Soil nutrient is one of the most important properties for improving farmland quality and product. Imaging spectrometry has the potential for rapid acquisition and real-time monitoring of soil characteristics. This study aims to explore the preprocessing and modeling methods of hyperspectral images obtained from an unmanned aerial vehicle (UAV) platform for estimating the soil organic matter (SOM) and soil total nitrogen (STN) in farmland. The results showed that: (1) Multiplicative Scattering Correction (MSC) performed better in reducing image scattering noise than Standard Normal Variate (SNV) transformation or spectral derivatives, and it yielded a result with higher correlation and lower signal-to-noise ratio; (2) The proposed feature selection method combining Successive Projections Algorithm (SPA) and Competitive Adaptive Reweighted Sampling algorithm (CARS), could provide selective preference for hyperspectral bands. Exploiting this method, 24 and 22 feature bands were selected for SOM and STN estimation, respectively; (3) The particle swarm optimization (PSO) algorithm was employed to obtain optimized input weights and bias values of the extreme learning machine (ELM) model for more accurate prediction of SOM and STN. The improved PSO-ELM model based on the selected preference bands achieved higher prediction accuracy (R² of 0.73 and RPD of 1.91 for SOM, R² of 0.63, and RPD of 1.53 for STN) than support vector machine (SVM), partial least squares regression (PLSR), and the ELM model. This study provides an important guideline for monitoring soil nutrient for precision agriculture with imaging spectrometry.

Geological disasters are responsible for the loss of human lives and for significant economic and financial damage every year. Considering that these disasters may occur anywhere—both in remote and/or in highly populated areas—and anytime, continuously monitoring areas known to be more prone to geohazards can help to determine preventive or alert actions to safeguard human life, property and businesses. Remote sensing technology—especially satellite-based—can be of help due to its high spatial and temporal coverage. Indeed, data acquired from the most recent satellite missions is considered suitable for a detailed reconstruction of past events but also to continuously monitor sensitive areas on the lookout for potential geohazards. This work aims to apply different techniques and methods for extensive exploitation and analysis of remote sensing data, with special emphasis given to landslide hazard, risk management and disaster prevention. Multi-temporal SAR (Synthetic Aperture Radar) interferometry, SAR tomography, high-resolution image matching and data modelling are used to map out landslides and other geohazards and to also monitor possible hazardous geological activity, addressing different study areas: (i) surface deformation of mountain slopes and glaciers; (ii) land surface displacement; and (iii) subsidence, landslides and ground fissure. Results from both the processing and analysis of a dataset of earth observation (EO) multi-source data support the conclusion that geohazards can be identified, studied and monitored in an effective way using new techniques applied to multi-source EO data. As future work, the aim is threefold: extend this study to sensitive areas located in different countries; monitor structures that have strategic, cultural and/or economical relevance; and resort to artificial intelligence (AI) techniques to be able to analyse the huge amount of data generated by satellite missions and extract useful information in due course.

Spaceborne interferometric synthetic aperture radar (InSAR) techniques are important for landslide detection and monitoring; however, several limitations and uncertainties, such as the unique north–south flying direction and side-look radar observing geometry, currently limit the ability of InSAR to credibly detect landslides, especially those related to high and steep slopes. Here, we conducted experimental and statistical analysis on the feasibility of time-series InSAR monitoring for steep slopes using ascending and descending SAR images. First, the theoretical (T_GNSS), practical (P_GNSS), and terrain (H_terrain) (T-P-H) indices for sensitivity evaluations of the slope displacement monitoring results from time-series InSAR were proposed for slope monitoring. Subsequently, two experimental and statistical studies were conducted for the cases with and without Global Navigation Satellite System (GNSS) monitoring data. Our experimental results of two high and steep open-pit mines showed that the defined theoretical and practical sensitivity indices can quantitatively evaluate the feasibility of ascending and descending InSAR observations in steep-slope deformation monitoring with GNSS data, and the terrain sensitivity index can qualitatively evaluate the feasibility of landslide monitoring results from ascending and descending Sentinel-1 satellite data without GNSS data. We further demonstrate the generalizability of these proposed indices using four landslide cases with both public GNSS and InSAR monitoring data and 119 landslide cases with only InSAR monitoring data. The statistical results indicated that greater indices correlated with higher reliability of the monitoring results, suggesting that these novel indices have wide suitability and applicability. This study can help to improve the practice of slope deformation monitoring using spaceborne InSAR, especially for high and steep slopes.

