Abstract
Abstract. Domain adaptation (DA) can drastically decrease the amount of training data needed to obtain good classification models by leveraging available data from a source domain for the classification of a new (target) domains. In this paper, we address deep DA, i.e. DA with deep convolutional neural networks (CNN), a problem that has not been addressed frequently in remote sensing. We present a new method for semi-supervised DA for the task of pixel-based classification by a CNN. After proposing an encoder-decoder-based fully convolutional neural network (FCN), we adapt a method for adversarial discriminative DA to be applicable to the pixel-based classification of remotely sensed data based on this network. It tries to learn a feature representation that is domain invariant; domain-invariance is measured by a classifier’s incapability of predicting from which domain a sample was generated. We evaluate our FCN on the ISPRS labelling challenge, showing that it is close to the best-performing models. DA is evaluated on the basis of three domains. We compare different network configurations and perform the representation transfer at different layers of the network. We show that when using a proper layer for adaptation, our method achieves a positive transfer and thus an improved classification accuracy in the target domain for all evaluated combinations of source and target domains.
Highlights
The first step to generate maps from remotely sensed data is pixel-wise classification of these data
ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume IV-2/W7, 2019 PIA19+MRSS19 – Photogrammetric Image Analysis & Munich Remote Sensing Symposium, 18–20 September 2019, Munich, Germany experiments show that our fully convolutional neural networks (FCN) model achieves results close to the state of the art and that our Domain adaptation (DA) approach achieves a positive transfer in all cases if an appropriate layer for representation transfer is used
To be consistent with the evaluation on the benchmark website, we report the overall accuracies (OA) and F1 scores determined without considering pixels near object boundaries in the reference, i.e., based on the eroded reference provided by the benchmark organizers
Summary
The first step to generate maps from remotely sensed data is pixel-wise classification (or semantic segmentation) of these data. One of the keys to the success of CNN was the availability of large collections of annotated images (Krizhevsky et al, 2012). There is only a limited amount of freely available data with annotations; see (Zhu et al, 2017) for a recent overview. The large variations of the appearance of objects, for instance due to seasonal effects, lighting conditions, geographical variability of objects and sensor properties, makes it impossible to apply classifiers trained on such existing data directly to new data without a significant drop of classification accuracy. Ground truth labels are usually generated by manual pixel-wise annotation based on subsets of the images to be classified, a very tedious and time-consuming task
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
