Open-Pit Mine Area Mapping With Gaofen-2 Satellite Images Using U-Net+

Abstract

Obtaining information on the surface coverage of open-pit mining areas (OPMAs) is of great significance to ecological governance and restoration. The current methods to map the OP-MAs face problems such as low mapping accuracy due to complex landscapes. In this article, we propose a hybrid open-pit mining mapping (OPMM) framework with Gaofen-2 (GF-2) high-spatial resolution satellite images (HSRSIs), using an improved U-Net neural network (U-Net+). By concatenating the previous layers with each subsequent layer to ensure that there is a maximum of feature maps of each layer in the network, the U-Net+ can reduce the loss of feature information and make the extraction capability of the network more powerful. Two independent OPMAs were selected as the study area for the OPMM. By taking advantage of GF-2 HSRSIs, a total of 111 open-pit mine sites (OPMSs) were mapped and each OPMS boundary was validated by field surveys. Then, these OPMSs were used as input to assess the accuracy of the OPMM results obtained by the U-Net+. By comparing our results with those provided by five state-of-the-art deep learning algorithms: Fully Convolutional Network (FCN), SegNet, U-Net, Residual U-Net (ResU-Net), and U-Net++, we conclude that the proposed framework outperformed these methods by more than 0.02% in Overall Accuracy, 0.06% in Kappa Coefficient, 0.03% in Mean Intersection over union, 8.36% in producer accuracy and 4.44% in user accuracy. Therefore, the proposed framework thus exhibits very promising applicability in the ecological restoration and governance of OPMAs.

Employing Remote Sensing, Data Communication Networks, AI, and Optimization Methodologies in Seismology

“…The role of satellite imaging has been also extended using a mapping of two independent hybrid open-pit mining along with Gaofen-2 [25]. That work planned the confronting issues to the mapping of hybrid open-pit mining areas because the complex types of landscapes led to low accuracy.…”

Open-Pit Mining Area Extraction from High-Resolution Remote Sensing Images Based on EMANet and FC-CRF

“…Researchers have used convolutional neural networks to analyze high-resolution remote sensing data and obtain land cover information from open-pit mines [13]. For instance, Chen et al [14] proposed an improved deep learning network, U-Net+, with multilayer feature associations. They aimed to obtain the surface coverage of the open-pit mining area, and the experimental results indicated that the proposed framework outperforms the original U-Net by 0.02% overall.…”

Identification of open-pit mines and surrounding vegetation on high-resolution satellite images based on improved bilateral segmentation network semantic segmentation model

“…Through the classification of the ground features in the mining area, the damage to the ecological environment in the mining area can be effectively monitored dynamically 7 , 8 . However, due to the various sizes, complex quantities, and scattered distribution of the types of ground features in the mining area, the traditional manual sampling method is inadequate for accurately classifying the types of ground features in the mining area 9 . Exploring high-performance feature classification methods based on remote sensing data is gradually becoming an essential tool for ecological monitoring in mining areas 10 …”

“…In recent years, traditional machine learning algorithms, such as decision tree, 19 support vector machines, 8 and deep belief networks, 20 have been used to extract information of typical ground features, such as mining areas. However, due to the irregular spatial terrain distribution in open-pit mining areas, 9 the accuracy of the terrain classification model based on traditional machine learning is low. Because of its strong feature extraction and self-learning ability, deep-learning model is more suitable for the analysis of complex surface mining areas 21 .…”

“…7,8 However, due to the various sizes, complex quantities, and scattered distribution of the types of ground features in the mining area, the traditional manual sampling method is inadequate for accurately classifying the types of ground features in the mining area. 9 Exploring high-performance feature classification methods based on remote sensing data is gradually becoming an essential tool for ecological monitoring in mining areas. [10][11][12][13][14] The traditional methods for classifying features from remote sensing data are the pixel-based or object-oriented classification methods, which focus on detecting the distribution of features using spectral mixture analysis techniques.…”