As a researcher deeply engaged in transportation engineering, I have long been committed to advancing the quality assessment of highway infrastructure. Subgrade compaction is a critical parameter that directly determines the bearing capacity, stability, and service life of highways. Traditional detection methods such as the ring knife method and sand replacement method are limited by sparse sampling points, low efficiency, and potential damage to the subgrade structure. To overcome these limitations, I have systematically investigated the application of China UAV aerial survey data for high-precision subgrade compaction detection. This paper presents my comprehensive study, including the analysis of influencing factors, detailed methodologies, and practical recommendations for accuracy enhancement.
China UAV technology has rapidly evolved and is now widely adopted in civil engineering. The advantages of unmanned aerial vehicles—flexibility, non-destructive data acquisition, and large-area coverage—make them ideal for subgrade compaction monitoring. In my research, I focus on four primary detection approaches: 3D modeling analysis, multispectral image interpretation, LiDAR fusion, and time-series data comparison. Each method has unique principles, strengths, and limitations, which I will elaborate on with supporting mathematical formulas and comparative tables.
Factors Affecting the Accuracy of China UAV Aerial Survey Data
The precision of subgrade compaction detection based on China UAV data is influenced by multiple factors across hardware, environmental conditions, and data processing algorithms. I have categorized these factors into three main groups and summarized them in Table 1.
| Category | Specific Factors | Impact Mechanism |
|---|---|---|
| Equipment Parameters | Sensor resolution, lens focal length, IMU accuracy, data transmission stability | High-resolution sensors capture fine texture; precise IMU ensures correct attitude; lens distortion affects geometric accuracy |
| Environmental Conditions | Wind speed, illumination, atmospheric visibility, temperature gradient | Wind causes image blur; strong light leads to overexposure; low visibility reduces contrast; temperature affects LiDAR beam refraction |
| Processing Algorithms | Image matching, point cloud generation, coordinate transformation, filtering threshold | Robust matching reduces mismatches; appropriate reconstruction preserves micro-topography; accurate georeferencing minimizes offset |
The IMU drift error can be modeled as a time-dependent function:
$$ \Delta \phi(t) = \alpha t + \beta t^2 + \epsilon $$
where $\Delta \phi(t)$ is the attitude angle drift at time $t$, $\alpha$ and $\beta$ are drift coefficients, and $\epsilon$ is random noise. Calibration procedures must correct these errors to maintain sub-centimeter accuracy.
Another critical factor is the ground control point (GCP) distribution. The coordinate transformation error follows:
$$ E_{\text{GCP}} = \sqrt{\frac{1}{n}\sum_{i=1}^{n}\left[(X_i – \hat{X}_i)^2 + (Y_i – \hat{Y}_i)^2 + (Z_i – \hat{Z}_i)^2\right]} $$
where $(X_i, Y_i, Z_i)$ are true coordinates and $(\hat{X}_i, \hat{Y}_i, \hat{Z}_i)$ are estimated coordinates. Insufficient GCPs increase $E_{\text{GCP}}$ significantly.
Detection Methods Based on China UAV Data
3D Modeling Analysis Method
Principle: I acquire high-overlap images using China UAV, then apply Structure from Motion (SfM) to generate dense point clouds. After mesh reconstruction and texture mapping, a high-precision 3D model of the subgrade surface is obtained. The compaction degree is correlated with micro-topographic parameters such as elevation variation rate, slope, and curvature. In well-compacted areas, the coefficient of variation (CV) of these parameters is low; in poorly compacted areas, the CV is high. I build a regression model:
$$ C(x, y) = a_0 + a_1 \cdot CV_{\text{elev}}(x, y) + a_2 \cdot CV_{\text{slope}}(x, y) + a_3 \cdot CV_{\text{curv}}(x, y) + \varepsilon $$
where $C(x,y)$ is the compaction degree at location $(x,y)$, $a_0, a_1, a_2, a_3$ are regression coefficients, and $\varepsilon$ is the residual.
| Aspect | Advantages | Limitations |
|---|---|---|
| Coverage | Full-area visualization, overcomes point-based limitation | Requires clear surface without vegetation or debris |
| Efficiency | Rapid automated modeling, short processing time | Accuracy influenced by image resolution and overlap |
| Applicability | Ideal for fill sections during construction, especially in arid regions | Less effective in high-vegetation or muddy conditions |
Applicable Scenarios: This method is best suited for subgrade surfaces that are flat and unobstructed. During the construction phase of urban roads or expressways, I can generate 3D models periodically to monitor deformation after each rolling pass and identify weak compaction zones.
