Earlier, UAV systems applied in the exploration phase were reviewed via their classification according to the methodologies for acquiring data. However, it was difficult to classify existing literature according to the type of data acquired in the exploitation stage. This section classifies existing literature, according to the purpose of using a UAV, by topographic surveying at the open-pit and underground mine, rock slope analysis, and work environment analysis.
3.1. Topographic Surveying at Open-Pit Mines
It is inconvenient to survey open-pit mines, which are progressing rapidly. This is because topographic surveying requires expensive surveying equipment (total station and terrestrial laser scanner) and professional technicians. However, if a UAV is used, it is possible to rapidly survey a wide area of the open pit within a reasonable budget. Most of the studies reviewed herein acquired aerial photographs using UAVs and used these photographs to generate DEM, DSM, and 3-D models of open-pit mines.
Wang et al. [
56] investigated the usefulness of low-altitude UAVs for obtaining high-resolution images to reconstruct the geometry of an opencast mine. In addition, the relative error was evaluated via a comparison of the DSMs generated from the UAV point cloud and those generated from the terrestrial laser scanner (TLS) data. They developed UAV point clouds via image modeling techniques, such as SfM and a patch-based multi-view stereo (PMVS) algorithm, and then developed a DSM from the study area. While comparing these results with the DSM generated based on the TLS data, the discrepancy map of the 3-D distance based on the DSM showed that the most deviations were less than ± 4 m and that the relative error of the volume was 1.55% [
56].
Cho et al. [
57] verified the applicability of aerial triangulation by applying a small UAV to a mine site. To achieve this, they investigated a smectite mine located in Gyeongju-si, Gyeongsangbuk-do, Korea. They used six ground control positions to perform aerial triangulation for the study area. In addition, 448 aerial photographs were captured in the measurement area of dimensions 600 × 380 m. A 3-D terrain model was constructed using these. It was observed that the use of the ground control positions enabled an aerial triangulation with an accuracy of less than 1 cm.
Lee and Choi [
58,
59] performed topographic surveys of different open-pit mines using rotary- and fixed-wing UAVs. Aerial photographs were acquired of the study area, and orthomosaic images and digital surface models were generated using these. In addition, the position coordinates measured using a differential global positioning system (DGPS) with respect to the ground reference point were compared with those extracted via UAV photogrammetry. The root mean square errors (RMSEs) of the X, Y, and Z-coordinates were 10 (rotary wing) and 15 cm (fixed wing).
Fixed-wing and rotary-wing UAVs differ in the characteristics, such as flight altitude, speed, time, and performance, of the attached cameras. Lee and Choi [
60] compared the results of aerial photogrammetry, which differed for each aircraft type, in the same study area. To achieve this, a construction site located in Yangsan-si, Gyeongsangnam-do was set as the research area. Orthomosaic images and DSM with a 4-cm/pixel spatial resolution were produced using the acquired aerial photographs. Furthermore, coordinates were measured on the ground using DGPS for seven ground control points (GCPs). From a comparison between the aerial survey and ground survey, the RMSE was analyzed to be approximately 10 cm. The fixed wing showed a relatively negligible error when the results of the two types of aerial surveying were compared.
Figure 4 shows the orthographic images and DSMs of the study area for the two types of UAVs.
Chen et al. [
61] investigated the characteristics of iron open-pit mines located in Beijing district, China, using high-resolution topography and landscape metric. They calculated the orientation of the terrace wall in addition to the correlation length. Furthermore, they tested a simple empirical model to derive the percentage of artificial surfaces. First, the main topographic information (DSM) was derived using UAV and SfM photogrammetry techniques. Then, the terraced areas of open-cast and open-pit mines were expressed using the slope local length of autocorrelation (SLLAC) proposed by Sofia et al. [
62], a new landscape metric for the identification of terraced sites.
Rossi et al. [
63] proposed a method to reconstruct the quarry terrain by utilizing nadir and oblique aerial photographs captured using a UAV, and conducted a feasibility analysis thereafter. They set the quarry in Bari, southern Italy, as the research area. They first used nadir images, and then added images obtained from off-nadir angles. They also investigated the accurate set of GCPs for georeferencing and process validation. It was observed that the final position of the point clouds, which represent the main geometries of the quarry environment in the topography reconstruction of the quarry, can achieve an accuracy of a few centimeters. A comparison of the surveys performed using a nadir and oblique image with surveys performed using a total station revealed the advantage of a better description of the shape of the quarries, surface discontinuities, and sub-vertical walls.
