Field and Technical Report THE POTENTIAL FOR LIDAR USING SUPPORT VECTOR MACHINE (SVM) TO DETECT ARCHAEOLOGICAL STONE-WALLED STRUCTURES IN KHUTWANENG, BOKONI

Interest in precolonial stone-walled structures in Mpumalanga began in the 19th century AD when European settlers and travellers encountered these stone-walled complexes. Since then these sites, which are associated with African farming communities, have been investigated using ground-based surveys and aircraft-based remote sensing, including LiDAR (Light Detection and Ranging). Remote sensing using aerial photographs has been central to research in Bokoni since the 1970s. Through this, stone-walled structures of various shapes, sizes and functions have been identified. These include circular homesteads, agricultural terraces and stone-walled roads. This study employed machine learning - support vector machines (SVM) - to determine whether such structures can be automatically extracted from LiDAR data and classified using computerised techniques. In this paper, we present our innovative research on the use of machine learning applications for the detection of stone-walled structures in the Bokoni village of Khutwaneng, Thaba-Chweu, Mpumalanga Province. The results show that the total (overall) accuracy of the auto detection of stone-walled structures is 95% and the Kappa coefficient (K) accuracy is 91%, which is acceptable in terms of both overall and K accuracy. The object-oriented machine learning technique was found to have good classification accuracies for detecting stone-walled structures.