Multispectral Image Interpretation Method
Principle: I mount a multispectral sensor on a China UAV to capture reflectance data across visible and near-infrared (NIR) bands. The compaction state alters soil pore structure and moisture content, thus affecting spectral reflectance. Normalized Difference Vegetation Index (NDVI) and moisture indices are used to map compaction:
$$ \text{NDVI} = \frac{R_{\text{NIR}} – R_{\text{Red}}}{R_{\text{NIR}} + R_{\text{Red}}} $$
$$ \text{NDWI} = \frac{R_{\text{Green}} – R_{\text{NIR}}}{R_{\text{Green}} + R_{\text{NIR}}} $$
I use a dual-camera multispectral imaging system (as shown in the figure below) that combines LED near-infrared and halogen white light sources, with a beam splitter and filters to simultaneously capture color and NIR images. The multispectral differences reflect compaction conditions. Machine learning algorithms such as Random Forest or Support Vector Machine are then applied to classify compaction levels.
| Material Type | NIR Reflectance (High Compaction) | NIR Reflectance (Low Compaction) |
|---|---|---|
| Loess | 0.25–0.30 | 0.35–0.42 |
| Silty Sand | 0.20–0.27 | 0.30–0.38 |
| Clay | 0.18–0.24 | 0.28–0.35 |
Advantages & Limitations: This method is non-contact and covers large areas rapidly. However, it is sensitive to surface moisture variation and works best for homogeneous subgrade soils. For mixed fills with gravel or construction debris, spectral signatures become ambiguous.
Applicable Scenarios: I recommend this method for early- to mid-stage subgrade construction on homogenous soils. It is particularly effective for expressway projects requiring rapid uniformity assessment over large areas.
LiDAR Fusion Method
Principle: A China UAV equipped with a LiDAR system emits laser pulses to obtain high-precision elevation data, which are then fused with simultaneously captured optical images. The LiDAR point cloud is unaffected by lighting conditions and can measure surface roughness and micro-topography with sub-centimeter accuracy. The optical image provides texture and color information. After spatial registration, I extract roughness index ($R_q$) and elevation standard deviation ($\sigma_z$) to build a multi-dimensional compaction evaluation model:
$$ C_{\text{LiDAR}} = b_0 + b_1 R_q + b_2 \sigma_z + b_3 I_{\text{return}} $$
where $I_{\text{return}}$ is the laser return intensity, which correlates with soil density. Studies show this method can detect compaction conditions within the top 5–10 cm of the subgrade surface.
| Parameter | Range for Well-Compacted | Range for Poorly Compacted |
|---|---|---|
| $R_q$ (mm) | 2.0–4.5 | 6.0–12.0 |
| $\sigma_z$ (mm) | 3.0–6.0 | 8.0–15.0 |
| $I_{\text{return}}$ (digital number) | 180–220 | 130–170 |
Advantages & Limitations: The LiDAR fusion method provides high accuracy and is robust to light variations. It can even penetrate sparse vegetation to reveal the true subgrade surface. The main limitation is higher equipment cost and more complex data processing.
Applicable Scenarios: I apply this method for critical sections such as bridge approach subgrades and high-fill embankments where structural safety is paramount. It is also valuable for mountainous highways with complex terrain and for post-construction quality acceptance.
Time-Series Data Comparison Method
Principle: I conduct repeated China UAV surveys over the same subgrade section at different time intervals. By comparing multi-temporal 3D models or elevation data, I calculate the surface settlement and settlement rate. Poorly compacted areas exhibit larger post-construction settlement. The relationship between settlement and compaction is modeled as:
$$ S(t) = S_{\infty}(1 – e^{-kt}) + \delta(t) $$
where $S(t)$ is cumulative settlement at time $t$, $S_{\infty}$ is ultimate settlement, $k$ is a consolidation coefficient, and $\delta(t)$ is random fluctuation. The compaction degree $C$ can be inversely related to $S_{\infty}$ via:
$$ C = c_0 + c_1 \cdot \frac{1}{S_{\infty}} + c_2 \cdot k $$
Fixed monitoring benchmarks are essential to eliminate systematic errors from environmental factors.
