With the development of global economies, fixed-wing surveying drones have seen increasingly widespread applications across civil and military domains. These surveying UAVs offer efficient, safe, and environmentally friendly inspection capabilities, becoming integral to industries like power, oil, and natural gas. However, existing fixed-wing drones face significant challenges during large-scale inspection missions, including limited takeoff/landing sites and restricted operational ranges. These constraints severely impact inspection efficiency and coverage area. To address these limitations, this study proposes an innovative solution enabling cross-site operations for surveying UAVs.
Our approach integrates several key technologies to achieve robust autonomous operations. The core innovation lies in combining Fourier transform-based gradient direction histograms with active disturbance rejection control (ADRC) algorithms within an embedded Debian OS environment. This enables the surveying drone to handle cross-site operations while maintaining stability under challenging wind conditions. The hardware backbone features a Samsung S5P6818 octa-core Cortex-A53 processor with Mali-400 graphics engine, providing the computational power necessary for real-time image processing and flight control.
System Requirements Analysis
The operational demands for modern surveying UAVs necessitate solutions that overcome three fundamental limitations: cross-site takeoff/landing capability, extended operational range, and intelligent target recognition. Traditional VTOL (Vertical Take-Off and Landing) fixed-wing drones increase system complexity, leading to higher failure rates and maintenance costs. Battery replacement requirements further disrupt continuous operation during large-area inspections. Our analysis identified these critical requirements for next-generation surveying drones:
| Requirement Category | Technical Specifications | Performance Target |
|---|---|---|
| Takeoff/Landing | Runway independence, wind resistance | Operation in 10m/s crosswinds |
| Navigation | Autonomous path planning | 200km range per mission |
| Target Detection | Small object recognition | ≤5cm objects at 100m altitude |
| Data Processing | Real-time image analysis | <500ms processing latency |
The proposed surveying UAV system addresses these requirements through a combination of adaptive control algorithms and computer vision techniques. Key innovations include a probabilistic target proposal (FPTP) method for candidate target extraction and Fourier-HOG (FHOG) features for rotational-invariant target confirmation.
Overall System Architecture
The surveying drone system comprises airborne and ground control segments. The airborne system features a distributed architecture with the following components:
$$ \text{System}_{\text{airborne}} = \sum_{i=1}^{n} (\text{Sensor}_i + \text{Processing}_i + \text{Control}_i) $$
Ground control stations communicate via 4G/5G networks and LoRa (using SX1262 chips) for long-range command transmission. The navigation system employs a multi-loop control structure with the following path planning logic:
| Flight Phase | Control Parameters | Navigation Method |
|---|---|---|
| Takeoff | δr = Kφ(φr – φ) + KiΔS | Runway alignment |
| Cruise | Waypoint tracking | GPS/INS fusion |
| Landing | θ = e-l/4θr | Flare path optimization |
Where δr represents rudder deflection, φ denotes heading angle, ΔS indicates lateral deviation, and θ defines pitch angle during landing descent. The landing control strategy employs exponential pitch adjustment relative to altitude (l) with reference pitch θr = 5.5° initiating at lr = 3.5m.
Hardware Architecture
The surveying UAV’s hardware platform integrates specialized components for autonomous operation:
| Subsystem | Components | Specifications |
|---|---|---|
| Processing | S5P6818 Cortex-A53 | Octa-core, Mali-400 GPU |
| Sensing | DJI Zenmuse X5S, FLIR T640, Ouster OS1-16 | 20.4MP/4K, 640×480 IR, 16-beam LiDAR |
| Communication | SX1262 LoRa, 4G/5G | 10km range, 150kbps |
| Ground Station | Intel i9-10900K, 32GB RAM | 10-core, 5.3GHz |
Power management employs dynamic voltage scaling to optimize energy consumption during different flight phases. The mechanical design incorporates carbon-fiber composites, resulting in the following prototype specifications:
$$ \text{Mass}_{\text{takeoff}} = 6\text{kg}, \quad \text{Wingspan} = 2.2\text{m}, \quad \text{Aspect Ratio} = 5.8 $$
Control Algorithm Design
Wind disturbance rejection during takeoff/landing utilizes a novel Active Disturbance Rejection Controller (ADRC). The controller structure consists of three components:
$$ \begin{align*}
\text{Tracking Differentiator (TD):} & \quad \dot{x}_1 = x_2, \quad \dot{x}_2 = r \cdot \sin\left( \frac{x_1 + v(t) – x_2^2}{r} \right) \\
\text{Extended State Observer (ESO):} & \quad \dot{a}_1 = a_2 – \alpha_1 f(e,z_1,l) \\
& \quad \dot{a}_2 = a_3 – \alpha_2 f(e,z_2,l) \\
& \quad \dot{a}_3 = – \alpha_3 f(e,z_3,l) \\
\text{Nonlinear Feedback (NLF):} & \quad u = \frac{u_0 – a_3}{b}
\end{align*} $$
Where v(t) is the reference signal, x1/x2 are TD outputs, a1-3 represent ESO states, and αi are adjustable gains. The compensation factor b is derived from system dynamics, while c and l are fine-tuned parameters. This structure enables the surveying UAV to maintain ±0.5m positioning accuracy in 10m/s crosswinds.