Chirico and DeWitt [
64] evaluated the effectiveness of high-resolution orthoimage and DSMs obtained from small low-cost UAS and SfM photogrammetry, in mapping and monitoring mining sites in west Africa. The main objective of the study was to develop two-dimensional and three-dimensional geospatial data that can be used to observe mining pits and to distinguish the features of mine sites. They collected high-resolution orthoimage and DSMs using the wide-angle and narrow-field features of a view camera system and compared these. It was observed that DSMs with high precision and accuracy can be obtained using UAS images and SfM photogrammetry technology.
Gil and Frąckiewicz [
65] performed spatial analysis to optimize the location of the observation network points in an open-pit mine. They used UAVs for aerial photography and Quantum GIS open-source software for spatial analysis. First, an aerial photograph was captured using a UAV; based on this, a digital terrain model (DTM) was generated. The DTM includes heights, elevations, other geographical elements, and natural features, such as rivers and ridgelines. However, it does not represent the elevation of the reflective surfaces of trees, buildings, and other features elevated above the “Bare Earth” that is provided by the DSM. Subsequently, a spatial query and an analysis of the open-pit mine were performed using the Quantum GIS program and DEMs. The spatial analysis revealed that the use of the program substantially facilitates the initial selection of the location area for points.
In addition, various studies have been conducted to measure open-pit mines using UAV or to analyze the results. Tscharf et al. [
66] proposed a workflow for automated image-based reconstruction that can generate detailed and accurate 3-D models from aerial images captured by UAVs and unordered multi-view datasets. They applied and demonstrated the proposed workflow in an open-pit mine. Xiang et al. [
67] established the open-pit mine in Beijing, China, as the research area and took high-resolution aerial photographs twice using a UAV to evaluate geomorphic variations. DEMs were generated through SfM photogrammetry. The surface of the open-pit mine was analyzed by calculating the difference between two DEMs on a cell-by-cell basis and via the SLLAC method. Beretta et al. [
68] evaluated the uncertainty that may appear in DSMs obtained through UAV photogrammetry. To achieve this, UAV aerial photogrammetry was performed in a small open-pit quarry located in southern Brazil. Then, it was compared with a survey method using a real-time kinetic (RTK) GNSS and laser scanning. Kršák et al. [
69] carried out photogrammetry by applying a low-cost commercial UAV to open-pit mines and evaluated the accuracy of the DEM constructed through this. Ge et al. [
70] presented the results of the work performed by applying UAVs in the Ulan open-cut mine and Tahmoor underground mines in New South Wales (NSW), Australia. They used a UAV to evaluate the volume of stockpile, monitor the safety of highwall slopes, and map the underground mine subsidence in these mines. Esposito et al. [
71] used UAV photogrammetry to quantify the volume mined at the Sa Pigada open-pit mine in Sardinia, Italy, and to evaluate the alterations in the mine surface.
Table 4 summarizes the reviewed literature on the application of UAVs to topographic surveying in the phase of mineral exploitation. Among the 15 reviewed studies, 8 applied rotary-wing UAVs, 4 used fixed-wing UAVs, and the rest used both rotary- and fixed-wing UAVs. All 15 studies acquired digital camera images to survey the topography. Twelve articles for the target area scale were analyzed on the mine scale and three on the zonal scale. Additionally, five articles applied an autonomous control type and two articles used a semi-autonomous control type. One paper used a manual mode to operate UAVs. For the remaining studies, the type is unknown.
3.2. Analysis of Rock Slope
UAVs can be used for rock slope analyses (including the stability and discontinuity analysis of rock slopes that are challenging to access directly) and lithological classification of rock masses. All the existing literature reviewed in this paper used drone-based aerial photographs to map the highwall, analyze the characteristics of the slope, and survey the landslides.
McLeod [
72] and McLeod et al. [
73] studied the feasibility of obtaining a 3-D point cloud from video images acquired with a UAV, using SfM photogrammetry software. They used a lightweight vertical take-off and landing UAV with a miniature video camera. First, the performance of the system was evaluated by capturing aerial photographs of the city. The actual applicability was evaluated in the Wollastonite mine in Ontario, Canada. In this task, the overall low point cloud concentration is shown because the SfM processing method is applied at a relatively low pixel resolution (640 × 480). In addition, it was shown that the point on the well-represented flat surface can quantify the accuracy for a range of approximately 0.5 m at a distance of 25–30 m. Although the point cloud developed by the technology was sparse, they anticipated that the overall result could be improved by enhancing the resolution of the original image and the image processing.
