Deep learning-driven semantic segmentation and spatial analysis of quarry relic landscapes using point cloud data: insights from the Shaoxing quarry relics

Quarry relic landscapes hold significant historical and cultural value, yet current research often lacks the depth to understand their complex spatial structure. This study addresses this gap by utilizing 3D point cloud data and deep learning to analyze quarry relic landscapes, focusing on the Shaoxing quarry relics. In this paper, point cloud models of four quarry relic landscapes were established, as well as the performance of the PointNet + + network in segmenting complex and variable quarry relic landscape spaces. Based on the semantic segmentation results, quantitative and clustering analyses were conducted on various landscape elements of the four quarry relics, thereby exploring the cultural value of Shaoxing quarry relic's heritage. The study demonstrates the following key findings: 1. The feasibility of combining 3D laser scanning and UAV photogrammetry to gather detailed site data for documenting quarry relic landscapes has been proven. 2. The PointNet + + deep learning network is particularly effective for the semantic segmentation of landscape elements associated with quarry relics. 3. The Shaoxing quarry relic exhibits a composite spatial form, with a nearly equal ratio of positive to negative space (approximately 1:1). Plants and bare rocks predominantly occupy the positive space, while rocks and stone pits exhibit the highest heritage value. 4. The development of the QLIM&PMS system has facilitated the comprehensive digitalization of the quarry relic landscape, offering examples and technical support for the preservation and utilization of quarry relic sites.