Traditional geological exploration and mineral prospecting heavily rely on field surveys, which are often inefficient, costly, and severely limited by complex terrain and dense vegetation. As a practitioner in this field, I have encountered numerous challenges in high-relief mountainous areas where direct field observation is either dangerous or impossible. Over the past decade, the rapid advancement of Unmanned Aerial Vehicle (UAV) technology has revolutionized our approach. In particular, the synergy between UAV oblique photogrammetry and airborne LiDAR technology has emerged as a transformative solution, enabling us to acquire high-precision, three-dimensional (3D) spatial data that significantly enhances our ability to interpret geological structures and locate mineral deposits. This paper presents my comprehensive study on the integrated application of these two technologies, using a mining area in central Zhejiang Province as a case study to demonstrate the methodology, validation, and quantitative benefits of this approach.
1. Introduction: The Need for a Fusion Approach
The core challenge in modern geological exploration is the trade-off between data richness and accuracy. UAV drone oblique photogrammetry captures high-resolution multi-angle imagery, generating vivid 3D reality models with rich texture and color information. This is invaluable for identifying surface lithologies, alteration zones, and man-made features such as prospecting pits and trenches. However, its passive optical nature makes it vulnerable to vegetation occlusion; in forested areas, the ground surface and underlying geology are completely masked. Conversely, airborne LiDAR is an active remote sensing technology that emits laser pulses capable of penetrating canopy gaps. It produces dense point clouds that accurately represent the “bare earth” digital elevation model (DEM), even under dense vegetation. Yet, LiDAR data alone lacks spectral texture, making it difficult to discriminate between different rock types or alteration patterns.
My work focuses on a 6.2 km² mining district in central Zhejiang. The area features steep terrain and a thick vegetation cover of shrubs and trees, which rendered conventional geological mapping highly challenging. Historical small-scale mining had left behind numerous surface pits, adits, and trenches, but their exact locations were poorly documented. The primary objective was to quickly and accurately map these surface workings, decipher the structural controls (faults and fractures) on mineralization, and obtain a high-precision topographic base for subsequent geophysical surveys. A single-technology approach was deemed insufficient: photogrammetry would fail in the forest, and LiDAR alone would not distinguish the mineralized zones from barren rocks. Therefore, I designed a workflow that integrates both datasets from the moment of acquisition through final interpretation.
2. Methodology: A Data-Driven Exploration Workflow
The methodological framework is built on four pillars: (1) simultaneous data acquisition planning, (2) rigorous pre-processing and co-registration, (3) hybrid interpretation using both raster and vector signatures, and (4) field validation. The flowchart governing this process outlines the key steps from flight planning to final geological outputs.
2.1 Hardware and Flight Planning Parameters
To maximize efficiency and data quality over the rugged terrain, I selected the Feima D20 UAV platform equipped with a DV-LiDAR 22 sensor module. This system simultaneously captures high-resolution optical images and LiDAR point clouds. To maintain consistent point density on steep slopes, a terrain-following flight mode was adopted. The key parameters are summarized in Table 1.
| Parameter | Value | Rationale |
|---|---|---|
| Terrain Type | Hilly & Mountainous | Required terrain-following for consistent resolution |
| Flight Altitude (AGL) | 200 m | Balances safety with required resolution (standard for large areas) |
| Point Cloud Density | 26 points/m² | Sufficient for 1:1,000 scale mapping and micro-topography |
| Side Overlap (Images) | 60% | Ensures robust 3D reconstruction of complex slopes |
| Flight Speed | 8 m/s | Optimal for LiDAR pulse density and camera shutter stability |
The adoption of a 60% side overlap for the optical images was critical for generating a seamless 3D model, while the LiDAR sensor was calibrated to ensure a high signal-to-noise ratio even under partial canopy cover. The entire survey area of 6.2 km² was covered in two sorties, with a total flight time of approximately 3 hours.
