Spectral imaging from UAVs under varying illumination conditions T. Hakala1, E. Honkavaara1, H. Saari2, J. Mäkynen2, J. Kaivosoja3, L. Pesonen3, I. Pölönen4

1

Finnish Geodetic Institute, Department of Remote Sensing and Photogrammetry Technical Research Center of Finland, Photonic Devices and Measurement Solutions 3MTT Agrifood Research 4Department of Mathematical Information Tech., University of Jyväskylä 2VTT

Introduction • Objective: to investigate methods for quantitative radiometric processing of images taken under varying illumination conditions • Why: to expand the range of weather conditions during which successful imaging flights can be made. • Empirical study in a precision agriculture application using realistic data collected in difficult illumination conditions. • Light-weight Fabry-Perot interferometer filter based camera • Data cubes with adjustable spectral properties in a rectangular image format • Weigh 600 g -> suitable for light Unmanned Airborne Vehicles (UAVs) and light Manned Airborne Vehicles (MAVs) 2

UAV remote sensing using FPI spectral camera technology

3

Fabry-Perot interferometer based tuneable spectral imager, 2012 prototype • Hyperspectral imagery in frame format • Developed by VTT Technical Research Finland (Heikki Saari et al.)

• Spectral data cube by changing the width of Fabry-Perot air gap • A burst of images, each with different filter setting • Image size: 1024 x 648 pixels (2xbinned), Pixel 11 µm • C=10.9 mm, F-number < 3.0 • Spectral resolution 10-40 nm • Spectral range 500-900 (typical settings) 4

UAV operation UAV •Autopilot •IMU •GPS

Payload •Spectral imager •High spatial resolution imager •GPS •Irradiance sensors

Ground control station •Mission design and control •Insitu reference measurements: irradiance, reflectance targets,

•In typical flight 100-500 data cubes with 20-40 spectral layers •Georeferencing data •Irradiance data •Insitu data 5

Processing of FPI image data

6

Output products: hyperspectral image mosaics, DSMs, point clouds

0.6 2724 0.5

2512 2028

Reflectance

0.4

1497 0.3

994

0.2

632

Honkavaara, E., et. al. 2012. Hyperspectral reflectance and point clouds for precision agriculture by light weight UAV imasignatures aging system, ISPRS Ann. Photogramm. Remote Sens. Spatial Inf. Sci., I-7, 353-358, doi:10.5194/isprsannals-I-7-353-2012, 2012.

0.1 0 500

700 Wavelength (nm)

900

Empirical study

8

Vihti campaigns • MTT agricultural test area in Vihti • July 2, 2012 10:39 and 10:50 local time (UTC +3). • Poor illumination conditions with fluctuating levels of cloudiness • Flying altitude 140 m -> GSD of 14 cm • Flying speed: 3.5 m/s. • Block: five image strips and a total of 80 images; the forward and side overlaps were 78% and 67%, respectively • Ground control points, reflectance targets

9

Correction method based on insitu irradiance measurement • Simplified equation Ljc(λ)at_sensor = Lj(λ)at_sensor Cj(λ)

• Correction factor: Cj(λ) = Eref(λ)/Ej(λ) • Ej(λ) irradiance measured during image j • Eref(λ) irradiance measured during reference image

• Two different reference instruments • ASD FieldSpec Pro with irradiance optics • Spectral measurement 3 nm FWHM @ 350-1000 nm

• Onboard irradiance sensors on the UAV • Broadband 400-1000 nm

10

A block adjustment based method for reflectance image generation of frame images • Data • Overlapping spectral rectangle format data cubes

• Tasks • Eliminate radiometric disturbances caused by sensor instability and illumination/atmosphere • BRDF compensation • Reflectance calibration

• Approach • Radiometric model parameters using radiometric block adustment with a network of radiometric tie points • Optional insitu irradiance measurements • Reflectance images using reflectance targets Light source

Observer θi

θr

C G E

D

North

A A

φr

B F O

φi

11

Results

12

Image data

SNR 120

80

SNR

• A total of 42 spectral layers in the original raw data • 30 smile-corrected spectral layers.

100

60 40 20 0 500

550

600

650

700

750

800

850

Wavelength (nm)

1 0.8 0.6 0.4 0.2 0 450

500

550

600

650 700 750 Wavelength (nm)

800

850

900

950

13

900

Irradiance measurements Wide-band irradiance from UAV Relative irradiance 9000 8000 7000 6000 5000 4000 3000 2000 1000 0 Image

Spectral irradiance on ground 7

Relative irradiance

16 29

8.00E-01 7.00E-01 6.00E-01 5.00E-01 4.00E-01 3.00E-01 2.00E-01 1.00E-01 0.00E+00 Image

14

Correction factors Correction factors for each image

• The UAV based correction factor had a dependence on the flying direction • The factor based on radiometric block adjustment showed slight drift behavior. • Average factor was dependent on object.

3

ground, 29 uav

2.5 BA: relA, 29 average, 29

2 1.5 1 0.5 0

Image

Difference to ground irradiance measurement %

50

BA: relA, 29

40

average, 29

30

uav, 29

20 10 0 -10 -20 -30 -40 -50

Image

15

Image mosaics and sample spectra No corr

Ground

UAV

Image based

16

Coefficient of variation • Average coefficient of variation in tie points • Without correction 0.14-0.18 • UAV: 0.1-0.12 • Ground: 0.06-0.09 • Block adjustment: 0.04-0.07

Coefficient of variation

no corr

0.2 0.18 0.16 0.14 0.12 0.1 0.08 0.06 0.04 0.02 0

ground

uav BA: relA

500

600

700

800

900

Wavelength (nm)

17

Biomass estimation by knn-estimator No correction R2: 0.64 NRMSE: 24.9%

Ground irradiance R2: 0.74 NRMSE: 17.8%

Block adjustment R2: 0.74 NRMSE: 16.8%

18

Conclusions • Fabry Perot Interferometer camera is very promising technology • High spatial resolution, stereoscopic, spectrometric image data • FPI camera is operational from UAV platforms, suitable also for MAV platform • Well suited for time-critical and monitoring applications, such as water quality, agriculture, mining environments, disasters

• Radiometric processing technology for images collected in diverse weather conditions is needed, radiometric aspects need to be carefully considered • A new method based on insitu measurement of irradiance either from UAV platform or from ground was developed. • Method was tested in a precision agriculture application using realistic data collected in difficult illumination conditions. • Results were very promising, indicating that quantitative UAV based remote sensing could be operational in diverse conditions, which is prerequisite for many environmental remote sensing applications.

19

Thank you!

20