To date, comprehensive reviews and discussions of the strengths and limitations of Remote Sensing (RS) standalone and combination approaches, and Deep Learning (DL)-based RS datasets in archaeology have been limited. The objective of this paper is, therefore, to review and critically discuss existing studies that have applied these advanced approaches in archaeology, with a specific focus on digital preservation and object detection. RS standalone approaches including range-based and image-based modelling (e.g. laser scanning and SfM photogrammetry) have several disadvantages in terms of spatial resolution, penetrations, textures, colours, and accuracy. These limitations have led some archaeological studies to fuse/integrate multiple RS datasets to overcome limitations and produce comparatively detailed outcomes. However, there are still knowledge gaps in examining the effectiveness of these RS approaches in enhancing the detection of archaeological remains/areas. Thus, this review paper is likely to deliver valuable comprehension for archaeological studies to fill knowledge gaps and further advance exploration of archaeological areas/features using RS along with DL approaches.

This thesis explores the potential of Remote Sensing (RS) combination approaches to improve the detection of archaeological remains. Previous studies have found that standalone data are vital for prospecting archaeological features. However, combining multi-datasets derived from different RS sources is not commonly applied in digital archaeology to reveal uncovered remains. This research builds on previous studies by combining Light Detection and Ranging (LiDAR) with photogrammetric datasets to identify potential archaeological remains. The results demonstrated that integrated raster images can be used not only to confirm the findings from the RS standalone raster but also to further explore buried archaeology. The RS data applied in this research were gathered through international collaboration from four study sites: the Ancient City of Babylon in Iraq, Chun Castle and Chun Quoit in the UK, and the Cahokia Mounds in the USA.
This work addresses three significant research questions. The first question asks to what extent can RS standalone approaches (Laser scanning and photogrammetry) provide detailed data and can combination approaches deliver relatively greater Levels of Detail (LOD) for digital preservation. By applying Structure-from-Motion and Multi-View-Stereo (SfM-MVS) photogrammetry, as well as Terrestrial Laser Scanning (TLS), I analysed various factors of the generated point clouds including occlusion, LOD, data density, and roughness. The integration (cloud-to-cloud) and fusion (image-to-image) approaches were then applied to compensate for the drawbacks of the photogrammetric and TLS outputs. Although both the integration and fusion approaches overcome the limitations of the standalone results, there remain gaps in the models. This led to the conclusion that fusing aerial images with synthetic images is the most appropriate approach for digital preservation.
The second question investigates whether RS combination approaches improve the detection and interpretation of archaeological remains compared to standalone approaches. This opens a novel examination of the potential of the combination (fusion and integration) approaches to enhance the detection of archaeological remains in comparison to standalone methods. I combined various raster images derived from SfM photogrammetry and LiDAR data that have different specifications (e.g. different cameras, sensors, and spatial resolutions). Five new different integrated/fused datasets were originally created in this study to evaluate the most appropriate method to discover new archaeological traces. I found that the integrated datasets enhanced the raster images and relatively more archaeological remains were detected than with the standalone and fused VATs. The third research question sought the most appropriate Convolutional Neural Networks (CNN) model based on Deep Learning (DL) algorithms to be applied in the detection of archaeological remains. The single and combined raster images were used in the DL pipeline. Various CNN models, such as Squeeze Net, ResNet34, ResNet18, Inception, and Google Net were applied to estimate the training and validation accuracy of individual models and to assess the most appropriate model to be employed for object detection. In this way, the outcomes from this advanced technique can be applied as a stepping-stone toward the auto-detection of new archaeological features.
The results of this thesis suggest that combination approaches, particularly integration approaches, should be applied more widely by the archaeological community to enhance the evaluation of archaeological change and understanding of understudied archaeological areas worldwide. This thesis demonstrates the importance of generating combined datasets derived from multi-sensors and suggests that RS combination approaches along with object detection in DL are likely to cause a paradigm shift in digital archaeology.