Two types of UAV platform were used within the STARS project. These were (1) the Sensefly eBee and (2) the Geokonzept Octocopter Geo-X 8000. The first is a fixed-wing system, the latter is an 8-engines helicopter model. Both models allow mounting different cameras. We flew the eBee with an RGB camera but also with a multispectral camera. The Geo-X 8000 was flown with an RGB, a multispectral and a thermal camera. Obviously, these camera uses were determined by purpose: for instance, soil water content was important in Bangladesh and therefore a thermal cammera was mounted on the Geo-X 8000.
For these platforms, imaging capabilities obviously depend on the mounted camera. Cameras capture data in different portions of the electromagnetic spectrum. This ranges from only in the visible portion (red, blue, green), to visible-plus-NIR (multi-spectral) and the thermal portion of the EM. Hyperspectral cameras also exist and record data in a wider range of spectral channels; these were not used in our project.
In the sections below, we share our experiences in operating the two platforms and a range of cameras in different farming environments around the world.
First, we introduce the different UAV platforms/cameras that we used:
Tetracam MiniMCA (Multispectral Camera Array, here with the sixth channel used for an incident light sensor)
Canon S110 RGB / NIR
Canon S110 NIR
Canon S110 NIR and S110 RGB/NIR
These two cameras were used on-board of the eBee platform. Whilst visual inrterpretation with RBG data is fairly straightforward, that with images from the NIR (Near Infrared) camera is difficult, but image segmentation, e.g., using eCognition, works quite well with this band. It is easier to identify features such as trees on DEMs generated with images from an RGB (red, blue, green) camera than from those from a NIR camera. For this reason, it is advisable to fly the NIR and RGB together, from which visual interpretation can be made with the RGB while taking advantage of the NIR band for image segmentation and other processing tasks.
The multiSPEC 4C camera, flown on-board of the eBee, has a spatial resolution of 1.2 MP (megapixel) in four separate sensors: Green (550nm), Red (660nm), Red-Edge (735nm) and NIR (790nm) (Nebiker et al., 2016). The spatial resolution is 10 times less than that of the S110 NIR. The multiSPEC 4C is heavier than the S110, hence the allowable flight time decreases. Consequently, the number of flights required to cover the same area increases compared to that of the S110. The lenses of the camera are constantly open and unprotected and, therefore, scratch easily when landing on rough terrain. This requires great caution in choosing a landing location. The first version of the camera had a significant overheating problem, causing it to malfunction and not take pictures (Nebiker et al., 2016). This problem also frequently caused the eBee itself to veer off-course in an uncontrolled manner. This was fixed in the new version of the camera which came out in the third quarter of 2015. Old cameras were refurbished by the manufacturer for free. Nonetheless, excessive exposure to sunlight prior to flight could still cause overheating. When fast-moving clouds are present, the on-board calibration via the Incident Light Sensor (ILS) becomes unreliable, because of the large variation of sunlit and shaded areas.
Geo-X 8000 with Tetracam miniMCA
The Tetracam miniMCA is an array of six separate cameras, used on-board of the Geo-X 8000 octocopter. Each camera in the array measures in a small spectral band, and thus the array can measure six different spectra. We used the hardware version as purchased in mid-2014, which was presumably improved over earlier versions. The Bangladesh team operated two X 8000s, and the W and E Africa teams each one. The camera allows to choose one’s own spectral band combination, as these are determined by filters that one can place in each camera. In the case of STARS, one of the six available camera slots was sacrifized for an incident light sensor, which allows the cameras to calibrate registered reflection values in-flight.
With help of the Tetracam firm, we conducted a camera filter swap. The purpose of this swap was to obtain a spectral coverage that closely mimicked, and could be seen as a compromise of, the multispectral channels of WorldView and RapidEye satellite images. The cameras operated in Africa were equipped with narrow-band filters at 550, 680, 710, 740, and 800 nm, with the latter used as master channel. The Bangladesh miniMCA had the 550 nm filter replaced by a 530 nm filter, for reasons of the prominence of water surface studies in their project. The filter swap though doable, requires careful engineering, image alignment and calibration; our experiences foun their way in a separate report.
The Geo-X 8000 is a fairly heavy machine, which brings the advantage of in-flight positional stability resulting in sharp images. The flight time with octocopter platform is, however, relatively short and the area covered is much smaller than that of the eBee. In addition, its set-up and maintenance is much more involved than the eBee. The system is more sensitive to damage and wear, and we suffered from a few (minor) crashes that brought damages to the machine. The flight planning software is rather rudimentary and needs fast internet access to do any kind of planning. Pre-caching the imagery for use in the field without internet access is not possible. Lack of an on-board GPS, at least on the systems that we purchased, is a limitation and does not allow automatic geo-tagging of the acquired photographs. There is no direct communication between the on-board GPS and the camera. This requires laying out ground control points, which is not always practical, and this certainly slows down field procedures.
Especially for flight missions involving multispectral cameras, one is advised to take precautionary action into the spectral calibration challenge. As missions will typically be days or weeks apart, different light conditions will be the norm, and one wants to address these differences in the best way possible to allow spectral reflectance comparisons between images. To this end, we developed calibration panels, made from aluminium and painted in four grey tones, to be put out and imaged in the field with each flight mission. Each of our panels measured 75´75 cm, giving a set of four panels for each field team. These panels had been professionally spray-painted with NEXTEL paint, produced by Mankiewicz, Germany. We used the colors black (5% reflectance), granite gray (20%), steel gray (40%), and a self-constructed “pale gray” from a mix of 3 volume parts white and 2 volume parts steel gray (giving a 80% reflectance). Pre-gflight installation of the panels needs to be done with care, as angular effects may arise from non-flat positioning. A full report can be found here.