The global population increase, coupled with the effects of climate change, poses an imminent threat to food security. Changes in agricultural practices are necessary to meet the growing food demand while minimising negative impacts on water and soil. To make informed decisions, policymakers need timely, accurate, and efficient crop maps. Spectral similarities between crops are a major limiting factor in crop map accuracy. Crops that have similar spectral responses are difficult to distinguish using traditional multispectral image analysis. The aim of this study is to develop a system that is capable of accurately identifying crop maps regardless of the similarities that might exist between classes. To achieve this, we propose a hierarchical approach that uses NDVI time-series and ground truth data combined with machine learning and deep learning models. For each leaf of the hierarchical tree, a separate model is trained. A case study of the proposed system in the Gharb region of Morocco showed that hierarchical crop mapping reached an F1-measure of 0.90, outperforming commonly used flat classifiers by 12-17%.

Soil Total Nitrogen (TN) is a critical nutrient for plant growth, and its monitoring in agricultural soils is essential for sustainable nutrient management. Conventional laboratory-based TN analyses are accurate but remain laborious, costly, and limited for spatially continuous large-scale applications. Spaceborne hyperspectral remote sensing provides a promising alternative for mapping soil TN at different scales. However, its use is limited by spectral gaps caused by strong atmospheric water-vapor absorption, particularly in nitrogen-sensitive near-infrared (NIR) and shortwave infrared (SWIR) regions. This study evaluates the contribution of reconstructing these missing spectral domains for improving soil TN estimation from PRISMA (PRecursore IperSpettrale della Missione Applicativa) hyperspectral imagery. A spectral gap-filling framework based on a conditional generative adversarial network (cGAN) coupled with a self-supervised masked autoencoder-inspiredstrategy, pretrained using hyperspectral data acquired over the Meknes region (Sacs Plateau) was developed to reconstruct continuous reflectance spectra, including the water-vapor absorption intervals (1320-1500 nm and 1780-2050 nm) and the NIR-SWIR overlap region (950-990 nm). The reconstruction model achieved high accuracy, with coefficients of determination of R2 = 0.95 on PRISMA spectra and R2 = 0.91 when validated against laboratory spectroradiometer (ASD FieldSpec III) measurements. The reconstructed spectra were subsequently used for soil TN estimation using open-access ISDA soil TN map product at 30 m resolution. A total of 1037 samples were analyzed over three bare-soil croplands agricultural regions (i.e., Al Haouz, Doukkala plain, and Khouribga) in Morocco. Our results show that incorporating reconstructed bands improves TN prediction performance. In Al Haouz, R2 increased from 0.83 to 0.89, with RMSE decreasing from 0.14 to 0.11 g center dot kg-1 (RPIQ: 3.1-3.65; RPD: 2.49-2.91). In Sidi Bennour-Doukkala, R2 improved from 0.73 to 0.79 and RMSE from 0.11 to 0.08 g center dot kg-1 (RPIQ: 2.52-2.87; RPD: 2.06-2.14) using LASSO regression. In Khouribga, the best performance was obtained with Random Forest (R2 = 0.73; RMSE = 0.11 g center dot kg-1; RPIQ = 2.52; RPD = 1.98). Performance metrics were evaluated against ISDA-derived soil TN reference data and therefore reflect agreement with map-based predictions rather than direct laboratory measurements. Feature-selection analyses further revealed that the most informative wavelengths for TN estimation are predominantly concentrated in the NIR-SWIR regions of 1050-1450 nm, 1800-2100 nm, and 2300-2400 nm, including several reconstructed bands within water-vapor absorption intervals. These findings demonstrate that spectral gap filling enhances the coherence and usability of spaceborne hyperspectral data for soil TN monitoring, supporting more robust digital soil mapping and precision agriculture applications.

