Deep machine vision-based content prediction of rare earth elemental components

Rapid detection of elemental content in the rare earth extraction process is the primary condition affecting the quality of exported products, while the current detection instruments have problems with large detection delays and high maintenance costs. Given that some rare earth elements, such as Pr and Nd ions, have color characteristics, machine vision-based methods can be used to measure the elemental content. Different from the traditional machine vision method, this paper first introduced a convolutional neural network (CNN) to extract the abstract representation of the original image of the Pr/Nd mixed solution, and at the same time, a deep neural network was used to construct a regression model for predicting component content of each element in the Pr/Nd mixed solution. The experiment selected 1 210 images of Pr/Nd mixed solutions as the experimental data, and the data scale was increased nearly 12 times compared with the existing methods. The maximum relative error absolute value between the predicted component content and the real was 2.7738%, which met the accuracy requirement of the distribution change in the elemental content for actual extraction production and had a practical significance.