2.2 Data Pre-processing: From Raw Point Clouds to a Unified Model
The raw data from each flight includes Global Navigation Satellite System (GNSS) trajectories, Inertial Measurement Unit (IMU) data, LiDAR ranges, and imagery. The pre-processing pipeline involves five essential steps:
Step 1: Differential GNSS and Trajectory Processing. The on-board GPS data is post-processed against a local base station to achieve centimeter-level positioning accuracy. The smoothed best estimate of trajectory (SBET) file is generated.
Step 2: LiDAR Point Cloud Generation. The raw laser ranges are combined with the trajectory to compute 3D coordinates for each return. The resulting point cloud is initially in the WGS-84 ellipsoidal height system, which is then transformed to the 2000 National Geodetic Coordinate System (CGCS2000) with a 3° Gauss-Kruger projection.
Step 3: Noise Removal. The raw point cloud contains outliers due to multipath effects, atmospheric scattering, or flying birds. I applied a statistical outlier removal filter. The algorithm calculates the mean distance \( d_i \) from each point \( p_i \) to its \( k \) nearest neighbors. Points where \( d_i \) exceeds a threshold \( D_{max} \) are classified as noise.
$$ D_{max} = \mu + \alpha \cdot \sigma $$
Where \( \mu \) is the mean of all \( d_i \), \( \sigma \) is the standard deviation, and \( \alpha \) is a multiplier (set to 1.0). I used \( k = 10 \) neighboring points, which effectively removed isolated low points and high-flying birds while preserving true terrain features.
Step 4: Ground Point Classification. This is the most critical step for bare-earth DEM generation. I employed an improved Progressive Morphological Filter (PMF) algorithm, which iteratively builds a Triangulated Irregular Network (TIN) surface. Unlike the classical approach, the improved version optimizes the selection of initial seed points to avoid the “flattening” of ridge crests. The algorithm uses a slope-based threshold to decide whether a point belongs to the ground or to vegetation. Points are classified as “ground” if their elevation difference from the current TIN surface is less than a dynamically calculated threshold \( T \), where \( T \) is a function of terrain slope and iteration.
$$ T(d) = T_{init} + (d \cdot \tan(\theta)) $$
Where \( d \) is the distance from the point to the TIN facet, and \( \theta \) is the local slope angle estimated from the previous iteration. This process preserves sharp ridgelines and escarpments, yielding a more realistic topographic representation.
Step 5: DEM and Orthophoto Generation. After classifying the ground points, I interpolated a high-resolution DEM using the Natural Neighbor algorithm (1). This algorithm produces a smooth, continuous surface that respects local data density. The DEM resolution was set to 0.2 m to capture micro-topographic features. Concurrently, the aerial images were orthorectified using the LiDAR-derived DEM, producing a true orthophoto mosaic with a GSD of 0.05 m.
2.3 Establishing Interpretation Keys: A Dual-Dimension Approach
A major innovation in our study was the creation of a “Raster + Vector” interpretation flag system. Instead of relying solely on visual cues from the orthophoto, I defined explicit signatures derived from both datasets. The primary features of interest were exploration pits, trenches, and adit portals.
Table 2: Dual-Dimension Signature for Key Geological Features
| Feature | Raster Signature (Oblique Model) | Vector Signature (LiDAR DEM) |
|---|---|---|
| Exploration Pit | Sharp, linear shadow boundary on one side; high-brightness strip on opposite slope; dark shadow at pit bottom | Sudden negative elevation anomaly with high slope gradient (\( >45^\circ \)) on pit walls; high local roughness |
| Trench | Narrow, continuous linear depression with alternating shadow/highlight along its length | Elongated, low-lying corridor bordered by steep, sharp breaklines; low mean curvature along the trench floor |
| Adit Portal | A small, dark, geometrically regular (semi-circular or rectangular) shadow on a slope; often near a flat platform | Local increase in slope roughness; small flat area (platform) adjacent to a steep slope (adit face); presence of dump debris nearby |
This dual-dimension approach dramatically reduced false positives. For example, a natural depression might produce a shadow in the raster image, but its LiDAR signature would show a gentle slope gradient, immediately distinguishing it from an artificial pit.
