Domain Adaptation for Semantic Segmentation of Historical Panchromatic Orthomosaics in Central Africa

Abstract

Multitemporal environmental and urban studies are essential to guide policy making to ultimately improve human wellbeing in the Global South. Land-cover products derived from historical aerial orthomosaics acquired decades ago can provide important evidence to inform long-term studies. To reduce the manual labelling effort by human experts and to scale to large, meaningful regions, we investigate in this study how domain adaptation techniques and deep learning can help to efficiently map land cover in Central Africa. We propose and evaluate a methodology that is based on unsupervised adaptation to reduce the cost of generating reference data for several cities and across different dates. We present the first application of domain adaptation based on fully convolutional networks for semantic segmentation of a dataset of historical panchromatic orthomosaics for land-cover generation for two focus cities Goma-Gisenyi and Bukavu. Our experimental evaluation shows that the domain adaptation methods can reach an overall accuracy between 60% and 70% for different regions. If we add a small amount of labelled data from the target domain, too, further performance gains can be achieved.

Multiclass Land Cover Mapping from Historical Orthophotos Using Domain Adaptation and Spatio-Temporal Transfer Learning

“…This model adaptation capability is especially useful in remote sensing applications, where domain shifts are ubiquitous due to temporal, spatial and spectral acquisition variations, and is therefore subject to a growing body of research [56][57][58][59]. However, to the best of our knowledge, only one study has considered applying UDA using a dataset of historical panchromatic orthomosaics [4]. Moreover, no studies exist that combine domain adaptation and domain-specific pretraining for multiclass LULC extraction from historical orthoimagery.…”

“…Usually, for RGB to greyscale mapping, this is the standard option. However, in the case of historical orthophotos, this mapping may not be straightforward because of potential spectral noise, blur, distortions, camera lens marks, spatially depended brightness variations, or dust on the scanner when digitizing the aerial images [4]. Moreover, for the Sagalassos historical orthophotos, we do not have actual certainty regarding their spectral band(s).…”

“…However, very little research exists that considers evaluating deep learning models using historical EO imagery-i.e., archive aerial photography products-as a data source [4][5][6][7]. Presumably because the processing of historical photos into LULC maps is particularly challenging as, among others, (i) availability and accessibility of historical EO photos is limited and often restricted to private databases, (ii) in many cases, photos still need to be scanned and/or orthorectified before they can be employed in a digital mapping system or GIS, (iii) meta-data may be lacking, (iv) only a single spectral channel is available (monochromatic), (v) photos are often of degraded quality because of potential camera lens marks, poor physical conservation conditions or dust on the scanner, causing image noise, blur, distortions/displacements or artifacts, and (vi) large spectral differences between and within images may be present, making similar objects have varying appearances.…”

An Object-Based Markov Random Field with Partition-Global Alternately Updated for Semantic Segmentation of High Spatial Resolution Remote Sensing Image

“…Further, substituting Equations ( 15), ( 16), (21), and (22) into the above Equation ( 29), we get a node-by-node update strategy.…”

“…The former designs a classifier with undetermined parameters according to specific rules, and obtains a classifier that can identify specific categories by identifying training samples. This category contains many methods, such as: support vector machines (SVMs) [6][7][8], neural network-based algorithms [9][10][11][12][13][14][15], deep learning [16][17][18][19][20][21][22], etc. The latter is based on the image's own features, directly modeling the image's data and giving segmentation results, such as: super-pixel [23][24][25][26], Markov random field (MRF) [27][28][29][30][31][32][33][34], conditional random field (CRF) [35][36][37][38][39], level set [40][41][42], etc.…”