Discrimination of soil texture and contour recognitions during archaeological excavation using Machine Learning

Abstract

The most time-honoured tool for understanding the processes of the human past is represented by archaeological excavation. By examining an area at discrete temporal periods, archaeologists are literally able to look backwards in time: they can analyse incomplete material records in order to understand and reconstruct the cultural history of an area at particular moments in time. Since the digging process destroys the site forever, great care must be paid during both the excavation and the documentation. In general, after a stratum has been completely excavated, both the floors and walls are cleaned and made ready for documentation. Photos of both the sides and bedrock of a given excavation are collected, and several sketches of what the archaeologists have seen in the trenches are made. In these drawings are delineated the features and shapes of artefacts on the horizontal plane. In addition, depending on the colours and similarities of the textures, drawing are also made of the archaeological layers. This approach is time-consuming, is affected by human ability, and does not make possible a prompt digitization of the results. Within this context, the automatized identification of archaeological stratigraphy during excavation work is welcomed by archaeologists. Here, a k-means unsupervised machine learning algorithm has been used for colour clustering digital images of excavation sites. The algorithm that we have developed attempts to enhance the colour similarity while keeping the colours separate one from another as much as possible. The main idea is that pixels belonging to the same colour cluster are a part of the same layer. Once the layer has been identified, a statistical approach based on Haralick features is used to characterize each strata in terms of texture. Unsupervised machine learning combined with texture analysis could become a good practice in speeding up the documentation work of archaeologists and paving the way towards the creation of an “automated archaeologist”.