Scale parameter optimization through high-resolution imagery to support mine rehabilitated vegetation classification

Determining the appropriate scale of image objects is a critical aspect that affects segmentation quality in object-based classification. This research aims to assess the influence of three objective segmentation optimization methods for object-based vegetation classification through high-resolution imagery at mine rehabilitation site. These methods included estimation of scale parameter, segmentation error index, and Euclidean distance 2 index, which were employed to determine segmentation parameters, final classification accuracy, and optimal scale parameters for segmentation. The results showed that segmentation optimization may increase the classification accuracy of object-based analysis to extract information from artificial objects with regular shape, such as rehabilitated vegetation and urban green space. Given the high-resolution image with multi-spectral bands, the segmentation error and Euclidean distance 2 indices improved final classification accuracy relative to the control. Final classification accuracy obtained from each of four workflows further indicated that relatively small differences in scale parameters can lead to considerable differences in final classification accuracy. These methods automate the mail classification steps and improve the classification accuracy, only some parts of the classification require manual intervention, resulting in a more transferable approach with potentially less time required for classifying new imagery.