Automatic Inspection Software
The vision processing pipeline for the surveying drone employs a multi-stage approach for small target detection:
- Candidate extraction using Feature-based Probabilistic Target Proposal (FPTP)
- Rotation-invariant Fourier-HOG feature extraction
- Target confirmation via support vector machine classification
The FHOG transformation provides rotational invariance through polar coordinates. For pixel gradient d with direction ∇f(d), the distribution function h is expressed as:
$$ h(\theta) = \|l\| \delta(\theta – \nabla f(d)) $$
After convolution with spatial aggregation kernel P1 and normalization kernel P2, the Fourier coefficient transformation maintains rotational consistency:
$$ \widetilde{F}_m = \hat{f} \frac{P_1}{\|L\|^2} P_2 $$
When rotating FHOG features Wk,m by angle θ, the transformed features become:
$$ W’_{k,m} = e^{-i\theta(m-k)} W_{k,m} $$
This enables consistent target recognition regardless of surveying UAV orientation. The complete detection process achieves 94.2% accuracy for 5cm targets at 100m altitude.
System Implementation and Testing
Field testing involved multiple networked nests across a 50km2 operational area. The surveying drone completed 37 autonomous sorties with the following mission profile:
| Parameter | Takeoff Phase | Cruise Phase | Landing Phase |
|---|---|---|---|
| Duration | 108s (Fig 11a) | 42min average | 78s average |
| Altitude | 0-50m | 100-150m | 50-0m |
| Speed | 15-25m/s | 30m/s | 12-0m/s |
X-Plane simulations validated the control algorithms under various wind conditions. The simulated flight path incorporated 7 navigation waypoints and 3 inspection targets, demonstrating complete autonomous operation from takeoff to landing.
Performance Analysis
Comparative evaluation against existing methods demonstrated significant improvements in surveying UAV performance:
| Metric | Proposed System | Reference [1] | Reference [2] |
|---|---|---|---|
| Positioning Error (m) | 0.12 ± 0.02 | 0.35 ± 0.06 | 0.29 ± 0.07 |
| Inspection Efficiency | 94.2% | 84.2% | 73.7% |
| Coverage Area (km2/h) | 18.7 | 15.2 | 13.1 |
| Wind Resistance (m/s) | 12.5 | 8.2 | 9.1 |
The overall operational efficiency and coverage showed an 11.9% average improvement over existing solutions. Additional robustness metrics confirmed system reliability:
$$ \text{EMI Immunity} = 10\text{V/m}, \quad \text{ESD Protection} = 4\text{kV}, \quad \text{Fault Recovery} < 5\text{s} $$
Small target detection accuracy remained above 92% across lighting conditions and altitudes, enabling effective identification of infrastructure defects during power line inspections.
Conclusion
This research presents a comprehensive solution for large-scale autonomous inspection using fixed-wing surveying UAVs. The integration of ADRC control with Fourier-based vision processing enables reliable cross-site operations in challenging environments. Key achievements include:
- Successful demonstration of autonomous takeoff/landing at unprepared sites
- 11.9% improvement in operational efficiency over existing systems
- Robust performance in winds up to 12.5m/s
- Accurate detection of sub-5cm targets from 100m altitude
The implemented technologies provide a foundation for widespread adoption of surveying drones in energy infrastructure monitoring, environmental surveying, and disaster assessment. Future work will focus on swarm coordination and AI-based predictive maintenance using the collected inspection data.