Deep learning opened new possibilities for remote sensing image analysis using multiple neural nets layers. We introduce a hybrid pixel-based model that allows improving the unsupervised training with stacked autoencoders (SAE) by inserting convolutional neural networks (CNN) in the encoding and decoding steps. Inclusion of the convolution in the encoding and decoding steps allows a feature-based description of the pixel’s hyperspectral signature, suitable to perform an initial unsupervised classification. As one-dimensional (1D) filters are applied, the processing effort is lower than when using two-dimensional-CNN. Finally, to adapt the classifier to the desired classes, the parameters of the net are adjusted using training samples and fine-tuning followed by logistic regression using the softmax activation function. This combination explores the potential of both, autoencoders (AE) and convolutional nets, providing an alternative for the classification of hyperspectral data. To evaluate the performance of the proposed approach, it was compared to traditional machine learning algorithms such as support vector machine, artificial neural networks, CNN, and SAE. The results show that the use of the SAE-1DCNN method is more effective in terms of hyperspectral classification accuracy and more efficient in computational complexity and that it can be an alternative for hyperspectral data classification.
Remote sensing enables multitemporal information of the Earth’s surface and the dynamic processes that affect the environment. Given the considerable data availability, methods to summarize multitemporal datasets are needed to support the analysis. Our study introduces and compares methods to monitor temporal variations of water bodies based on multitemporal image composition. For this purpose, the presence of water at different dates is mapped applying the normalized difference water index using two encoding methods. The first one is based on the cumulative analysis of water in the pixel along time, and the second one uses the principle of binary encoding. The cumulative analysis helps to visualize more humid and dry areas, while binary encoding indicates the monthly variations of the lake surface, storing information about the dynamics of the phenomenon. The methods are compared using Landsat time series of Lake Poopó obtained between 2013 and 2019. The results showed that binary encoding allows detecting when and where severe droughts affect the water body and its recovery. In addition, it was possible to monitor the severe drought that affected the lake in 2016 and it was also noticed that its surface is still below the level registered before the drought in 2013.
This work addresses the topic of flow direction and flow accumulation simulations in urban areas over digital surface models derived from light detection and ranging (LiDAR) data and multispectral high-resolution imagery. LiDAR data are very dense point clouds that include many objects that, in a 2 1/2-dimensional model, may become false obstacles for runoff, such as power lines or treetops. The presence of such obstacles is a problem for the flow paths simulation, especially in urban areas. We describe a methodology to produce a surface model more suitable for runoff modeling, by filtering objects that are above the surface and should not influence the flow paths. In a first step, thin obstacles are suppressed by applying mathematical morphology to a raster surface model. In a second step, satellite multispectral data and LiDAR data are classified using a support vector machine to identify trees, which are also removed from the digital model, and produce a more coherent surface model for runoff simulation. To simulate and evaluate the results, the flow-routing algorithm Dinfinity was used. The results show that the filtering is necessary to achieve a better characterization of runoff paths and allows identifying places where runoff may accumulate, causing floods or other problems.
The process of desertification, which extends from a long time ago, became a reality in Brazil. This phenomenon can be
understood as land degradation, caused by factors including climatic changes and human activities. Besides being a
threat to biodiversity, causes loss of soil productivity, threatening the lives of thousands of people living in affected
regions. So, the identification of affected areas is essential to diagnose and prevent the problem. Satellite image has been
a source of relatively low cost and widely used in this task. Therefore, is proposed in this study, a method to extract
automatically areas heavily affected by desertification. The method is based on concepts of mathematical morphology,
vegetation index and classification of digital images. Experiments are conducted separately, with images of CBERS 2
and 2B, and subsequently compared. The validation is done by crossing the results obtained with a reference image,
created by a manual process.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.