Vrublová et al. [
74] documented landslides or difficult-to-reach areas of mines using UAV systems and digital terrestrial photogrammetry. The Czech Nástup Tušimice mine was set up as the study area, and aerial photographs were taken automatically using a UAV. Then, they created an orthophoto and built a 3-D model of the landslide area. In addition, they used digital terrain photogrammetry for a more detailed documentation of the region of interest.
Blistan et al. [
75] also indicated the usability of UAVs for the documentation of outcrops of geological rocks, including in inaccessible parts of mines. They set Slovakia’s Perlite deposit Lehôtka pod Brehmi as the research area, used UAVs to image outcrops that were approximately 75 m tall, and generated photogrammetry documentation. The flight was set at 35 m above the average height of the area, and a total of 58 images were taken. Eighteen GCPs with coordinates obtained by the GNSS RTK method were used for locating and registering each image. The commercial software "AGISOFT PhotoScan" (Agisoft LLC, St. Petersburg, Russia) was used for image processing, and a digital terrain model of the outcrop (which is the object of the survey) was created. A comparison between the correlation between the model obtained by applying the conventional method and the model obtained through aerial photogrammetry revealed that the geological boundaries identified in the two models do not coincide with each other: The difference is as high as 1 m. These differences were attributed mainly to the accuracy of the analog geological maps and sections and the errors in the scale and vectorization of historical maps.
Beretta et al. [
76] classified rocks on the slopes of open-pit mines using UAV photogrammetry and ML. They first reconstructed the terrain through aerial photogrammetry using UAV. Then, they developed a point cloud representing the mining area. The rocks were classified into four groups: Diorite, granite, soil, and vegetation. General ML algorithms (SVM, K-NN, RF, and GTB) were used to categorize the point cloud into four classes. An analysis of the results of each ML algorithm revealed that the SVM algorithm with the radial basis function kernel had the highest accuracy. This result shows the influence of the variability on any type of visual identification on lithological materials. The regular sampling issue in 3-D where samples cannot be directly read in accumulated views has been solved with pseudo-regular samples, with a spacing between the floating points that is approximately equal to the average distance of the original dense cloud [
76].
Katuruza and Birch [
77] used UAV technology for mapping opencast highwalls (which present difficulties in the extraction of information because of limited access for safety reasons) and demonstrated its effectiveness. The study was conducted in Isibonelo Colliery, Mpumalanga Province, South Africa. The highwall was mapped using a drone-based digital photogrammetry technique. The raw data obtained by the UAS was processed within a relatively short period of 48 h, and a 3-D model of the research area was generated. A comparison between the UAS-based 3-D model and the resource model revealed a good correlation.
Stead et al. [
78] reviewed the application of field and remote sensing approaches for the rock slope characteristics at various scales and distances over 15 years surveyed by the Engineering Geology and Resource Geotechnics Research Group at Simon Fraser University in Vancouver, Canada. They summarized potential applications, advantages, and limitations of various remote sensing techniques for comprehensive characterizations of rock slopes. The study showed the importance of remote sensing methods, UAVs, etc. in the investigation of rock slopes. However, the traditional field methods are still important for collecting intact rock and discontinuity condition data.
Table 5 summarizes the reviewed literature on UAV applications to analyze rock slope. A total of seven documents were reviewed. Most of the literature used the rotary-wing type, and the fixed wing was used in one study. All documents obtained digital camera images and conducted a zonal-scale study. In addition, two studies operated UAVs using an autonomous mode, and the flight control style of the remaining studies was unknown.
3.3. Analysis of Working Environments
In addition to taking aerial photographs, UAVs can perform various tasks when installed with thermal and dust sensors, communication modules, and lights. The papers reviewed below are related to the monitoring or analysis of the working environments and workers in mines, using various sensors.
Alvarado et al. [
79] developed a methodology to address the need for a more accurate approach to characterizing blasting plumes in mining sites in near-real time. Because the existing air quality monitoring method relies on a limited number of sampling locations, it is difficult to monitor the point where blasting has occurred, in real time. Therefore, dust data can be collected during flight by attaching an opto-electrical dust sensor to a small fixed-wing and multi-rotor UAV. The dust monitoring system presented in this study showed a technical performance comparable to those of industrial quality dust-monitoring devices. However, an individual calibration equation was required for the sensor to characterize the dust plumes. In addition, the tests described in this paper measured the concentration of PM10 (particulate matter with aerodynamic diameter < 10 μm) with a precision of 1 mg/m
3. However, it was observed that a more accurate measurement of the concentration requires the use of other optical sensors and reference calibration with more precise equipment.