Reliable crop-type mapping in heterogeneous agricultural landscapes remains challenging, particularly when models trained on one region are transferred to another. This study proposes an integrated framework that combines hyperspectral band selection with multimodal foundation-model embeddings to improve the spatial robustness of winter cereals mapping. A Spectral Attention Module was applied to an EnMAP image to identify 29 informative wavelengths consistent with vegetation biophysical properties. Complementary structural and climatic information was encoded using Presto embeddings derived from Sentinel-1, Sentinel-2, ERA5, and DEM data, compressed through PCA and fused with the selected hyperspectral bands. Three classifiers-SVM, Random Forest, and CatBoost-were evaluated under a nested K-means spatial cross-validation scheme and tested on an independent dataset. Results show that the embedding-enhanced scenario consistently improves Accuracy, F1-score, and ROC-AUC, with RF and CatBoost gaining more than 8-13 percentage compared to the spectral-only scenario. SVM remained the strongest performer overall, but all models exhibited improved spatial generalization. These findings demonstrate the value of combining hyperspectral features with multimodal embeddings for scalable and transferable crop-mapping applications.

Energy efficiency and product quality control are critical concerns in grinding mill operations, particularly within the innovative context of Mine 4.0. This study introduces a novel Genetic Algorithm (GA)-based optimization framework specifically developed to address these challenges. Given the mining industry's significant energy consumption, especially in grinding processes, the proposed approach optimizes key parameters such as feed composition, water flow rates, and power consumption levels, while maintaining sieve refusal near the target threshold of 20%. Using real operational data from a Moroccan plant, the GA achieved a Mean Absolute Error (MAE) of 0.47, outperforming Simulated Annealing (SA) and Particle Swarm Optimization (PSO), which yielded MAEs of 1.14 and 0.74, respectively. The GA also demonstrated superior convergence stability and robustness, as evidenced by lower variability in predicted power consumption. These results validate the effectiveness of the GA framework in navigating nonlinear, high-dimensional parameter spaces and improving energy efficiency while ensuring product quality consistency. Ultimately, this research confirms the potential of metaheuristic optimization in enhancing grinding mill efficiency and supports the broader shift towards intelligent and sustainable mining operations under the Mine 4.0 paradigm.

Identifying rock mass instabilities is crucial in geotechnical monitoring, particularly in underground structures such as deep underground mines. Conventional techniques for identifying unstable rock blocks rely on visual and acoustic assessments, which lack automation and objectivity. Consequently, there is an increasing demand for sophisticated monitoring techniques that fulfill the precision and effectiveness standards for underground geotechnical assessment. This study will employ deep learning-based computer vision models to propose an intelligent approach for detecting and segmenting unstable rock blocks in underground mines. It utilizes three variants of the U-shaped convolutional neural network, referred to as U-Net: Base-U-Net, Attention U-Net, and BCDU-Net (Backbone Convolutional Deconvolution U-Net), to identify and segregate unstable rock blocks in thermal images. A detailed dataset of 1502 annotated images was assembled from a video recorded at the Draa Sfar deep underground mine in Morocco using a FLIR A70 Research Development thermal camera for model training, validation, and testing. In this work, the Base-U-Net model achieved high training and validation accuracies, as well as an average performance during the testing phase. The results demonstrate the effectiveness of the U-Net model in segmenting unstable rock blocks in thermal images, particularly as a first attempt to apply this model in the field of mining geotechnics. Our research highlights the use of deep learning-based computer vision models to enhance geotechnical engineering applications and safety protocols in mining structures.

Foundation models are transforming Earth observation, but their potential for hyperspectral crop mapping remains underexplored. This study benchmarks three foundation models for cereal crop mapping using hyperspectral imagery: HyperSigma, DOFA, and Vision Transformers pre-trained on the SpectralEarth dataset (a large multitemporal hyperspectral archive). Models were fine-tuned on manually labeled data from a training region and evaluated on an independent test region. Performance was measured with overall accuracy (OA), average accuracy (AA), and F1-score. HyperSigma achieved an OA of 34.5% (+/- 1.8%), DOFA reached 62.6% (+/- 3.5%), and the SpectralEarth model achieved an OA of 93.5% (+/- 0.8%). A compact SpectralEarth variant trained from scratch achieved 91%, highlighting the importance of model architecture for strong generalization across geographic regions and sensor platforms. These results provide a systematic evaluation of foundation models for operational hyperspectral crop mapping and outline directions for future model development.

