Processing Multi-Spectral UAS Imagery

Introduction

  The images taken for this lab were done so using a Red Edge Sensor. A Red Edge Sensor is a camera which captures five distinct bands: Blue, Green, Red, Red Edge, and Near Infrared (NIR). A typical RGB sensor only captures three bands: Red, Green, and Blue. This makes the Red Edge Sensor very useful for added data analysis. Some of the important parameters for the Red Edge's camera lens include its focal length which is 5.5mm, its lens field of veiw which is 47.2° HFOV, its imager size which is 4.8 mm by 3.6 mm, and its imager resolution which is 1280 by 960 pixels. The Red Edge and NIR bands are extra bands which an average RGB sensor doesn't have. The proper order of these bands starting with band 1 is Blue, Green, Red, Red Edge,and NIR. The NIR, and Red Edge bands can be used to help create very precise quantitative data along with qualitative data. This lab will focus more on the qualitative analysis. Red Edge sensors are commonly used for vegetation health analysis. The purpose of this lab is to process UAS imagery and then to classify this imagery between permeable and impermeable. A series of maps will be created showing the false color NIR, false color Red Edge, RGB, and permeable versus impermeable surfaces. These maps will then be discussed as they relate to the Red Edge sensor.

Methods

Process the UAS Imagery
 First, the images had to be processed using Pix4D. A title was given to the project, 20160904_FallCreek, and the images from the lab folder were added.  Before processing the data, some of the usual options had to changed. These include choosing the Ag-Multispectral Processing template instead of the 3D model template, checking on GeoTiff, and GeTiff without transparency in the processing options. After the initial processes was complete, a quality report was generated shown below at left in figure 6.0. The quality check section shows that only 69% of all the images were calibrated. This is because the images were being taken as the Phantom was taking off. The area actually had good coverage by the Phantom as shown in figure 6.1 below at right in the overlap section of the quality report.

Fig 6.0: Quality Report for Initial Processing

Fig 6.1: Overlap Section in the Quality Report

  After reviewing this quality report. The data was ready for further processing. Once the Point Cloud and Mesh, and DSM, Orthomosaic and Index processing was complete there were a series of geoTiffs with the different bands specified (Blue, Red, Green, NIR, and RE) in the Fall Creek folder.

Create a Composite Image
  To create maps overlaying the different color bands, a composite image was generated. This was done by using the composite bands tool in ArcMap. For the input bands, all five of the bands are added in the following order: Blue, Green, Red Edge, and Near Infrared. The composite bands tool is used to create a new raster dataset with the bands aligned in a specific order. Once these bands are assigned a band number, they can be reordered and outputted through different band channels. This is how the RGB, False IR, and Falsue RE maps will be constructed.

Classify the Imagery Between Permeable and Impermeable
  Next, the image surfaces were classified between permeable and impermeable. This was done using a process very similar to the one used in the Using ArcPro and UAS data to Calculate Impervious Surface Area lab. To do this, first, the RGB composite had to be segmented using the Segment Mean Shift tool. This was used to generalize the image to help rid of unnecessary pixels. The segmented image is shown below in figure 6.2. This helps to make it easier to distinguish what surfaces are permeable and what surfaces are impermeable.

Fig 6.2: Segmented Image

   After that, the Training Sample Manager was used to create help classify the image. This was done by creating a series of rectangles on the roof, pavement, grass, shadows, and other areas and then labeling them appropriately. The use of the Training Sample Manager can be seen used below in figure 6.3. Once the rectangles were finished they were saved as a separate shapefile.

Fig 6.3: Using the Training Sample Manager

  Lastly, the Classify Raster tool was used to apply the training samples to the whole image. This would classify each part of the image between the categories specified in the training samples. Then, some of the categories were grouped together to create a impermeable and permeable surface distinction.

Results/Discussion

 Maps
  This first map shown below in figure 6.4 is the RGB composite. This map looks very much like an aerial orthomosaic. RGB is the way which our eyes see the spectral bands, so this image looks "normal". In the image there are four main features. The first is the road in the western part of the map. The second is the house and the driveway immediately surrounding it. The third is the brownish green vegetation south of the house. The fourth is the greener vegetation north of the house.

Fig 6.4: RGB Composite Map

 This next map in figure 6.5 shows a false color NIR of the area. This map has the bands arranged in the order: NIR, Red, and Green. Doing this makes the image a mix between red and blue hues. The healthier vegetation is represented by the darker shades of red, and the areas which have no vegetation are represented by the blue hues. This map does a good job of showing the density of vegetation. This can be seen by comparing the western edge of the map with the southern part of the map. In the western edge is a crop field full of crops and the southern part is just a grassy field which isn't very dense.

Fig 6.5: False NIR Map

  This map, shown below in figure 6.6, shows a false color Red Edge. It is very similar to the false NIR map in that it uses red and blue hues. However, the blue hue in this map is much more gray-like. The bands are arranged in the following order: Red Edge, Red, and Green. The patterns found in this map is identical to the ones found in the False NIR map.

Fig 6.6: False Red Edge Map

  This last map, shown below in figure 6.7, shows the permeable and impermeable surfaces. The permeable surfaces are shown in green and the impermeable surfaces are shown in light blue. Generally, the impermeable surfaces are where the roof, driveway, and road is, and the permeable areas are where the vegetation is. However, there are some impermeable surfaces embedded in the vegetation areas. This is because some of the pixels which were classified in the training samples were similar enough to some of the pixels in the composite to where it classified it as impermeable. One minor error in this map is that the outer area which surrounds the image. When classifying the the image, this was classified as impermeable. In future similar maps and analysis this area should be left out all together. For the purpose of this map, it can simply be ignored as the light blue was chosen so that it would match in with the background color, but be in contrast with the important permeable surface areas.

Fig 6.7: Permeable and Impermeable Surfaces Map

Red Edge Senor and Data Analysis
  These maps would not be possible if it were not for the Red Edge sensor. The Red Edge sensor allows for the capturing of 5 bands: Red, Green, Blue. NIR, and Red Edge. The two extra bands (NIR and RE) can be used to monitor vegetation health. This could be applied in a market setting to determine the health of a crop field and identify areas which need more water to grow healthier crops. Although only qualitative analysis is present in this lab, quantitative analysis is possible. An example of quantitative analysis would be to calculate the amount of impermeable surface area versus the amount of permeable surface area.

Conclusion

  In conclusion, the Red Edge sensor's abilities can be used for added data analysis by the process of capturing separate spectral bands. Besides the maps shown in this lab, the Red Edge sensor data can also be used to create a NDVI index which sees the water content in vegetation. Red Edge sensor data could be used by government agencies such as the DNR, Bureau of Land Management, Department of Agriculture, or by private businesses. In particular, the DNR would be more likely to be interested in calculating permeable and impermeable surface area. This could be used to see how much storm water flows into lakes and rivers which can fish and wildlife habitat, and water chemistry. Overall, the Red Edge sensor has the capability to collect important data regarding different color bands that the human eye cannot see which can be used for added data analysis by both private and public applications.

Sources
Red Edge User Manual PDF, Mica Sense
  https://drive.google.com/open?id=141-0wd2r80Q1T0u3oZiKIB_eaZHP8jMJ