Title: A rock classification system based on embedded platform

Authors: Xiangyuan Zhu; Jie Yang; Weiyang Zhi; Haifeng Lu

Addresses: School of Computer Science and Software, Zhaoqing University, Zhaoqing, Guangdong – 526061, China ' School of Computer Science and Software, Zhaoqing University, Zhaoqing, Guangdong – 526061, China ' School of Computer Science and Software, Zhaoqing University, Zhaoqing, Guangdong – 526061, China ' Information Center, Zhaoqing University, Zhaoqing, Guangdong – 526061, China

Abstract: Rock is a major material for the crust and upper mantle formation of the Earth. In Earth Sciences, rock image classification is an essential and critical task in the geological survey. Due to the scarcity of samples and unaffordability of rock classification systems, an embedded system was built to collect and identify rock images. The Raspberry Pi3B+ was applied as the micro controller unit and the Sony IMX219 image sensor was selected to shoot rock images. The new well-annotated dataset contains seven types of fresh rocks with 7,976 images. Based on the new dataset, a new rock classification model based on the ConvNeXt algorithm was proposed. To ensemble the local and global features of the rock images, a feature fusion strategy named super-image was designed. Compared with the prevalent models including VGG16, ResNet50, MobileNet V3, GoogleNet, and DenseNet121, our enhanced ConvNeXt method achieved the macro-average F1 and accuracy of 99.61% and 99.63%, respectively.

Keywords: rock image classification; embedded platform; ConvNeXt; super-image; dataset enhancement.

DOI: 10.1504/IJES.2023.134115

International Journal of Embedded Systems, 2023 Vol.16 No.1, pp.23 - 35

Received: 30 Aug 2022
Received in revised form: 01 Feb 2023
Accepted: 22 Mar 2023
Published online: 11 Oct 2023 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article