At the first stage of an aerial survey project, in collecting the data, it is the flying time that costs money. The costs are determined for example by the size of the area to be surveyed, the number of flight lines needed to cover the area, and good enough light and weather conditions allowing flying.
Another stage of the remote sensing project where the cost-effectiveness is critical is the data processing. Because of huge amount of indirect mapping methods and manual work needed in data classification, that stage can constitute up to 70 % of costs of an aerial survey project, and therefore huge savings can be achieved.
Increasing automation of data classification
Is it possible to integrate hyperspectral data with LiDAR point cloud and increase automation of data classification? Sean Anklam, the President at Exogenesis, the provider of big geospatial data says yes, it is possible. And by doing that, it is possible to decrease the cost of data processing. Sean Anklam, a specialist in integrating and synthesizing big geospatial data at Exogenesis, and additionally National Intelligence Council Adviser for the Office of the Director of National Intelligence, is not questioning the significance of LiDAR and hyperspectral data fusion.
“I think it’s the Holy Grail of remote sensing. The two different disciplines of remote sensing complement one another so well and each fill a critical information shortfall. LiDAR provides you with intensity information and height and volume and texture whereas hyperspectral provides you with material composition, chemical composition and you are then fully able to describe an object, whether it’s a tree or a house or a road in terms of what is made of and its actual size. You can describe a tree’s canopy structure, height, stem spacing, leaf size, types of leaves, tree health etc.“
Sean has created an algorithm that transforms and fuses both datasets into a frequency and then performs an inverse transformation to produce polysynthetic dataset, the truly fused HSI & LiDAR point cloud, where each point has a chemical signature associated with it. “This is probably the most powerful way to express those dual capabilities because it enables you to do stream processing classifying things like tree types, fire fuels for forest fire modeling etc”, he says. “Within the oil and gas industry it has tremendous value for not only mapping oil and gas infrastructure, but detecting spills and leaks and other signs of aging in the infrastructure.”
Currently, the manual classification of LiDAR point cloud using RGBi imagery can take up to 70 % of the total aerial survey project cost just. “The way that expense gets diminished is through the hyperspectral-LiDAR fusion process where the new dataset enables a whole battery of automated stream processing tools.”
For forest tree species classification Exogenesis has created a tool called Random Forest. It generates the automated decision entry classifiers to ride across the LiDAR and hyperspectral data in concert with one another, and it produces tremendously accurate results. According to Sean the cost of data processing can be halved by using data fusion.
Decreasing the number of flight lines
Before the data can be processed, it needs to be collected. The most important condition that has to be met in order to fuse the two different data sets afterwards is that the both sensor technologies must be in the same platform; aircraft, helicopter or UAV. Both of those data sets need to be georeferenced and it is cheapest to do if both of those sensors share the same GNSS/IMU.
Flying is expensive and aerial survey data collection costs typically 1000-2000 Euros an hour. Sometimes bad weather can keep the aircraft on the ground for days. When it finally can fly, the number of flight hours needed is determined by the size of the area to be surveyed and the number of flight lines needed to cover the area. Often the surveyed areas are large and therefore the possibility to reduce the number strips is crucial in cutting the costs. “The addition of LiDAR to a hyperspectral collection dramatically reduces errors caused by shadows and topographic distortion thereby requiring fewer flight lines to be collected”, Sean says.
The full spectrum hyperspectral sensors usually have 384 spatial pixels, and that is the case for example with SPECIM’s AisaFENIX. To be able to cut the costs of flying even more by reducing the number of flight lines, and at the same time retaining the pixel size, the hyperspectral sensor should have significantly more spatial pixels. By using SPECIM’s new full spectrum imager, AisaFENIX 1K, which does have 1024 spatial pixels, the number of flight lines can be reduced by 60 %. For example, if previously 15 hours of flying was needed to cover a specific survey area, with AisaFENIX it can be covered in 6 hours.
“In addition to the cost reduction, by combining hyperspectral imaging with LiDAR surveying it is possible for survey providers to create entirely new data applications and therefore new business”, Sean points out.