Detection of Fragmented Rectangular Enclosures in Very High Resolution Remote Sensing Images

Abstract

We develop an approach for the detection of ruins of livestock enclosures (LEs) in alpine areas captured by highresolution remotely sensed images. These structures are usually of approximately rectangular shape and appear in images as faint fragmented contours in complex background. We address this problem by introducing a rectangularity feature that quantifies the degree of alignment of an optimal subset of extracted linear segments with a contour of rectangular shape. The rectangularity feature has high values not only for perfectly regular enclosures but also for ruined ones with distorted angles, fragmented walls, or even a completely missing wall. Furthermore, it has a zero value for spurious structures with less than three sides of a perceivable rectangle. We show how the detection performance can be improved by learning a linear combination of the rectangularity and size features from just a few available representative examples and a large number of negatives. Our approach allowed detection of enclosures in the Silvretta Alps that were previously unknown. A comparative performance analysis is provided. Among other features, our comparison includes the state-of-the-art features that were generated by pretrained deep convolutional neural networks (CNNs). The deep CNN features, although learned from a very different type of images, provided the basic ability to capture the visual concept of the LEs. However, our handcrafted rectangularitysize features showed considerably higher performance.

Using deep neural networks on airborne laser scanning data: Results from a case study of semi‐automatic mapping of archaeological topography on Arran, Scotland

“…Also based on a DTM is an automatic pit filling method based on an inverted DTM to locate mound structures (Freeland et al, ); a combination of curvature estimates, topographic position index, and circular Hough transform to detect prehistoric barrows (Cerrillo‐Cuena, ); a combination of segmentation and template matching to detect grazing structures (Toumazet et al, ); and local contrast in the DTM at three different scales and a random forest classifier to detect burial mounds (Guyot et al, ). A study to detect rectangular enclosures in panchromatic satellite images (Zingman et al, ) concluded that bespoke methods in some cases perform better than using a pre‐trained deep CNN, but at the cost of much longer development time. However, the use of a deep CNN for charcoal burning platforms showed considerable improvement on an earlier template matching approach (Trier et al, ).…”

“…Recently, the PyTorch library has emerged as a better alternative to Caffe, offering more flexibility in training and classification. Accepting the published reservations that deep CNNs might not always be the best solution (Zingman et al, ), this approach has been used on Arran because transferability between datasets and the development of ‘general purpose’ archaeological CNNs is desirable if the discipline as a whole is to make better use of the methodology. For HES, the similarities in landscape forms, datasets and the basic morphology of archaeological monuments between some of the Norwegian case studies and the Scottish context meant that it was an attractive first step in establishing a proof of concept project on Arran.…”

Object‐based image analysis: a review of developments and future directions of automated feature detection in landscape archaeology

“…Morphological variables used for classification include: Circularity (Davis et al, ; Freeland et al, ; Witharana et al, ) Rectangularity (Zingman et al, ) Area (Davis et al, ; Magnini et al, ; Witharana et al, ) Length and width (Magnini et al, ; Toumazet, Vautier, Roussel, & Dousteyssier, ) Size (Cerrillo‐Cuenca, ; Davis et al, ; Zingman et al, ) Curvature (Cerrillo‐Cuenca, ) Edge detection (Traviglia & Torsello, ; Witharana et al, ; Zingman et al, ) Elevation (Davis et al, ; Guyot et al, ) …”

Landscape Pattern Detection in Archaeological Remote Sensing

“…The reasons for lack of trust in automated methods in heritage recognition often come down to uneasiness amongst archaeological scholarship of losing control of the interpretation process [12]. Examples of implementation of automated approaches to archaeological aero-photointerpretation for determined objects are thus still scarce [13] and they broadly fall into an even smaller number of classes: template matching-based methods [14], custom algorithms [15], (GE)OBIA-based methods [16], and machine learning-based methods [15], all involving previous knowledge of the shape and characteristics of the searched object. Recent developments in Machine Learning and the appearance on the scientific scene of Convolutional Neural Networks (CNN) approaches [17,18], which gained credit for dramatically improving previous recognition performances, are generating enthusiasm for the possibilities of these high-end methods for the (semi)automatic detection of archaeological surface or sub-surface objects.…”

“…Recent developments in Machine Learning and the appearance on the scientific scene of Convolutional Neural Networks (CNN) approaches [17,18], which gained credit for dramatically improving previous recognition performances, are generating enthusiasm for the possibilities of these high-end methods for the (semi)automatic detection of archaeological surface or sub-surface objects. Despite its potential, however, CNN has so far seen extremely limited development in the archaeological/cultural heritage field and only in applications where the shape of objects to detect (although variegated) were known and pertained to only one typology [15,19]. All these trials have in any case concentrated their attention on separated, single "entities" appearing with varying frequency within a landscape and not on systems of "related entities".…”