Mercedes E. Paoletti

Cloud implementation of the K-means algorithm for hyperspectral image analysis

Remotely sensed hyperspectral imaging offers the possibility to collect hundreds of images, at different wavelength channels, for the same area on the surface of the Earth. Hyperspectral images are characterized by their large volume and dimensionality, which makes their processing and storage difficult. As a result, several techniques have been developed in previous years to perform hyperspectral image analysis on high-performance computing architectures. However, the application of cloud computing techniques has not been as widespread. There are many potential advantages in exploiting cloud computing architectures for distributed hyperspectral image analysis. In this paper, we present a cloud implementation (developed using Apache Spark) of the popular K-means algorithm for unsupervised hyperspectral image clustering. The experimental results suggest that cloud architectures allow for the efficient distributed processing of large hyperspectral image data sets.

A Navigation Agent for Mobile Manipulators

Robot navigation and manipulation in partially known indoor environments is usually organized as two complementary activities, local displacement control and global path planning. Both activities have to be connected across different space and time scales in order to obtain a smooth and responsive system that follows the path and adapts to the unforeseen situations imposed by the real world. There is not a clear consensus in how to do this and some important problems are still open. In this paper we present the first steps towards a new navigation agent controlling both the robot’s base and the arm. We address several of theses problems in the design of this agent, including robust localization integrating several information sources, incremental learning of free navigation and manipulation space, hand visual servoing in camera space to reduce backslash and calibration errors, and internal path representation as an elastic band that is projected to the real world through measurements of the sensors. A set of experiments are presented with the robot Ursus in real and simulated scenarios showing some encouraging results.

Fast dimensionality reduction and classification of hyperspectral images with extreme learning machines

Recent advances in remote sensing techniques allow for the collection of hyperspectral images with enhanced spatial and spectral resolution. In many applications, these images need to be processed and interpreted in real-time, since analysis results need to be obtained almost instantaneously. However, the large amount of data that these images comprise introduces significant processing challenges. This also complicates the analysis performed by traditional machine learning algorithms. To address this issue, dimensionality reduction techniques aim at reducing the complexity of data while retaining the relevant information for the analysis, removing noise and redundant information. In this paper, we present a new real-time method for dimensionality reduction and classification of hyperspectral images. The newly proposed method exploits artificial neural networks, which are used to develop a fast compressor based on the extreme learning machine. The obtained experimental results indicate that the proposed method has the ability to compress and classify high-dimensional images fast enough for practical use in real-time applications.

Deep&Dense Convolutional Neural Network for Hyperspectral Image Classification

Deep neural networks (DNNs) have emerged as a relevant tool for the classification of remotely sensed hyperspectral images (HSIs), with convolutional neural networks (CNNs) being the current state-of-the-art in many classification tasks. However, deep CNNs present several limitations in the context of HSI supervised classification. Although deep models are able to extract better and more abstract features, the number of parameters that must be fine-tuned requires a large amount of training data (using small learning rates) in order to avoid the overfitting and vanishing gradient problems. The acquisition of labeled data is expensive and time-consuming, and small learning rates forces the gradient descent to use many small steps to converge, slowing down the runtime of the model. To mitigate these issues, this paper introduces a new deep CNN framework for spectral-spatial classification of HSIs. Our newly proposed framework introduces shortcut connections between layers, in which the feature maps of inferior layers are used as inputs of the current layer, feeding its own output to the rest of the the upper layers. This leads to the combination of various spectral-spatial features across layers that allows us to enhance the generalization ability of the network with HSIs. Our experimental results with four well-known HSI datasets reveal that the proposed deep&dense CNN model is able to provide competitive advantages in terms of classification accuracy when compared to other state-of-the-methods for HSI classification.