Abstract
Abstract. Pixel-wise classification of remote sensing imagery is highly interesting for tasks like land cover classification or change detection. The acquisition of large training data sets for these tasks is challenging, but necessary to obtain good results with deep learning algorithms such as convolutional neural networks (CNN). In this paper we present a method for the automatic generation of a large amount of training data by combining satellite imagery with reference data from an available geospatial database. Due to this combination of different data sources the resulting training data contain a certain amount of incorrect labels. We evaluate the influence of this so called label noise regarding the time difference between acquisition of the two data sources, the amount of training data and the class structure. We combine Sentinel-2 images with reference data from a geospatial database provided by the German Land Survey Office of Lower Saxony (LGLN). With different training sets we train a fully convolutional neural network (FCN) and classify four land cover classes (Building, Agriculture, Forest, Water). Our results show that the errors in the training samples do not have a large influence on the resulting classifiers. This is probably due to the fact that the noise is randomly distributed and thus, neighbours of incorrect samples are predominantly correct. As expected, a larger amount of training data improves the results, especially for the less well represented classes. Other influences are different illuminations conditions and seasonal effects during data acquisition. To better adapt the classifier to these different conditions they should also be included in the training data.
Highlights
Automatic classification and analysis of remote sensing imagery is highly interesting for current and future applications in land surveying and related disciplines such as navigation and city planning
In this work we address this problem by combining Sentinel2 imagery with an available topographic geospatial database to generate a large amount of training data, which results in two challenges that have to be considered: the images and the land cover data have to be compatible with respect to acquisition time, spatial resolution and class structure, and we have to consider the fact that the training data may contain a certain amount of incorrect labels
In the following we present a few approaches that are close to our field, dealing with noisy remote sensing data and convolutional neural networks (CNN) for semantic segmentation
Summary
Automatic classification and analysis of remote sensing imagery is highly interesting for current and future applications in land surveying and related disciplines such as navigation and city planning. Since the launch of the first Sentinel mission in 2014 as part of the European Union’s Copernicus project, image data has been continuously collected and made available free of charge (Fletcher, 2012) This data is already used for tasks like land cover classification and, due to its high temporal resolution, for change detection and rapid mapping in case of natural disasters. In this work we address this problem by combining Sentinel imagery with an available topographic geospatial database (referred to as maps in this paper) to generate a large amount of training data, which results in two challenges that have to be considered: the images and the land cover data have to be compatible with respect to acquisition time, spatial resolution and class structure, and we have to consider the fact that the training data may contain a certain amount of incorrect labels.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have