Neural Network-Based Spatial Autocorrelation Representation for Deep Learning-Driven Geospatial Object Detection from 3D Point Clouds: A Case Study with Urban and Rural Scenes

Spatial contextual features are effective to inform spatial object detection in terms of the different contextual nature of various objects. Traditional deep neural networks (DNN), such as convolu-tional based networks, only capture part of context by operating window-based convolutions on local neighborhood of pixels in 2D imagery or points in 3D point cloud. However, studies on spatial contextual feature detection on 3D point clouds are limited. This study presents a neural network-based spatial autocorrelation encoder, designed to enhance deep learning-driven 3D geospatial object detection by integrating neural network-derived spatial autocorrelation as a representation of spatial contextual features. The study investigated the effectiveness of such spatial contextual features by estimating the model performance trained on them alone. The results suggested that the derived spatial contextual information can help adequately identify some large objects in an urban-rural scene, such as buildings, terrain, and large trees. We further investigate how the spatial autocorrelation encoder can improve the model performance in terms of geo-spatial object detection. The results demonstrated significant improvements in detection accuracy across varied urban and rural environments, when compared to models without considering the spatial autocorrelation as an ablation. Moreover, the approach also outperformed the models trained by explicitly feeding traditional spatial autocorrelation measures (i.e., Matheron’s semi-variance). This study showcases advantages of the adaptiveness of the neural network-based en-coder in deriving a spatial autocorrelation representation. This advancement bridges the gap between theoretical geospatial concepts and practical AI applications. Consequently, this study demonstrated the potential of integrating geographic theories with deep learning technologies to address challenges in 3D object detection, paving a way for further innovations in this field.