Sorting industrial minerals with pattern recognition

• Autores: Industrial Minerals
• Localización: Industrial Minerals, ISSN 0019-8544, Nº. 558, 2014 (Ejemplar dedicado a: Marzo)
• Idioma: inglés
• Texto completo no disponible (Saber más ...)
• Resumen

  • Pattern recognition can be generally carried out in two different ways; the first being template matching and the second being feature detection. During template matching, a pattern is used to produce items of the same proportions. The template matching procedure is based on the incoming sample image which is compared with templates memorised in the system. If there is a match, the sample is identified. In this configuration, it is necessary to define templates of minerals which must be identified and removed or retained. This procedure requires more processing power, since the number of templates must be sufficient enough for the matching to be effective. The matching process must consider all combinations of all templates with all analysed particles.

    Pattern recognition models have been tried in the sorting of different minerals when both optical properties were used for identification, either separately or together with X-ray attenuation. The models were successfully tested and applied in the sorting of limestone, quartz, apatite, fluorspar, manganese ore, bauxite, barite, iron ore, chrome ore and coal. One of the examples is shown in Tables 1 and 2.

    Multiple sensor sorting provides the potential for continuous separation of different particles that have different size, colour, shape, texture and density. Proper signal processing is extremely important in providing an optimal route for the sensed information where the "obvious cases" are efficiently removed from the image and further processed by pattern matching algorithms. Such algorithms must also be optimised to eliminate lengthy calculation routines and so speed up the whole process. Application of both pattern matching methods can be considered for mineral sorting though this is dependent on the final target. Template matching can be more time consuming though more accurate than feature detection which can be faster but less detailed.