Rockfalls are among the frequent hazards in underground mines worldwide, requiring effective methods for detecting unstable rock blocks to ensure miners' and equipment's safety. This study proposes a novel approach for identifying potential rockfall zones using infrared thermal imaging and image segmentation techniques. Infrared images of rock blocks were captured at the Draa Sfar deep underground mine in Morocco using the FLUKE TI401 PRO thermal camera. Two segmentation methods were applied to locate the potential unstable areas: the classical thresholding and the K-means clustering model. The results show that while thresholding allows a binary distinction between stable and unstable areas, K-means clustering is more accurate, especially when using multiple clusters to show different risk levels. The close match between the clustering masks of unstable blocks and their corresponding visible light images further validated this. The findings confirm that thermal image segmentation can serve as an alternative method for predicting rockfalls and monitoring geotechnical issues in underground mines. Underground operators worldwide can apply this approach to monitor rock mass stability. However, further research is recommended to enhance these results, particularly through deep learning-based segmentation and object detection models.

Accurate and efficient crop maps are essential for decision-makers to improve agricultural monitoring and management, thereby ensuring food security. The integration of advanced artificial intelligence (AI) models with hyperspectral remote sensing data, which provide richer spectral information than multispectral imaging, has proven highly effective in the precise discrimination of crop types. This systematic review examines the evolution of hyperspectral platforms, from Unmanned Aerial Vehicle (UAV)-mounted sensors to space-borne satellites (e.g., EnMAP, PRISMA), and explores recent scientific advances in AI methodologies for crop mapping. A review protocol was applied to identify 47 studies from databases of peer-reviewed scientific publications, focusing on hyperspectral sensors, input features, and classification architectures. The analysis highlights the significant contributions of Deep Learning (DL) models, particularly Vision Transformers (ViTs) and hybrid architectures, in improving classification accuracy. However, the review also identifies critical gaps, including the under-utilization of hyperspectral space-borne imaging, the limited integration of multi-sensor data, and the need for advanced modeling approaches such as Graph Neural Networks (GNNs)-based methods and geospatial foundation models (GFMs) for large-scale crop type mapping. Furthermore, the findings highlight the importance of developing scalable, interpretable, and transparent models to maximize the potential of hyperspectral imaging (HSI), particularly in underrepresented regions such as Africa, where research remains limited. This review provides valuable insights to guide future researchers in adopting HSI and advanced AI models for reliable large-scale crop mapping, contributing to sustainable agriculture and global food security.

This study addresses the problem of early detection of leaf miner infestations in chickpea crops, a significant agricultural challenge. It is motivated by the potential of hyperspectral imaging, once properly combined with machine learning, to enhance the accuracy of pest detection. Originality consists of the application of these techniques to chickpea plants in controlled laboratory conditions using a natural infestation protocol, something not previously explored. The two major methodologies adopted in the approach are as follows: (1) spectral feature-based classification using hyperspectral data within the 400–1000 nm range, wherein a random forest classifier is trained to classify a plant as healthy or infested with eggs or larvae. Dimensionality reduction methods such as principal component analysis (PCA) and kernel principal component analysis (KPCA) were tried, and the best classification accuracies (over 80%) were achieved. (2) VI-based classification, leveraging indices associated with plant health, such as NDVI, EVI, and GNDVI. A support vector machine and random forest classifiers effectively classified healthy and infested plants based on these indices, with over 81% classification accuracies. The main objective was to design an integrated early pest detection framework using advanced imaging and machine learning techniques. Results show that both approaches have resulted in high classification accuracy, highlighting the potential of this approach in precision agriculture for timely pest management interventions.

