Contents I. Introduction
Archaeological sites, encompassing ancient structures, settlements, and artifacts, represent invaluable cultural heritage that provides insights into past civilizations and human history. Identifying and classification of archaeological sites from satellite imagery is fundamental in reconstruction, provision of insights on historical human use of lands and in conservation [1]. Conventional approaches of identifying archaeological site involves a manual survey on the ground, which is highly time consuming, especially when considering such factors as large areas of unsuitable environments, or even remote areas [2]. Hence, the use of high-resolution satellite imagery has wide usage in the methodologies of modern archaeology, which enables the large-scale identification and computation of potential archaeological sites [3]. Traditional techniques have greatly contributed to the automated identification of archaeological areas by using GIS for analyzing spectral signature and geographic features indicative to archaelogical areas [4]. However, these methods had dependency on the surrounding environment and extensive expertise in the domain in order to extract relevant features. Besides, traditional methods struggle to accurately or partially identify an archaeological feature especially in a complex landscape due to lack of distinction between archaeological and non-archaeological features [5]. Convolutional Neural Network (CNN) is a class of DL models that holds great potential in addressing the above challenges. CNNs has the ability to learn the hierarchical features from the raw pixel data; thus, it is well suited for a number of applications such as image classification and object detection [6]. The use of CNNs is demonstrated by training the networks with labelled data sets of satellite images and to identify patterns and features inherent in potential archaeological sites. Moreover, DL models are scalable and can be applied in large dataset which will enhance the identification of archaeological sites on large scale [7]. This also significantly contributes to the improvement of detection capabilities and offers a reliable solution for constant monitoring and conservation of archaeological sites [8]. The proposed research focuses on developing a DL framework for effective identification and classification of archaeological sites with key contributions as listed below:
Collecting geo-referenced satellite images as in input data and pre-processing using min-max normalization for efficient data analysis and to remove unwanted data.
To improve the feature extraction process, the SFE module is used where the spatial features are extracted.
Classifying he extracted features using FCDNN for effective identification and classification of archaeological sites.
