Ore Image Classification Based on Improved CNN. (April 2022)

• Record Type: Journal Article
• Authors: Zhou, Wenyan; Wang, Hao; Wan, Zhibo
• Is Part Of: Computers & electrical engineering. Volume 99(2022)
• Publication Date: 2022-04
• Subjects: Ore classification -- Deep learning -- Convolution neural network -- Transfer learning -- Data augmentation -- Squeeze-and-excitation networks; CNN Convolutional Neural Network -- DA Data Augmentation -- DL DeepLearning -- ReLU Rectified LinearUnit -- TL TransferLearning -- FC Fully Connected layer -- AM MetaheuristicAlgorithm -- SENet Squeeze-and-Excitationnetworks -- CVD Cardiovascular Disease; Computer engineering -- Periodicals; Electrical engineering -- Periodicals; Electrical engineering -- Data processing -- Periodicals; Ordinateurs -- Conception et construction -- Périodiques; Électrotechnique -- Périodiques; Électrotechnique -- Informatique -- Périodiques; Computer engineering; Electrical engineering; Electrical engineering -- Data processing; Periodicals; Electronic journals; 621.302854
• DOI: 10.1016/j.compeleceng.2022.107819
• Languages: English
• ISSNs: 0045-7906

Abstract

Abstract: The identification of ore deposits is an important technical task in mining and excavation. However, conventional techniques are time-consuming and tedious. Therefore, data augmentation and transfer learning were used in this topic to improve the classification accuracy of ore deposit datasets. The model was pre-trained using 957 images of seven different ore types, which were acquired from a Public Kaggle dataset. Five different convolutional neural network (CNN) models were selected for weight acquisition, including AlexNet, VGG16, ResNet50, InceptionV3, and MobileNet. The comparative experiments demonstrated the use of transfer learning to be effective against improving classification performance, such as reaching 94% with the MobileNet model. The classification performance was further improved by training a Squeeze-and-Excitation Networks (SENet) classifier using features calculated with MobileNet, resulting in an accuracy of 96%. These results prove that the proposed technique, which combines a CNN with transfer learning, data augmentation, and SENet, to be an effective new tool for automated ore classification, offering higher accuracy with less training data.