Bamford et al. [
80] presented a concept for measuring rock fragmentation using UAVs and conducted laboratory-scale tests. The purpose of the study was to highlight the advantages of aerial rock fragmentation analysis using a UAV in terms of prediction accuracy and time effort. They explained the procedure for collecting data using the UAV system, and the UAV system configuration. Under the laboratory-scale tests, they first collected rock and pile distribution and information using a UAV system, and manually photographed the rocks and piles for comparative analysis. A comparison of the method using the UAV system and the existing method revealed that the UAV system can measure rock fragmentation within 6% of the accuracy of the existing method, which deviates by up to 14% from the actual distribution. In addition, the time required was within 20%. Based on these results, it is anticipated that rock fragmentation analysis using UAV can improve the reliability of measurement and reduce sampling errors without disturbing the mining process.
Bamford et al. [
81] also investigated the application of a UAV using artificial lighting to measure rock fragmentation in poor lighting conditions, such as those during night work or underground mine work. They investigated the effect of lighting conditions on the analysis of aerial rock fragments through indoor and outdoor experiments. The experiment revealed that the lighting conditions substantially influence the accuracy of the image analysis technique for measuring rock fragmentation. Furthermore, it was verified that the prediction accuracy was improved when artificial lighting was installed on a rock pile or attached to a UAV system.
Motepe [
82] proposed a UAV system capable of conducting search and rescue missions at an accident site of a mine. This was aimed at reducing human injury, direct fatalities, and fatalities owing to delays in treatment that may occur in the mine. The UAV communicates with the control station located on the ground via Wi-Fi and is equipped with a vision system that includes an algorithm that can detect people, so that the rescuer can be informed of their presence. The human detection system is based on the Haar–Cascade classifiers, and the developed model has a very low false alarm rate. During testing of the UAV system, it was possible to detect humans with a probability of approximately 97%, and false detections occurred with a probability of 2.5%.
Péter et al. [
83] analyzed the usefulness of drones for rescue from the perspective of an open-air mine disaster. They surveyed and compared a variety of parameters, such as flight operational time, distance, and working loads, across a variety of products from leading drone companies. In addition, they were compared to rescue helicopters applied to real open-pit mines. Drones would not be able to completely replace rescue helicopters because drones are incapable of transporting people or objects or travelling long distances similar to rescue helicopters. However, drones have the advantage of having the capability to conveniently access areas difficult to reach by helicopters. This is because of the rapid take-off and landing capability and the small size of the aircraft. Therefore, it was concluded that it is highly efficient to operate a drone as an assistance unit of a rescue helicopter.
Because mines have complex topologies, setting up wireless communication with a traditional rescue robot is a highly complex task. Therefore, in the case of a mine emergency, it is highly challenging to perform rescue and recovery operations in a timely manner. Ranjan et al. [
84] proposed a UAV-based multihop emergency communication system to support miners and rescue team members in an emergency. They tested the emergency communication system by dividing the experiment into three scenarios: No UAV, one UAV, and two UAVs. To evaluate each scenario, performance indicators were considered, such as end-to-end packet error rate (PER), successful end-to-end per packet delay, and the number of retransmissions. The evaluation verified that the performance of the UAV-based system was improved in terms of the packet error rate, end-to-end delay, and per packet retransmission.
Table 6 summarizes the six reviewed studies on the application of UAVs to analyze working environments in the phase of mineral exploitation. For the UAV type, four applied rotary-wing UAVs, one used both, and another was unknown. Among the six reviewed studies, three studies obtained digital camera images and two papers acquired dust data and communication data. In terms of the target area scale, three papers were analyzed on the mine scale, two on the zonal scale, and one on the regional scale. In addition, four studies used an autonomous flight mode, one study used a semi-autonomous flight mode, and the other used the manual flight mode.
3.4. Surveying at Underground Mines
There are few UAVs and instruments that are dedicated to underground environments. Similar environments typically have low-visibility conditions, confined openings, magnetic interference, and an absence of GPS coverage [
85]. However, if UAV systems are equipped with high-resolution cameras, LED lights, and thermal sensors, useful information, such as image (thermal, spectral, etc.), distance, inertial measurement unit (IMU), and sound navigation and ranging (SONAR) data, can be obtained in areas that are difficult to be accessed by mine workers.