The spatial characteristics of discontinuity traces play an important role in evaluations of the quality of rock masses. Most researchers have extracted discontinuity traces through the gray attributes of two-dimensional (2D) photo images or the geometric attributes of three-dimensional (3D) point clouds, while few researchers have paid attention to other important attributes of the original 3D point clouds, that is, the color attributes. By analyzing the color changes in a 3D point cloud, discontinuity traces in the smooth areas of a rock surface can be extracted, which cannot be obtained from the geometric attributes of the 3D point cloud. At the same time, a necessary filtering step has been designed to identify redundant shadow traces caused by sunlight on the rocks’ surface, and a multiscale spatial local binary pattern (MS-LBP) algorithm was proposed to eliminate the influence of shadows. Next, the geometric attributes of the 3D point cloud were fused to extract the potential discontinuity trace points on the rocks’ surface. For cases in which the potential discontinuity trace points are too scattered, a local line normalization thinning algorithm was proposed to refine the potential discontinuity trace points. Finally, an algorithm for establishing a two-way connection between a local vector buffer algorithm and a connectivity judgment algorithm was used to connect the discontinuity trace points to obtain the discontinuity traces of the rock mass’s surface. In addition, three datasets were used to compare the results extracted by existing methods. The results showed that the proposed method can extract the discontinuity traces of rock masses with higher accuracy, thereby providing data support for evaluations of the quality of rock masses and stability analyses.

During the extraction and processing of coal, a large amount of solid waste, collectively known as gangue, is produced. This gangue has a low carbon content but a high ash content, accounting for approximately 15 to 20% of the total coal yield. Before coal is used, coal and gangue must be effectively separated to reduce the gangue content in the raw coal and improve the efficiency of coal utilization. This study introduces a classification method for coal and gangue based on a combination of laser-induced breakdown spectroscopy (LIBS) and deep learning. The method employs Gramian angular summation fields (GASF) to convert 1D spectral data into 2D time-series data, visualizing them as 2D images, before employing a novel deep learning model-GASF-CNN-for coal and gangue classification. GASF-CNN enhances model focus on critical features by introducing the SimAM attention mechanism, and additionally, the fusion of various levels of spectral features is achieved through the introduction of residual connectivity. GASF-CNN was trained and tested using a spectral data set containing coal and gangue. Comparative experimental results demonstrate that GASF-CNN outperforms other machine learning and deep learning models across four evaluation metrics. Specifically, it achieves 98.33, 97.06, 100, and 98.51% in the accuracy, recall, precision, and F1 score metrics, respectively, thereby achieving an accurate classification of coal and gangue.

To address the global phenomenon of the salinisation of large land areas, a quantitative inversion model of the salinity of saline soils and soil visible–near-infrared (NIR) spectral data was developed by considering saline soils in Zhenlai County, Jilin Province, China as the research object. The original spectral data were first subjected to Savitzky–Golay (SG) smoothing, multiplicative scattering correction (MSC) pre-processing, and a combined transformation technique. The pre-processed spectral data were then analysed to construct the difference index (DI), ratio index (RI), and normalised difference index (NDI), and the Spearman rank correlation coefficient (r) between these three spectral indices and the salt content in the samples was calculated, while a combined spectral index (r > 0.8) was eventually selected as a sensitive spectral index. Finally, a quantitative inversion model for the salinity of saline soils was developed, and the model’s accuracy was evaluated based on partial least squares regression (PLSR), the random forest (RF) algorithm, and the radial basis function (RBF) neural network algorithm. The results indicated that the inversion of soil salt content using the selected combination of spectral indices based on the RBF neural network algorithm was the most effective, with the prediction model yielding an R² value of 0.950, a root mean square error (RMSE) of 1.014, and a relative percentage deviation (RPD) of 4.479, which suggested a good prediction effect.