3. Case Study: Central Zhejiang Mining District
3.1 Geological Context
The study area is underlain by the lower Cretaceous Xishantou Formation (K1x⁴), which consists of volcanic tuff, andesite, and breccia. The mineralization, primarily vein-type fluorite, is structurally controlled by a network of faults and fractures. Historical mining from the 20th century had left behind a diffuse set of small-scale workings, but no detailed map existed. Our task was to use the integrated remote sensing data to locate these workings and interpret the structural framework.
3.2 Results of Feature Extraction and Validation
Using the interpretation keys defined in Table 2, I systematically analyzed the 3D model and high-resolution DEM for the four sub-areas. A total of 20 exploration pits, 1 large trench, and 2 adit portals were identified. To validate the interpretation, field teams were dispatched to 24 locations predicted by our remote sensing analysis. The results are shown in Table 3.
| Feature Type | Interpreted (Count) | Field Verified (Count) | Accuracy (%) |
|---|---|---|---|
| Exploration Pit | 20 | 17 | 85% |
| Trench | 1 | 1 | 100% |
| Adit Portal | 2 | 2 | 100% |
| Total | 23 | 20 | 87% |
Three interpreted pits were found to be natural sinkholes during field check, which were later filtered out by applying a stricter LiDAR-based roughness classifier. The overall accuracy of 87% confirms that our method is highly reliable for locating surface mine workings in dense vegetation.
3.3 Structural Interpretation and Ore Control
The spatial distribution of the verified surface workings provided a direct proxy for mineralized structures. By plotting their orientations on the LiDAR-derived DEM, three dominant fracture sets were recognized: (1) NE-SW trending, (2) NEE-SWW trending, and (3) NW-SE trending. These three fault systems appear to be the primary ore-hosting structures. The NE-trending set is the most prominent, with several aligned pits forming a linear zone over 800 m long. The NEE and NW sets are less continuous but show cross-cutting relationships that controlled the distribution of higher-grade lenses.
This pattern is entirely consistent with a fracture-controlled (epithermal) deposit model. The integration of the texture-rich orthophoto (which highlights the surface expression of these faults as linear valleys or scarps) with the LiDAR-derived micro-topography (which reveals subtle lineaments not visible in the imagery) allowed me to trace the faults beneath the soil cover. Using this combined approach, the accuracy of structural lineament interpretation improved by 38% compared to using either dataset alone.
4. Discussion: Synergistic Advantages and Technological Innovation
4.1 The “1+1>2” Effect of Data Fusion
The core finding of this study is that the fusion of oblique photogrammetry and airborne LiDAR yields a synergy far exceeding the sum of its parts. I have summarized this in a fundamental comparison of the three possible scenarios: using UAV drone photogrammetry alone, using LiDAR alone, and their fusion. The quantitative benefits are substantial.
Table 4: Performance Comparison of Three Technical Approaches
| Metric | UAV Drone Photogrammetry Only | Airborne LiDAR Only | Integrated Fusion (This Study) |
|---|---|---|---|
| Terrain Accuracy (RMSE) in Vegetated Areas | Very Poor (masked) | Excellent (0.08 m) | Excellent (0.06 m) |
| Structural Interpretation Accuracy | Baseline (60%) | Moderate (70%) | Significantly Improved (83%) |
| Lithology Discrimination | Good | Poor | Excellent |
| Point Cloud Density (ground only) | N/A | High (26 pts/m²) | High + texture attribute |
| Time for Data Processing | Fast | Moderate | Higher (10-20% more) |
The improvement in structural interpretation accuracy is particularly noteworthy. For this mined region, the LiDAR data alone allowed me to map the topographic lineaments, but could not help me decide which lineament corresponded to a mineralized fault versus a sterile joint. However, when I overlaid the oblique model, the color and texture of the rock (e.g., presence of fluorite float, oxidation staining, or quartz vein stockwork) enabled me to rank the structures by their mineralization potential.
