Search Results (2)

Despite the popularity of electroencephalograms (EEGs) as tools for assessing brain health, they can sometimes be abstract and prone to noise, making them difficult to interpret. The following work aims to implement a Quaternion Convolutional Neural Network (QCNN) to detect abnormal EEGs obtained from a database that includes both people with excellent mental health and individuals with different types of mental illnesses. Unlike other approaches in which the QCNN is used exclusively for image processing, in the present work, a unique architecture with mainly quaternionic layers is proposed, specifically designed for the classification of time-varying signals. Using the database “The TUH EEG Abnormal Corpus”, the signals are preprocessed using the Wavelet Transform, a mathematical tool capable of performing simultaneous time and frequency analysis, configured with a level 4 decomposition value. Subsequently, the results are subjected to a partial spectrogram-type treatment to integrate the energy parameter into the analysis. They are then conditioned in each of the elements of the quaternion and processed by the QCNN, leveraging quaternion algebra to maintain the relationships between its elements, both in the input and in the convolutional product. In this way, it is possible to obtain significant percentages in the precision, recall, and accuracy metrics with values higher than 77%. Its performance, which uses 4 times less computational memory, allows the QCNN to be considered an alternative for classifying EEG signals. Finally, a comparison of the proposed model was made with other architectures commonly used in the literature, as well as with developments in other research and with a hybrid model whose performance places it at the highest classification standard, not to mention the ability of the QCNN to preserve multi-channel dependencies in EEG signals in a more natural way, achieving parameter efficiencies by leveraging quaternion algebra, reducing the computational cost compared to real-valued CNNs. Full article

Polarimetric synthetic aperture radar (PolSAR) image classification plays an important role in various PolSAR image application. And many pixel-wise, region-based classification methods have been proposed for PolSAR images. However, most of the pixel-wise methods can not model local spatial relationship of pixels due to negative effects of speckle noise, and most of the region-based methods fail to figure out the regions with the similar polarimetric features. Considering that color features can provide good visual expression and perform well for image interpretation, in this work, based on the PolSAR pseudo-color image over Pauli decomposition, we propose a supervised PolSAR image classification approach combining learned superpixels and quaternion convolutional neural network (QCNN). First, the PolSAR RGB pseudo-color image is formed under Pauli decomposition. Second, we train QCNN with quaternion PolSAR data converted by RGB channels to extract deep color features and obtain pixel-wise classification map. QCNN treats color channels as a quaternion matrix excavating the relationship among the color channels effectively and avoiding information loss. Third, pixel affinity network (PAN) is utilized to generate the learned superpixels of PolSAR pseudo-color image. The learned superpixels allow the local information exploitation available in the presence of speckle noise. Finally, we fuse the pixel-wise classification result and superpixels to acquire the ultimate pixel-wise PolSAR image classification map. Experiments on three real PolSAR data sets show that the proposed approach can obtain 96.56%, 95.59%, and 92.55% accuracy for Flevoland, San Francisco and Oberpfaffenhofen data set, respectively. And compared with state-of-the-art PolSAR image classification methods, the proposed algorithm can obtained competitive classification results. Full article