TL;DR
This paper tackles the challenge of accurately measuring the backscatter intensity of hyperspectral LiDAR (HSL) by developing a new radiometric calibration model that accounts for the effect of the laser's incident angle. This Lambertian-Beckmann model significantly improves calibration accuracy, especially for surfaces with specular reflection, making HSL data more reliable for applications like target classification.
What Problem Does This Paper Address?
Hyperspectral LiDAR (HSL) is a powerful remote sensing technology that captures both spatial and spectral information, offering a "colorful" point cloud that traditional LiDAR lacks. However, the intensity of the backscattered laser signal, a crucial piece of information for identifying target attributes, is heavily influenced by various factors, including the angle at which the laser beam strikes the target (the incident angle). This "incident angle effect" can lead to inaccurate measurements, especially for surfaces that exhibit specular reflection (like glossy surfaces). Existing calibration methods, particularly those relying solely on the Lambertian model (which assumes diffuse reflection), are insufficient for accurately correcting this effect across diverse natural surfaces. This paper aims to address this gap by proposing a more comprehensive radiometric calibration method for HSL that specifically targets the incident angle effect, improving the reliability and usability of HSL data.
Methods
The authors propose an adaptive threshold Lambertian-Beckmann model for radiometric calibration. This model is an extension of the traditional Lambertian model, incorporating the Beckmann model to account for specular reflection. The core idea is that natural surfaces are a combination of diffuse and specular reflection.
Here's a breakdown of their approach:
- Lambertian-Beckmann Model: This model combines the Lambertian model (for diffuse reflection) and the Beckmann model (for specular reflection) using a linear combination. They introduce a diffuse reflection coefficient ($k_d(\lambda)$) and a specular reflection coefficient ($k_s(\lambda)$), where $k_d(\lambda) + k_s(\lambda) = 1$.
- Piecewise Equation: The model is further refined into a piecewise equation that considers an angle threshold ($\theta_T$). When the incident angle is below this threshold, both diffuse and specular reflections are considered. Above the threshold, only diffuse reflection is assumed to be significant.
- Parameter Estimation: They use nonlinear curve fitting (least squares method) to estimate the model parameters ($k_d(\lambda)$, $m$ (roughness parameter), and $\theta_T$) for each wavelength and sample.
- Adaptive Threshold Calibration: An adaptive threshold radiometric calibration method is designed based on the fitted parameters to correct the backscatter intensity. This method aims to correct the intensity to a "normal direction" equivalent, making it independent of the incident angle.
Datasets
The study utilized a variety of experimental samples collected under controlled laboratory conditions. These samples were chosen to represent a range of surface characteristics, including:
- Standard Reference Targets: Three targets with different reflectance values (99%, 70%, 40%).
- Common Materials: Wood product, sidewalk brick, white floor tiles (ceramics), marble tiles, and a car shell sample.
- Plant Leaf Samples: Six types of common plant leaves from Beijing (Fraxinus pennsylvanica, yellow leaf of Fraxinus pennsylvanica, Eucommia ulmoides, Magnolia denudate, Ficus elastic, and Codiaeum variegatum).
The HSL system used in the study had 26 wavelength channels ranging from 650 nm to 900 nm with a 10 nm step.
Key Insights and Ideas
- Natural Surfaces are Complex: The study reinforces that natural surfaces are rarely purely diffuse reflectors. The interplay of diffuse and specular reflection is crucial for accurate intensity measurements.
- Specular Reflection Matters: Glossy surfaces, like floor tiles and car shells, exhibit significant specular reflection that the Lambertian model alone cannot handle. The Beckmann model component is essential here.
- Green Leaves and Specular Reflection: An interesting finding is that green leaves show a noticeable specular reflection effect in the visible wavelength range (around 650–720 nm), potentially linked to chlorophyll's internal microstructure. This highlights the need for wavelength-specific calibration.
- Angle Threshold is Key: The concept of an angle threshold ($\theta_T$) is vital. It acknowledges that at larger incident angles, the specular component might become negligible or undetectable, simplifying the model.
- Improved Accuracy: The proposed Lambertian-Beckmann model offers a significant improvement in radiometric calibration accuracy compared to the Lambertian model, particularly for surfaces with specular characteristics.
Key Results
- Calibration Accuracy Improvement: The Lambertian-Beckmann model achieved an average improvement rate of 22.67% for the tested samples compared to the Lambertian model.
- Significant Improvement for Specular Surfaces: For samples with obvious specular reflection (e.g., car shell, floor tile), the improvement was even more pronounced. The maximum improvement rates were 88.97% for the car shell sample and 61.90% for Eucommia ulmoides.
- Correction for Larger Angles: When comparing to results before calibration with incidence angles less than 70°, the Lambertian-Beckmann model showed an average improvement rate of 62.26%.
- Reduced Reflectance Standard Deviation: Table 3 demonstrates that the mean of reflectance standard deviations across wavelengths was significantly reduced after calibration with the Lambertian-Beckmann model, indicating more stable and reliable measurements. For instance, the floor tile showed a reduction from 0.326 (before calibration) to 0.0486 (after Lambertian-Beckmann calibration).
- Leaf Specifics: The model proved effective for leaf samples, with the Lambertian-Beckmann model outperforming the Lambertian model, especially in the visible wavelength range where specular reflection is more prominent.
Limitations and Future Work
- Complex Scenes: The current research focused on simple targets and leaves. The authors acknowledge that applying the method to more complex scanning scenes, such as forests with trees, will be challenging, particularly in obtaining accurate incident angles.
- Multiple Scattering: The influence of multiple scattering, which can affect backscatter intensity, was not explicitly addressed and warrants further investigation.
- Surface Roughness and Grain Size: For targets with no specular reflection but significant roughness or grain size variations, their reflection characteristics might not perfectly follow the Lambertian model, requiring further study.
- Instrument System and Environment: The paper notes that backscatter intensity is influenced by instrument system characteristics, distance, and environmental factors. Future work should aim to account for these influences to obtain more accurate backscatter intensity information.
- Incidence Angle Range: While the model performs well, the authors observed that calibration effects worsen for incidence angles greater than 60° due to changes in laser footprint size. They suggest that in practical applications, Airborne LiDAR incidence angles are generally kept below 60°.
Summary updated