Determining Ideal Parameters in Object Based Classification for Multiband UAS Imagery

Abstract

As Unmanned Aerial Systems (UAS) are applied to an ever-increasing array of remote sensing applications, a greater value is being placed on added data analysis including object based classification. Imagery captured using a UAS platform has very high spatial resolution imagery (5.67cm and 3.97cm in this study) compared to the spatial resolution of traditional piloted aircraft (~1 m) and satellite imagery (30 m). This research investigates the ideal parameters in performing object based classification on UAS captured imagery. The imagery in this project was captured using a Red Edge sensor. The figure below shows the spectral resolution of the imagery collected by the Red Edge Sensor while the figure above shows what the sensor looks like itself. Many steps are needed to perform object based classification, and therefore ideal parameter selection is essential. The main steps include segmenting imagery, collecting and refining training samples, classifying the imagery, and assessing output accuracy. These parameters are beneficial to potential real-world applications, such as vegetation restoration in extraction-based industries, which could rely upon object based classification to lower overhead costs, improve survey accuracies, and streamline workflows.