PIMSIS detects TPI by comparing detected events in an EO/SAR reference image with detected events in a new acquired EO/SAR image. For this we did implement a processing chain with the following steps: - The construction of a reference image starts by retrieving from the image store of the satellite provider one or more images covering the complete pipeline route to be monitored. The images are pre-processed to correct for differences in terrain elevation and differences in satellite orbits. Then the actual reference image is constructed by cutting a strip around the route of the pipeline to be monitored. The reference image can be based on one or more images acquired at a specific date but may also, for a stable area, be based on images acquired at 2 or 3 successive dates (for suppression of noise). Additional the following processing steps are applied to the reference image.
- Using phase signatures manmade structures are separated from the natural environment.
- Manmade structures are detected (target detection).
- When after a revisit by the satellite new images of the pipeline route become available these images are retrieved from the image store of the satellite provider. The same pre-processing and processing steps are applied as described above. Subsequently the detected targets in the new acquired images and in the reference image are compared. Also a second detection algorithm is applied based on the coherence between the new acquired image and the reference image.
- Than the differences in detected targets (i.e. appearances or disappearances of events) are sent as notifications via a secure internet connection to a pipeline integrity management system at the site of our clients. The notifications have a location and a quality measure, which describes the quality of the detected changes.
- The new acquired images becomes (part of) the reference image and we continue to step 2 of the processing chain.
The complete PIMSiS processing chain as described above operates in a fully automatic manner, including a number of quality controls to test intermediate results and to optimise specific parameters of the algorithms applied to the data. Target satellitesFor our PIMSiS services we use EO/SAR satellites or constellations of EO/SAR satellites with the following characteristics: - Wide swath (i.e. the width of the image).
- Reasonable geographic resolution.
For our PIMSiS services we initially selected as target satellites the constellation of the two ESA Sentinel-1 satellites (Sentinel-1A and -1B) which fulfil these requirements in combination with fast data delivery and free access to data. ESA will launch Sentinel-1A in April 2014 and Sentinel-1B in the second half of 2015.
Currently we are expanding our processing chain from data sourcing from a single EO/SAR platform to data sourcing from multiple, different EO/SAR platforms (including the constellations of TerraSAR-X and COSMO-SkyMed satellites and the RadarSAT-2 satellite). This new processing chain will also allow us to use both ascending and descending orbits of EO/SAR satellites. With the new processing chain we can dramatically increase the update frequency of the notifications for Third Party Interferences and also offer a flexibility in resolution and costs for satellite data. Results from development phaseThe test site presented here was located in Monster, south of The Hague and close to the North Sea shore. The Monster test site is an area for urban development, due to the financial crisis in 2013, it is at the time of testing without any construction activity. The soil is sand (see Figure 1). We used data acquired in standard mode by the EO/SAR satellite RadarSAT-2. The standard mode of RadarSAT-2 has very similar specifications as the target satellite constellation Sentinel-1. For every revisit of the RadarSAT-2 satellite we did simulate at these test sites all kind of Third Party Interference events. The acquired satellite images were used to develop and optimise our processing chain. Figure 2 is the ground truth map for the Monster test site for the revisit of the RadarSAT-2 satellite on 01/09/2013. After applying our processing chain we did detect the TPI targets (Figure 3). The objects are represented by the red pins in the image. By studying successive images we were able to detect changes (i.e. appearing or disappearing of TPI targets). It is to be noted that the PIMSiS processing chain we can separate manmade objects from responses from the natural environment. Figure 1: Monster test site 1Figure 2: Ground truth map for 01/09/2013Figure 3: Automated target detection 01/09/2013 |