HYPOD – Hyperspectral Oil Leak Detection for airborne pipeline monitoring

Hyperspectral remote sensing and image data analysis

Fraunhofer IOSB and its Scene Analysis department developed the methods for hyperspectral detection of oil in the HYPOD project. This involved the application of findings that had already been obtained during a measurement campaign in 2015 (see Lenz et al. 2015). 

However, other detection algorithms have also been tested for the reliable and automatic detection of oil. The basic principle of the detection method in hyperspectral remote sensing is as follows:

Ordinary (multispectral) cameras are sensitive in the range of visibly reflected solar radiation. More sophisticated cameras sometimes also record near-infrared or thermal signals. A hyperspectral camera, on the other hand, measures the entire wavelength range of visible, near-infrared and short-wave infrared light. The sensor not only has three or more channels available (as with a normal cell phone/RGB camera), but several hundred to a thousand channels. The resulting reflection spectrum forms a spectral signature that is specific to the object and material. Oil also has a specific reflection behavior. The difficulty of hyperspectral remote sensing lies in the fact that the mixing of oil and soil substrates/vegetation also overlaps the reflection properties of these materials. In addition, there are external factors such as atmospheric influences and distortions of the recorded images due to the flight movement of the flight platform.

In order to develop a reliable detection method, a test scenario was set up that was as close to reality as possible. For this purpose, several oil drip pans were filled with different soil substrates and mixed with three different types of crude oil and diesel. Some of these oil pans were used to simulate false alarms caused by plastic objects in order to provoke unintentional confusion with objects of similar material properties. For hyperspectral measurement from the air, Fraunhofer IOSB uses the latest hyperspectral sensor technology, whose ever decreasing weight now even allows installation on drones (e.g. DJI Matrice 600 Pro). Due to its low weight of 2.83 kg, the Hyperspec Co-Aligned VNIR-SWIR from Headwall enables a 12 to 18-minute flight per battery set, which corresponds to an area coverage of approx. 4.2 ha at a flight altitude of 80 m and results in an approximate pixel resolution of 4 cm.

First results

The preliminary results show that reliable detection of the three types of crude oil and diesel is possible with a low number of false alarms. False alarms can be suppressed in the surrounding area and within the oil catch pans with the plastic objects, while simultaneously preserving the good detection rate. Darker soil substrates and vegetation cover slightly reduced the detection performance. For this reason, machine learning approaches will be used in hyperspectral oil detection in future studies, which could additionally reduce the false alarm rate.