Africa’s rapidly growing population is driving unprecedented demands on agricultural production systems. However, agricultural yields in Africa are far below their potential. One of the challenges leading to low productivity is Africa‘s poor soil quality. Effective soil fertility management is an essential key factor for optimizing agricultural productivity while ensuring environmental sustainability. Key soil fertility properties—such as soil organic carbon (SOC), nutrient levels (i.e., nitrogen (N), phosphorus (P), potassium (K), moisture retention (MR) or moisture content (MC), and soil texture (clay, sand, and loam fractions)—are critical factors influencing crop yield. In this context, this study conducts an extensive literature review on the use of hyperspectral remote sensing technologies, with a particular focus on freely accessible hyperspectral remote sensing data (e.g., PRISMA, EnMAP), as well as an evaluation of advanced Artificial Intelligence (AI) models for analyzing and processing spectral data to map soil attributes. More specifically, the study examined progress in applying hyperspectral remote sensing technologies for monitoring and mapping soil properties in Africa over the last 15 years (2008–2024). Our results demonstrated that (i) only very few studies have explored high-resolution remote sensing sensors (i.e., hyperspectral satellite sensors) for soil property mapping in Africa; (ii) there is a considerable value in AI approaches for estimating and mapping soil attributes, with a strong recommendation to further explore the potential of deep learning techniques; (iii) despite advancements in AI-based methodologies and the availability of hyperspectral sensors, their combined application remains underexplored in the African context. To our knowledge, no studies have yet integrated these technologies for soil property mapping in Africa. This review also highlights the potential of adopting hyperspectral data (i.e., encompassing both imaging and spectroscopy) integrated with advanced AI models to enhance the accurate mapping of soil fertility properties in Africa, thereby constituting a base for addressing the question of yield gap.

The control of the froth flotation process in the mineral industry is a challenging task due to its multiple impacting parameters. Accurate and convenient examination of the concentrate grade is a crucial step in realizing effective and real-time control of the flotation process. The goal of this study is to employ image processing techniques and CNN-based features extraction combined with machine learning and deep learning to predict the elemental composition of minerals in the flotation froth. A real world dataset has been collected and preprocessed from a differential flotation circuit at the industrial flotation site based in Guemassa, Morocco. Using image-processing algorithms, the extracted features from the flotation froth include: the texture, the bubble size, the velocity and the color distribution. To predict the mineral concentrate grades, our study includes several supervised machine learning algorithms (ML), artificial neural networks (ANN) and convolutional neural networks (CNN). The industrial experimental evaluations revealed relevant performances with an accuracy up to 0.94. Furthermore, our proposed Hybrid method was evaluated in a real flotation process for the Zn, Pb, Fe and Cu concentrate grades, with an error of precision lesser than 4.53. These results demonstrate the significant potential of our proposed online analyzer as an artificial intelligence application in the field of complex polymetallic flotation circuits (Pb, Fe, Cu, Zn).

In the dynamic landscape of modern manufacturing, the pursuit of efficiency, reliability, and optimal performance has prompted the integration of cutting-edge technologies. Among these, digital twins (DT) have emerged as transformative tools, offering a virtual representation of physical processes, systems, and equipment. This paper delves into the pivotal role of digital twins in advancing the monitoring and supervision of manufacturing processes, focusing specifically on process digital twin (PDT) and its application in the domain of froth flotation in minerals processing. Our data-driven digital twin, replicating the behavior of a flotation cell, was developed using a combination of industrial and simulation data anchored by Artificial Neural Networks. This approach provides precise process emulation of the flotation process. Industrial evaluations of the AI model within the Digital Shadow demonstrated an overall 94% accuracy in estimating insightful information regarding the flotation cell operations with 2 s in response time. This research significantly contributes to the practical implementation of digital twins in industrial processes, highlighting their potential to revolutionize process control and enhance efficiency in the industrial sector.

