Pr/Nd Component Content Detection Method Based On Color Feature Modeling

China has largest rare earth reserve and production in the world, rare earth elementsbecome indispensable raw materials or additives of modern industrials due to their excellentphysical and chemical properties, and with reputation of "industrial vitamins". In recent years,Thanks to the guide of cascade extraction theory which proposed by Professor Xu Guangxian,solvent extraction separation crafts developed rapidly, and has been widely used in the rareearth extraction enterprises. However, cascade extraction process has many stages and verycomplex mechanism, process control with multivariable, strong coupling, and time-varyinglag, which lead component content detection and control is difficult to achieve online. Thereal-time knowledge and understanding the component content change at all levels and thekey monitoring points of rare earth extraction process is premise to implement extractionprocess parameters adjustment and control, therefore, the establishment of accurate andefficient component content detection method will help ensure long-term continuous andstable operation of the production process and obtain high-quality products.Aiming to the problem what component content detection of rare earth extraction process,this paper taken Pr/Nd extraction process that with ions characteristic color as the researchobject, with the application of machine vision and digital image processing technology, wepropose Pr/Nd component content detection methods based on color characteristics modeling.(1) In-depth rare earth extraction separation site, we choose mixed solutions which withions characteristic color in all levels clarifying tank of Pr/Nd production line as study object,combining the characteristics of rare earth extraction process, we designed a common videoimage capture system of mixed solution, and analysis the influence factors of color featurecontent, conclude the shape of sample cells, light source types, and concentration of solutions,determine the image acquisition conditions.(2) Acquisition images of Pr/Nd mixed solutions, extract color feature component ofimages after preprocessing, analysis with correlation coefficient method, determine H, S andrelative redness component as auxiliary variables to establish the component content detectionmodel.(3) Establish the RBF, wavelet and BP neural network detection model to forecast Pr/Ndcomponent content, BP neural network has higher prediction accuracy by comparing, itsabsolute error of test samples prediction less than1.2%and relative error less than3%,so itcan meet the accuracy of Pr/Nd extraction component content detection.Simulation test results based on the samples which collected at the scene show thatcomponent content detection method based on color feature with good stability and higher detection accuracy, which can provides real-time, reliable information for parametersadjustment of Pr/Nd extraction process, and the method also can be extended to a class of rareearth elements component content detection with ionic character color.