Main page › Implementations › Industrial machine vision › Shape classification system for diamonds
Intelligent (self-learning) system for detection and classification of diamonds by shape based on the optoelectronic stereoscopic method.
In 2013-2014, Mallenom Systems completed the project for development of mathematical algorithm and software for a pilot industrial sample of automatic sorting machine for natural diamonds by shape. This project was carried out for the order placed by ALROSA, and the general contractor for this order was NPP Burevestnik JSC. Optoelectronic stereoscopic method for control, machine vision technologies and machine learning all have been at the core of this system.
Dr Vladimir Tsarev, the head of the project research division, had the following to say: “Our company developed the method, algorithms and the software for extrapolation of the diamond shape in 3D space and its further classification based on the spatial properties. In addition, we designed an original optoelectronic device for diamond registration. This device allows to compose the diamond projection image using information from 9 cameras when the diamond is free falling. This visual information is utilised as a preliminary data for extrapolation of diamond shape and its further classification across 10 technological groups”:
Number | Group | Subgroup by degree of distortion | Subgroup by nature of surface edges |
1 | Whole crystals: octahedrons, dodecahedrons and rhombic dodecahedrons | Isometric: degree of distortion is between 1.0 and 1.5 | With smooth edges and with slight roundness and relief of edges |
With slight and pronounced roundness and relief of edges | |||
With very pronounced and sharp edge relief | |||
Distorted: degree of distortion is greater than 1.5 | With smooth edges and with slight roundness and relief of edges | ||
With slight and pronounced roundness and relief of edges | |||
With very pronounced and sharp edge relief | |||
2 | Crystal fragments, spinel crystal twinning and their fragments | Isometric: degree of distortion is between 1.0 and 1.5 | With marginal and unpronounced relief of edges |
With pronounced and sharp edge relief | |||
Distorted: degree of distortion is greater than 1.5 | With various degree of edge relief | ||
3 | Flat: whole crystals, crystal fragments, spinel crystal twinning and their fragments | Heavily distorted, flattened | With various degree of edge relief |
The rate of diamond classification with size ranging from 1.5 to 6 mm with this approach can reach 15-20 crystals per second.
Mr Yuri Bakhvalov, the developer of mathematical algorithms and head of research and development, had the following to say:
“We were able to address number of crucial mathematical challenges. First of all, we developed efficient calibration algorithms for stereoscopic system that is the part of final solution. During automatic calibration we evaluate all parameters of optical chain, connect 9 cameras into unified coordinate space. Second, we developed the approach for precise extrapolation of the diamond 3D shape based on its spatial projections. And, most importantly, we developed a method and tools to teach the diamond classification system using various introduced categories and natural diamond samples. Thus, we were able to create an automated solution for complex task of precise object classification”.
Synchronous live gathering of 9 diamond projections
Diamond projection a Diamond 3D model
Mr Alexey Potapov, the leading software engineer, had the following to say:
“Testing that took place at the beginning of 2014 for the pilot industrial sample on the premises of ALROSA has shown high results of the developed system efficiency. The results have shown the classification accuracy of 95-96% for various diamond samples of the total mass of up to 20 thousand carats. Based on these test results, ALROSA decided to purchase developed technology for industrial application in 2014. ALROSA and NPP Burevestnik JSC also have plans for future cooperation with Mallenom Systems to develop the whole range of similar systems for sorting based on the shape, color and quality of the diamond. Implementation of these systems on a large scale would allow to reduce the use of manual labour drastically, and improve productivity and quality of sorting substantially”.