Article ID | Journal | Published Year | Pages | File Type |
---|---|---|---|---|
9952354 | Computers & Geosciences | 2018 | 21 Pages |
Abstract
Previous efforts on analyzing the sorting level of rock particles rely on particle segmentation, which is time-consuming and inaccurate due to lighting and intra-particle statistical variations. With high-level features learned from a deep neural network, we directly conduct the classification of sorting in rocks. Our approach avoids the need for laborious segmentation and is entirely automatic. We use an off-the-shelf convolutional neural network (CNN) model that has been pre-trained on a large scale image dataset to extract feature representations for our rock images. Then we trained a support vector machine (SVM) classifier with the feature representations as input. The experiments show that the off-the-shelf CNN features lead to significantly improved results for the classification compared with handcrafted features and low-level K-means features.
Keywords
Related Topics
Physical Sciences and Engineering
Computer Science
Computer Science Applications
Authors
Lei Shu, Gordon R. Osinski, Kenneth McIsaac, Dong Wang,