Currently, analyzing satellite images requires an unsustainable amount of manual labor. Semiautomatic solutions for land-cover classification of satellite images entail the incorporation of expert knowledge. To increase the scalability of the built solutions, methods that automate image processing and analysis pipelines are required. Recently, deep learning (DL) models have been applied to challenging vision problems with great success. We expect that the use of DL models will soon outperform shallow networks and other classification algorithms, as recently achieved in multiple domains. Here, we consider the task of land-cover classification of satellite images. This seems particularly appropriate for deep classifiers due to the combined high dimensionality of the data with the presence of compositional dependencies between pixels, which can be used to characterize a particular class. We develop a pipeline for analyzing satellite images using a deep convolutional neural network for practical applications. We present its successful application for land-cover classification, where it achieves 86% classification accuracy on unseen raw images.
State Visual Evoked Potentials (SSVEPs) arise from a resonance phenomenon in the visual cortex that is produced by a repetitive visual stimulus. SSVEPs have long been considered a steady-state response resulting from purely oscillatory components phase locked with the stimulation source, matching the stimulation frequency and its harmonics. Here we explore the dynamical character of the SSVEP response by proposing a novel non-stationary methodology for SSVEP detection based on an ensemble of Echo State Networks (ESN). The performance of this dynamical approach is compared to stationary canonical correlation analysis (CCA) for the detection of 6 visual stimulation frequencies ranging from 12 to 22 Hz. ESN-based methodology outperformed CCA, achieving an average information transfer rate of 47 bits/minute when simulating a BCI system of 6 degrees of freedom. However, for some subjects and stimulation frequencies the detection accuracy of CCA exceeds that of ESN. The comparison suggests that each methodology captures different features of the SSVEP response: while CCA captures purely stationary patterns, the ESN-based approach presented here is capable of detecting the non-stationary nature of the SSVEP.
Transcranial Direct Current Stimulation (tDCS) is a non-invasive neuromodulation technique with a wide variety of clinical and research applications. As increasingly acknowledged, its effectiveness is subject dependent, which may lead to time consuming and cost ineffective treatment development phases. We propose the combination of electroencephalography (EEG) and unsupervised learning for the stratification and prediction of individual responses to tDCS. A randomized, sham-controlled, double-blind crossover study design was conducted within a clinical trial for the development of pediatric treatments based on tDCS. The tDCS stimulation (sham and active) was applied either in the left dorsolateral prefrontal cortex or in the right inferior frontal gyrus. Following the stimulation session, participants performed 3 cognitive tasks to assess the response to the intervention: the Flanker Task, N-Back Task and Continuous Performance Test (CPT). We used data from 56 healthy children and adolescents to implement an unsupervised clustering approach that stratify participants based on their resting-state EEG spectral features before the tDCS intervention. We then applied a correlational analysis to characterize the clusters of EEG profiles in terms of participant’s difference in the behavioral outcome (accuracy and response time) of the cognitive tasks when performed after a tDCS-sham or a tDCS-active session. Better behavioral performance following the active tDCS session compared to the sham tDCS session is considered a positive intervention response, whilst the reverse is considered a negative one. Optimal results in terms of validity measures was obtained for 4 clusters. These results show that specific EEG-based digital phenotypes can be associated to particular responses. While one cluster presents neurotypical EEG activity, the remaining clusters present non-typical EEG characteristics, which seem to be associated with a positive response. Findings suggest that unsupervised machine learning can be successfully used to stratify and eventually predict responses of individuals to a tDCS treatment.