Motor imagery (MI) electroencephalography (EEG) classification is an important part of the brain-computer interface (BCI), allowing people with mobility problems to communicate with the outside world via assistive devices. However, EEG decoding is a challenging task because of its complexity, dynamic nature, and low signal-to-noise ratio. Designing an end-to-end framework that fully extracts the high-level features of EEG signals remains a challenge. In this study, we present a parallel spatial–temporal self-attention-based convolutional neural network for four-class MI EEG signal classification. This study is the first to define a new spatial-temporal representation of raw EEG signals that uses the self-attention mechanism to extract distinguishable spatial–temporal features. Specifically, we use the spatial self-attention module to capture the spatial dependencies between the channels of MI EEG signals. This module updates each channel by aggregating features over all channels with a weighted summation, thus improving the classification accuracy and eliminating the artifacts caused by manual channel selection. Furthermore, the temporal self-attention module encodes the global temporal information into features for each sampling time step, so that the high-level temporal features of the MI EEG signals can be extracted in the time domain. Quantitative analysis shows that our method outperforms state-of-the-art methods for intra-subject and inter-subject classification, demonstrating its robustness and effectiveness. In terms of qualitative analysis, we perform a visual inspection of the new spatial–temporal representation estimated from the learned architecture. Finally, the proposed method is employed to realize control of drones based on EEG signal, verifying its feasibility in real-time applications.
Objective . Motor imagery (MI) electroencephalography (EEG) classification is regarded as a promising technology for brain–computer interface (BCI) systems, which help people to communicate with the outside world using neural activities. However, decoding human intent accurately is a challenging task because of its small signal-to-noise ratio and non-stationary characteristics. Methods that directly extract features from raw EEG signals ignores key frequency domain information. One of the challenges in MI classification tasks is finding a way to supplement the frequency domain information ignored by the raw EEG signal. Approach . In this study, we fuse different models using their complementary characteristics to develop a multiscale space-time-frequency feature-guided multitask learning convolutional neural network (CNN) architecture. The proposed method consists of four modules: the space-time feature-based representation module, time-frequency feature-based representation module, multimodal fused feature-guided generation module, and classification module. The proposed framework is based on multitask learning. The four modules are trained using three tasks simultaneously and jointly optimized. Results . The proposed method is evaluated using three public challenge datasets. Through quantitative analysis, we demonstrate that our proposed method outperforms most state-of-the-art machine learning and deep learning techniques for EEG classification, thereby demonstrating the robustness and effectiveness of our method. Moreover, the proposed method is employed to realize control of robot based on EEG signal, verifying its feasibility in real-time applications. Significance . To the best of our knowledge, a deep CNN architecture that fuses different input cases, which have complementary characteristics, has not been applied to BCI tasks. Because of the interaction of the three tasks in the multitask learning architecture, our method can improve the generalization and accuracy of subject-dependent and subject-independent methods with limited annotated data.
Small object detection is a major challenge in the field of object detection. With the development of deep learning, many methods based on deep convolutional neural networks (DCNNs) have greatly improved the speed of detection while ensuring accuracy. However, due to the contradiction between the spatial details and semantic information of DCNNs, previous deep learning methods often meet problems when detecting small objects. The challenge can be more serious in complex scenes involving similar background objects and/or occlusion, such as in remote sensing imagery. In this paper, we propose an end-to-end DCNN called the multi-component fusion network (MCFN) to improve the accuracy of small object detection in such cases. First, we propose a dual pyramid fusion network, which densely concatenates spatial information and semantic information to extract small object features via encoding and decoding operations. Then we use a relative region proposal network to adequately extract the features of small objects samples and parts of objects. Finally, to achieve robustness against background disturbance, we add contextual information to the proposal regions before final detection. Experimental evaluations demonstrate that the proposed method significantly improves the accuracy of object detection in remote sensing images compared with other state-of-the-art methods, especially in complex scenes with the conditions of occlusion.