In minerals processing, the froth flotation is one of the widely used process that separates valuable mineral components from their associated gangue materials. The efficiency of this process relies on several factors, such as feed characteristics, particle size, pulp flow rate, pH, conditioning time, aeration, reagents system and many other affecting parameters. These processing parameters significantly impact the overall performance of the flotation process and influence the quality of the final concentrate. For instance, improper pulp flow and reagent dosing systems can result in metal loss and waste, particularly when dealing with frequently changing ore compositions. In this work, we established an Artificial Intelligence-based system which goal is to intelligently monitor flotation circuits and to recommend set-points for the process’s manipulated variables in order to achieve optimal performance.The system has been developed and evaluated within an industrial flotation plant that processes complex Pb-Cu-Zn sulfide ores. Leveraging an Artificial Neural Network-based Mixture of Experts (MoEs) predictive model, the system accurately estimates the mineral grades in the final concentrate and tailing of the flotation circuit. Moreover, using a Genetic Algorithms-based optimization pipeline, the system recommends set-points for the manipulated variables of the process for a maximum recovery and optimal product quality.The industrial validation of the predictive component demonstrated a 94% accuracy with a rapid 3s response time. Furthermore, the hypothetical simulation of the optimization component indicated a potential 5% increase in circuit recovery and a 4% increase of lead (Pb) grade in the circuit’s final concentrate. This developed system aims to enhance the control of froth flotation process, stabilize the product quality, and improve the overall economic benefits of production efficiency. This research contributes to the field of manufacturing systems by providing practical data-driven application for the advanced monitoring, optimization and control of industrial processes with a specific emphasis on the froth flotation process.

Safety in underground mining is critically challenged by environmental conditions and the need for rigorous adherence to safety protocols. Draa Sfar, the deepest mine in Morocco, presents extreme conditions that test the effectiveness of Personal Protective Equipment (PPE) compliance. This study addresses the gaps in real-time safety monitoring and compliance in such challenging environments. The primary objective of this research is to enhance PPE compliance detection in underground mines using advanced computer vision techniques. The study aims to develop a system that not only detects PPE but also ensures its proper use through pose estimation. The study involved collecting and annotating a unique dataset from the Draa Sfar mine, characterized by its harsh environmental conditions. Pose estimation was performed using the newly developed You Only Live Once (YOLO) Pose v8 algorithm, tailored for miners in underground settings. For PPE detection—specifically helmets, safety vests, gloves, and boots—we employed and compared several models including YOLO v8, v9, v10, Real-Time Detection Transformer (RT-DETR), and YOLO World. PPE compliance was then assessed by integrating pose estimation keypoints to filter out false detections effectively. The integrated approach successfully identified and verified the use of PPE with high accuracy. Comparative analysis showed that newer versions of YOLO alongside RT-DETR provided substantial improvements in detection rates under varied lighting and spatial conditions prevalent in underground mines. The findings demonstrate that combining pose estimation with advanced object detection frameworks significantly enhances PPE compliance monitoring in underground mines. This dual approach reduces the risk of false positives and ensures a more reliable safety system. By improving the accuracy and reliability of safety equipment detection in one of the most challenging mining environments, this research contributes to reducing occupational hazards and enhancing miner safety. The implications extend to other high-risk industries where environmental conditions complicate safety monitoring.

In the mineral processing industry, specifically in froth flotation, the extraction of detailed information from bubble images is imperative for effective monitoring of the flotation process and its associated production indicators. This study delves into a range of semantic segmentation methods and algorithms, notably including YOLO, Watershed, and Thresholding, to accurately process these images. Our investigation leads to the proposal of an innovative cloud-based segmentation architecture, seamlessly integrated with the Internet of Things (IoT). This integration not only enhances the segmentation process but also supports a comprehensive monitoring application, offering a significant advancement in the real-time analysis and optimization of the flotation process. The article presents an empirical comparison of the segmentation methods and demonstrates the efficacy of the proposed cloud-based system in a practical industrial setting.

