Pattern identification and analysis for the traditional village using low altitude UAV-borne remote sensing: multifeatured geospatial data to support rural landscape investigation, documentation and management

Abstract

• Low-altitude multispectral UAV imagery provides an effective way for landscape investigation at multiple spatial scales. • High resolution and multispectral data captures multiple features of landscapes to facilitate the refined landscape detection. • Object oriented image analysis with heritage landscape structure regulation to achieve hierarchical targeted landscape classification. • Quantitative analysis to unveil the structure and spatial pattern of the landscapes in the traditional village; Spatial pattern of landscapes is viewed as the fabric and structure of traditional village. Accurate detection and analysis for the landscapes pattern plays key role in understanding the sociocultural milieu and human-natural relations. Current methods perceive their problems. Pedestrian survey is labor and time consuming. Meanwhile, the derived data tend to be subjective, qualitative and monotonous, which can hardly be used for further analysis. Remote sensing technique has been successfully applied in the field of heritage protection for its ability in object detection. But these methods are limited by visiting circle, spatial resolution and data richness. Therefore, the scientific methods of landscape pattern detection, documentation and analysis for the traditional village has long been under discussion. By taking Baojiatun castle village as a case study, the present paper aims to detect and analyze spatial pattern of traditional village by the geospatial data from a low altitude UAV-borne remote sensing. A four-leveled hierarchical landscape recognition scheme and the corresponding landscape category regulation were established. Based on the derived data and the established scheme, a three-level classification model was construct by using Object-Oriented Image Analysis method (OBIA) method and machine learning classifiers (Random Forest classifier and SVM classifier). The model was proven to be accurate and stable by ten-fold cross validation, and five major heritage landscape elements of the village were finally extracted. Furthermore, spatial pattern characteristics and distribution differences of targeted landscapes were unveiled based on distance statistics and clustering analysis. Lastly, further discussion is fostered, which focuses on the usefulness of remote sensing technique in the field of heritage landscape investigation, documentation and management.