Image Segmentation Method on Quartz Particle-Size Detection by Deep Learning Networks

In the beneficiation of quartz sand, hydraulic classification is a primary way to obtain production in various size fractions. It essential for plants measure particle sand during classification, time evaluate efficiency. However, traditional manual-screening method consumes labor and time, while particle-size analyzer expensive. Thus, size-detection quartz-sand proposed this paper, which based on deep learning semantic-segmentation network Fully Convolutional Networks (FCN)-ResNet50. The FCN-ResNet50 segments images, average obtained after converting pixel-particle physical-particle size. Using learning, with sizes −40 + 70 (0.212–0.38 mm), −70 100 (0.15–0.212 −100 140 (0.109–0.15 −140 400 (0.038–0.109 mm) meshes, can be measured directly. results showed that validation accuracy was over 97%, loss value approximately 0.2. Compared UNet-Mobile Deeplab-Xception, error detection 0.01 mm, close manual calibration-software results. This has advantages quick sampling low equipment costs, increasing hydraulic-classification efficiency promoting automation concentrator.