This study aims to assess the squeezing rock severity at Draa Sfar deep underground mine in Morocco. In fact, this phenomenon is one of the frequent geotechnical issues occurring in deep underground mines, especially those excavated in weak rock formations. Therefore, this study is necessary to evaluate the squeezing potential of pelites, which are the most dominant formations at Draa Sfar mine’s deep levels, and to suggest ways for the management of this phenomenon in the mine. In this paper, previous studies related to squeezing rock phenomenon in both mining and tunneling contexts are presented. In addition, five empirical methods for squeezing rocks severity assessment are presented and compared. It was found that the method based on foliation spacing—stress to strength ratio matrix is the most accurate one to use in this study regarding the parameters it involves and its suitability to the underground mining context. Applying this method to Draa Sfar underground mine’s data shows that the squeezing rocks severity in the mine can go from a minor to a severe level depending on the interception angle. This allows for the comparison of the mine to other cases worldwide where the squeezing rock phenomenon was reported. These findings are useful to assess the environmental risk related to the squeezing of pelites at Draa Sfar underground mine and to adapt the galleries’ design to avoid its occurrence. Integrating our results into similar mining projects will serve to reduce the impact of the squeezing rock issue to ensure better mining conditions and to increase the safety of miners worldwide.

Digitalization is crucial for achieving product quality, cost reduction, and timely delivery in manufacturing. With the rise of Industry 4.0, Digital Twin technology has gained popularity, enabling real-time monitoring, predictive maintenance, process optimization, and assumption testing. Manufacturing Digital Twins utilize digital models to represent physical processes, facilitating continuous simulation, interaction, correction, and optimization for autonomous control. However, implementing Digital Twins can be challenging due to diverse data requirements and multiple services. This paper proposes a scalable, service-oriented multi-layered architecture for developing Digital Twins in mineral processing, which can be adapted to other manufacturing systems. The paper provides an overview of Digital Twins, their data-centric workflow, and industrial applications. Additionally, a case study on implementing a Digital Twin for the froth flotation process in the mining industry is presented based on the proposed architecture.

The goal of this paper is to develop a deep learning model for predicting citrus yield. The data used consists of two sources: (1) field data that includes information on fertilization and phytosanitary treatment products, water quantities used for irrigation, climatic data (temperature, precipitation, humidity, wind speed, and solar radiation), parcel sizes, and rootstock types for each parcel. (2) The second source comprises images representing the normalized difference vegetation index (NDVI) and the normalized difference water index (NDWI), extracted from Sentinel-2 images taken before the harvest period. The data was collected over a period of 5 years, from 2015 to 2019, and pertains to 50 parcels within a Moroccan orchard. Following data preparation, we constructed a deep learning neural network model with multiple layers and parameters. This model takes the information from each parcel as input for training purposes. Subsequently, we evaluated the model using new data obtained from additional parcels located at various sites within our orchard. The test phase resulted in the following scores: 0.0458 (Mean Squared Error), 0.1450 (Mean Absolute Error), and 0.10 (Percentage Error). These scores reflect the strong predictive capability of our approach.

The mining industry deals with complexity and high energy consumption in mineral production. Artificial intelligence technology has been integrated to manage and optimize energy usage in mining equipment. A special focus is given to grinding mills, which are essential but difficult to model due to their energy-intensive nature. Data-driven machine learning approaches have emerged as effective solutions, enabling energy consumption prediction and optimization. These methods offer valuable insights to improve the overall energy efficiency of grinding mills in mining. In this article, a comparative study is conducted to investigate the use of machine learning models for predicting the power consumption of grinding mills. The study starts by providing a detailed description of the dataset used and the predictive models employed in the research. Notably, the results of power prediction for grinding mills are compared, with the Random Forest model emerging as the top performer, boasting an impressive R2 value of 0.94. The article underscores the significance of this predictive capability in helping mining experts make informed decisions to optimize energy consumption in grinding mills. Finally, the article concludes by offering insightful recommendations and identifying promising avenues for future research in this field. Overall, this study sheds light on the potential of machine learning to enhance energy efficiency in the mining industry.