Kanellakis and Nikolakopoulos [
86] used UAVs to enable vision-based mine inspections. Thereby, they presented an assessment of the current technology of visual localization systems for underground mining. In addition, they verified whether UAV-based localization technology can be applied in harsh and challenging environments, while using reliable and low-cost existing methods and technologies. To achieve this, an experiment was conducted on a visual localization system at an iron ore mine in Kiruna, Sweden, one of the largest mines in Europe. The experimental results showed that stereo cameras can be used for automated UAVs. However, drift errors accumulated in the results. Meanwhile, it has been demonstrated that the monocular camera approach can be used successfully under certain constraints.
Freire and Cota [
87] designed a UAV to capture images of areas of underground mines that are rendered inaccessible due to mining and blasting. This UAV features a balloon filled with helium gas, a quadcopter propeller with remote-control powerful LED lighting, a rechargeable battery, a remote-control camera, an image stabilizer, and a radio frequency transceiver for control and image visualization. In addition, owing to the characteristic of the underground mine, the equipment could fly without using GPS and at variable low speeds to prevent collisions with rock walls and support elements. The test was conducted at a sublevel stope with a width of 5–6 m, dip of 45°, and with 20 m between sublevels. The LED lighting attached to the UAV was sufficient to reveal the details of the rock walls, rock mass structures, evidence of blasting, and support elements. In addition, no signal loss occurred during flight even when the balloon disappeared from direct view.
Raj [
88] proposed a monitoring system using drones to improve the safety of monitoring equipment, deliver fast and real-time monitoring results, and minimize human exposure to unsafe underground conditions. In addition, this study demonstrated that it is possible to survey and monitor using drones in underground mines where GPS is non-functional, light conditions are low, and spaces are confined. The main contents of his research are as follows: (1) Development of image capture and control technology for application in narrow spaces and underground environments; (2) development of a solution that can utilize drones even in dark environments, using specially designed lighting attached to the drones; (3) image capture with significant quality and quantity to create three-dimensional point clouds; and (4) evidence of geotechnical rock mass characteristics and rock mass movement by processing the generated point clouds data. The study revealed that the point clouds created using the drone were highly similar to those created using the light detection and ranging (LiDAR) scanner. However, the noise appears to occur more while using the drone. Notwithstanding the relatively lower point resolution, important geotechnical information was obtained, including fracture direction and variation detection.
Turner et al. [
85] created a 3-D model by thermal imagery using a UAV in an underground mine. They also conducted a study on the identification of geological data in a photogrammetric model. This study showed that off-the-shelf technologies could be used to obtain high-quality data for geotechnical analysis. The UAV system includes obstacle detection, lighting, thermal imagery, and software. Depending on the geological structure or fracture in the rock mass, the material may deform, and a difference in temperature may occur. Consequently, a thermal camera can be used to verify the temperature contrasts and assess the stability of the underground opening. The use of a UAV was tested at Barrick Golden Sunlight Mine, Whitehall, Montana, USA. The study revealed that the thermal image lacks the refinement and point density of the RGB model. However, it was more convenient to detect loose ground and detailed structures. The study revealed that a good combination of off-the-shelf technologies with UAVs can effectively perform photogrammetry and identify geological data in underground mining environments.
Turner et al. [
89] also investigated the detection and quantification of geological discontinuities in hard rock masses, using thermal and multispectral images obtained using UAVs. Multiple thermal; multispectral; red, green, and blue (RGB); and LiDAR data sets were acquired in the same study area. They used this data to generate georeferenced 3-D point clouds and meshes, and to map discontinuities.
In addition to these studies, Mitchell and Marshall [
90] presented potential applications of UAVs in underground mines. They also developed a prototype of UAVs capable of automatic rotation for underground mining scanning. Azhari et al. [
91] analyzed the construction of a UAV sensor suite that can generate rough 3-D models in real time using SONAR data and provide operators with high contextual awareness.
Table 7 summarizes the reviewed literature on UAV applications of surveying in underground mines. All seven reviewed studies conducted the study by using UAVs of the rotary-wing type. Various data, such as distance data, digital camera images, thermal images, and SONAR data, were acquired for topographic surveying in underground mines. Half the papers reviewed were analyzed on a mine scale and the other half on a zonal scale. In addition, two studies used an autonomous flight mode, two papers used a semi-autonomous mode, and the rest of the studies were unknown.