| Time Interval | Settlement Rate (mm/day) – Well Compacted | Settlement Rate (mm/day) – Poorly Compacted |
|---|---|---|
| 0–30 days | 0.1–0.3 | 0.5–1.2 |
| 30–90 days | 0.05–0.15 | 0.3–0.8 |
| 90–365 days | 0.01–0.05 | 0.1–0.4 |
Advantages & Limitations: This method provides dynamic insight into long-term compaction stability and can predict potential pavement distress. However, it requires multiple survey campaigns and careful alignment of data from different epochs.
Applicable Scenarios: I use this method for monitoring existing roads after opening to traffic, as well as for validation of new construction methods. It is particularly useful when evaluating the effectiveness of different rolling patterns.
Recommendations for Improving Detection Accuracy
Through extensive field experiments and data analysis, I have identified several key measures to enhance the precision of China UAV-based subgrade compaction detection. These are summarized below.
Optimize Data Acquisition Plan
I design the flight mission based on subgrade characteristics. The overlap rate between adjacent images should be at least 80% forward and 60% sideward to ensure robust 3D reconstruction. Flight altitude is set to achieve a ground sampling distance (GSD) of 2–3 cm. To mitigate wind effects, I limit flight operations to wind speeds below 5 m/s. The optimal time window is between 10:00 and 14:00 local time when solar elevation is high and shadows are minimal.
Standardize Data Processing Workflow
A standardized processing pipeline is critical. My workflow includes the following steps:
- Raw image correction: lens distortion and radiometric calibration.
- Multi-scale feature matching: using a combination of SIFT and deep-learning features.
- Point cloud filtering: statistical outlier removal followed by radius filtering.
- 3D reconstruction: selecting appropriate algorithm based on subgrade type (e.g., adaptive density for fill sections, edge-preserving for cut sections).
- Coordinate transformation: using GCPs with RMSE less than 5 cm.
The accuracy of the final 3D model can be quantified by:
$$ \text{RMSE}_{3D} = \sqrt{\frac{1}{N}\sum_{j=1}^{N}\left[ (X_j – X_j^{\text{ref}})^2 + (Y_j – Y_j^{\text{ref}})^2 + (Z_j – Z_j^{\text{ref}})^2 \right]} $$
In my practice, $\text{RMSE}_{3D}$ is consistently below 3 cm.
Strengthen System Calibration
Regular calibration of all sensing components ensures long-term accuracy. I perform monthly IMU calibration using static and dynamic tests. Camera radiometric calibration is conducted quarterly using a standard gray-scale panel. LiDAR range accuracy is verified weekly on a test field with known distances. Before each mission, a joint calibration of all sensors is performed on a calibration site to optimize exterior orientation parameters. All calibration records are archived for traceability.
Establish Data Quality Verification Mechanism
I implement a multi-level quality verification system. Internal verification includes checking reprojection errors (must be <1 pixel) and point cloud vertical accuracy (<3 cm). External verification uses ground checkpoints measured by total station or RTK GPS. The criteria are:
| Indicator | Acceptance Threshold |
|---|---|
| Plane position RMSE | ≤10 cm |
| Elevation RMSE | ≤5 cm |
| Compaction correlation coefficient (vs. conventional method) | ≥0.85 |
| Sampling ratio for verification | ≥2% of area |
Data failing to meet these thresholds are reprocessed or reacquired. This rigorous verification ensures the reliability of China UAV-based compaction detection in practical engineering.
Conclusion
My research demonstrates that China UAV aerial survey data provides a powerful, non-destructive, and efficient approach for highway subgrade compaction detection. The four methods—3D modeling, multispectral interpretation, LiDAR fusion, and time-series comparison—each have distinct strengths and are suitable for different engineering contexts. By optimizing data acquisition, standardizing processing workflows, performing regular system calibration, and implementing robust quality verification, the detection accuracy can reach centimeter-level and satisfy the stringent requirements of highway construction and maintenance. As China UAV technology continues to advance toward higher precision and greater intelligence, it will play an increasingly vital role in smart transportation infrastructure management, supporting the development of high-quality road networks across China.