4.2 Algorithmic Innovation: Self-Correction in Data Fusion
One practical challenge I faced was the systematic bias between the photogrammetric model and the LiDAR point cloud. The photogrammetric model sometimes drifts in elevation over long distances due to propagation of matching errors. To overcome this, I developed an iterative correction algorithm that uses the LiDAR ground points as a “rigid reference” to re-adjust the photogrammetric model. The process implements a variant of the Iterative Closest Point (ICP) algorithm, but applied here to correct the 3D mesh derived from images.
$$ E(R,t) = \sum_{i=1}^{N} \| (R \cdot p_i + t) – q_j \|^2 $$
where \( R \) is the rotation matrix, \( t \) is the translation vector, \( p_i \) are vertices of the photogrammetric mesh, and \( q_j \) are the corresponding LiDAR ground points (found by nearest-neighbor search). By minimizing \( E(R,t) \), I corrected the vertical drift in the photogrammetric model, improving its elevation accuracy from meter-level to centimeter-level (sub-0.1 m). This correction was critical for the subsequent high-precision structural analysis and volume calculations for the mining pits.
4.3 Future Directions and Intelligent Transformation
The successful application of this integrated method points toward a broader transformation in geoscience. We are moving from field-intensive “experience-driven” exploration to a “data-intelligent-driven” paradigm. The next logical step is to incorporate machine learning and deep learning into the interpretation pipeline. For instance, Convolutional Neural Networks (CNNs) trained on the fused dataset (RGB + LiDAR feature layers) can automatically identify key geological features such as fault scarps, breccia zones, and alteration halos. Furthermore, coupling the fusion data with large language models (like DeepSeek) could enable automated querying of 3D models, answering questions like “show me all NE-trending fractures in the andesite unit.”
However, challenges remain. The cost of high-end LiDAR systems (over millions of RMB) is still a barrier for many small projects. The processing of large volumes of data (point clouds + imagery) requires powerful computing resources and specialized software expertise. Additionally, the lack of a standardized national interpretation key library for geological features limits the portability of our method to new districts. I recommend that the geological survey community invest in building open-source interpretation databases, akin to those in the medical imaging field, to accelerate the adoption of this transformative technology.
5. Conclusion
This research demonstrates that the integrated application of UAV drone oblique photogrammetry and airborne LiDAR technology provides a robust and highly effective solution for geological exploration in complex, vegetated terrains. The key conclusions are:
- Synergistic Data Acquisition: The fusion of high-resolution texture from oblique imagery with sub-decimeter geometric accuracy from LiDAR creates a dataset that is far superior to either source used independently. This synergy enables the detection of subtle geological structures that would otherwise be invisible.
- Validated Interpretation Accuracy: In our Zhejiang case study, we achieved an 87% success rate in locating historical mining features (pits, trenches, adits). The derived structural patterns (NE, NEE, and NW trends) were confirmed by field mapping, validating the fracture-controlled model for ore deposition.
- Precision Modeling: By using LiDAR as a rigid reference to correct photogrammetric drift, we elevated the accuracy of terrain modeling from meter-level to centimeter-level, which is essential for high-resolution geological cross-sections and resource estimation.
- Technological Paradigm Shift: Our workflow signifies a tangible shift from traditional field-dependent exploration to a modern, data-centric approach. The combined use of these two remote sensing technologies reduces field time, improves safety by minimizing exposure to hazards, and provides a permanent, high-precision digital record of the study area.
In conclusion, the integrated use of these two technologies is not merely an incremental improvement but a fundamental leap forward for the field of exploration geology. I strongly advocate for the adoption of this methodology as a standard practice for regional mineral surveys, particularly in the forested and mountainous regions of China. As hardware costs decrease and intelligent processing algorithms mature, this approach will become the cornerstone of efficient, precise, and sustainable mineral resource exploration.