19 July 2024 in IEEE Transactions on Cognitive and Developmental Systems
Speech imagery-based Brain-computer interface (BCI) using Electroencephalogram (EEG) signal is a promising area of research for individuals with severe speech production disorders. Recent advances in deep learning (DL) have led to significant improvements in this domain. However, there is a lack of comprehensive review that covers the application of DL methods for decoding imagined speech via EEG. In this paper, we survey speech imagery and DL literature to address critical questions regarding preferred paradigms, preprocessing necessity, optimal input formulations, and current trends in DL-based techniques. Specifically, we first search major databases across science and engineering disciplines for relevant studies. Then, we analyze the DL-based techniques applied in speech imagery decoding from five main perspectives: dataset, preprocessing, input formulation, DL architecture, and performance evaluation. Moreover, we summarize the key findings of this work and propose a set of practical recommendations. Finally, we highlight the practical challenges of DL-based imagined speech decoding and suggest future research directions.
16 September 2024 in IEEE Sensors Journal
Weakly Supervised Semantic Segmentation (WSSS) aims to achieve segmentation using weak labels, thereby reducing annotation costs. Current mainstream WSSS methods utilize Class Activation Map (CAM) to generate pseudo-labels. However, CAM primarily focuses on discriminative regions beneficial for classification, which adversely affects the final segmentation performance. In this paper, inspired by learning mechanism in cognitive science theories, we propose a novel Hierarchical Activation Dual Backbone Network (HADBN) to obtain high-quality CAM. HADBN employs serial dual-backbones to decompose the CAM activation process into a free-training main-body learning part and an iterative detail supply part. Through this decomposition, HADBN highlights detailed regions that conventional CAM often overlooks, and achieves significant performance improvements through the precise application of algorithms. Furthermore, HADBN only takes supervised training with lightweight module on the main-body part of CAM to avoid introducing excessive computational cost. In extension experiments on the VOC and COCO datasets, HADBN significantly improves the quality of CAM and achieves final segmentation results with mIoU scores of 73.6 and 45.3, respectively, which are competitive with recent state-of-the-art methods. Additionally, the framework of HADBN demonstrates a certain level of versatility, making it well-suited for adaptation to other methods and improving their performance.
5 October 2025 in Neural Networks