Analysis of the application of an advanced classifier algorithm to ultra-high resolution unmanned aerial aircraft imagery - a neural network approach

Mapping the existing land use is the essential activity in the management of an area, especially in densely urbanized areas. Knowing the development, amount, and extent of specific land use will be very helpful in management activities. The availability of geospatial data acquisition technology such as unmanned aerial systems (UAS) is currently beneficial for monitoring and inventory activities. Geospatial data with ultra-high resolution are now easily obtained using UAS. This study evaluated the performance of advanced classification algorithms on ultra-high-resolution UAS aerial imagery data based on the different number of regions of interest (ROIs) with two different algorithms, namely, Multilayer Perceptron (MLP) and Radial Basis Function Neural Network (RBFNN). Evaluation was carried out regarding both the performance of computing time and accuracy. The final result shows that the number of ROIs affects the results of classification accuracy as well as the computing time. The MLP algorithm provides inconsistent accuracy but fast computing time, while the RBFNN algorithm provides consistent accuracy with slow computing time. The MLP algorithm is suitable if the researcher prioritizes the computational speed performance, but if the researcher prioritizes the accuracy, the RBFNN algorithm is the best choice.