not give a satisfactory result. Therefore, we began to actively use artificial neural networks that allow us to successfully solve the problem of localization and classification of complicated objects in arbitrary, changing external conditions.

Rough diamond classification by shape and clarity

Rough diamond sorting systems according to shape and clarity were developed for ALROSA, the leader of the world diamond mining industry. In both systems diamonds are sorted in free fall at a rate of up to 20 diamonds per second.

To sort diamonds according to their shape, the stereoscopic system snapshots free-falling diamonds with 9 cameras. To sort the stones by clarity, only 3 cameras are used.  

Both systems were developed by Mallenom Systems. Classification accuracy - up to 99%.

Determination of nanofiltration membranes pore parameters

Nanofiltration membranes are used to filter water, blood plasma and other liquids. The pores in the membranes are holes made as a result of membranes bombarding with neutrons. The size of the holes is measured in nanometers.

In membranes production, it is very important that the pores are distributed evenly across the membrane and large holes (holes) do not occur.

Mallenom Systems has developed the software that allows localizing pores on the membrane image obtained with an electron microscope, calculating their sizes and thus detecting holes of improper size.

Identification of industrial workpieces by surface texture

When manufacturing parts for heavy mechanical engineering, cylindrical workpieces are used, which need to be traced. Since the workpieces are processed at high temperatures and high pressure, the use of marking is not possible.

Mallenom Systems developed the system that identified workpieces by “papillary” lines that occur at butt end of the workpiece during radial forging. The pattern of these lines is unique for each workpiece and allows to unambiguously differentiate one workpiece from another.

Quality assessment of sugar beet in the back of a truck

When accepting raw materials at a sugar factory, the quality of sugar beet is assessed in the back of a truck. The quality of raw materials is determined by contamination, the amount of beet greens and inclusions, the number of chips and the presence of frostbitten tubers.

An automatic classification system has been developed to assess the quality. Cameras that shoot the back of a truck are mounted at the location where the vehicles stop. Then, using neural networks, the contents of the back of a truck are localized and dirt, greens, chips and snow are searched in the images. According to the data obtained, the classification of raw materials by quality categories is performed.

Vehicles counting and classification 

The software is designed for calculation of the intensity and classification of traffic on the video image. Video comes directly from surveillance and special cameras online or can be downloaded from a video file.

Video analysis is performed using an artificial neural network, which provides statistics on the number of vehicles and their categories.

The software works with a wide range of video cameras, with any optical schemes and in the most difficult weather conditions. Accuracy of an assessment reaches 99%. Automatic analysis is carried out in all directions and lanes in the frame. Automatic analysis is carried out in respect to all moving directions and lanes in the frame.


Counting railcars by car couplings

To solve the issues of traceability and commercial accounting of railway transport, it is necessary to divide the train into railcars very precisely. Traditionally, inductive wheelsets sensors are used for this purpose, which are rather expensive.

Mallenom Systems developed the software that allows detecting car couplings on video images. This allows to uniquely determine the number of cars in a passing train.

Since at different locations different optical schemes are used, the background and degree of pollution are changing, and the systems themselves operate in outdoor conditions, artificial neural networks were chosen for the coupling detection algorithms. Accuracy of car coupling detection when analyzing video from one camera is higher than 99.9%.

Determination of coke fractional composition

Coke is used in the production of metal and iron. A very important parameter of coke is the fraction size and its homogeneity. Therefore, it is required to control the fractional composition of coke at the production site.

Since the coke pieces have a complex arbitrary shape, the use of standard algorithms to localize individual pieces and calculate their size does not give a satisfactory result. In Mallenom Systems, artificial neural networks were used to solve this problem, which allowed determining the fractional composition of coke on its image. Neural networks are resistant to changes in ambient light and coke quality.

For additional information, please contact:

Tsareva Ekaterina, Project Manager
e-mail: tsareva@mallenom.ru
phone: +7 (8202) 20-16-39, cell: +7-921-251-62-96