Background
The two sensors used in an unmanned aerial vehicles
Active and passive sensors
This machine-learning method combines many decision trees to classify imagery
Random Forest
Using only these three image bands limited the amount of spectral information available
Red, green and blue bands (RGB)
These aerial platforms allowed researchers to collect very high-resolution imagery at relatively low cost
UAVs (drones)
The definition of UAV remote sensing
The use of Unmanned Aerial Vehicles fitted with sensors to collect spatial, spectral, geophysical, or atmospheric data
This type of UAV imagery uses red, green, and blue bands but no near-infrared band
RGB imagery
These image characteristics significantly improved vegetation classification when combined with RGB data
texture features
Texture analysis helped compensate for the lack of this type of information in RGB imagery
spectral information
The following is an example of a passive UAV sensor
Multispectral camera
This was the approximate spatial resolution of the UAV imagery
7 cm per pixel
These window sizes produced the highest classification accuracy because they balanced detail and contextual information
medium-sized texture windows
These were the only two areas classified in the study
Recreational park and residential area
This generally happens when UAV flight altitude increases
The area covered increases but detail/resolution decreases
This software was used to register, orthorectify, and mosaic the UAV images
Pix4D
Random Forest performed similarly to this more complex image classification approach
Object-Based Image Analysis (OBIA)
Despite good results, RGB imagery still lacks this compared with multispectral imagery
spectral richness
Ground Sampling Distance (GSD) describes
How large each pixel is on the ground
The researchers used this many training pixels for each land-cover class
500
Compared with Random Forest, this traditional classifier produced lower classification accuracy
Maximum Likelihood Classification (MLC)
The lack of this band in the off-the-shelf RGB camera reduced accuracy in land cover classifications
Near-infrared