Xiang Xie

This paper proposed a Multi-Channel Multi-Domain (MCMD) based knowledge distillation algorithm for sleep staging using single-channel EEG. Both knowledge from different domains and different channels are learnt in the proposed algorithm, simultaneously. A multi-channel pre-training and single-channel fine-tuning scheme is used in the proposed work. The knowledge from different channels in the source domain is transferred to the single-channel model in the target domain. A pre-trained teacher-student model scheme is used to distill knowledge from the multi-channel teacher model to the single-channel student model combining with output transfer and intermediate feature transfer in the target domain. The proposed algorithm achieves a state-of-the-art single-channel sleep staging accuracy of 86.5%, with only 0.6% deterioration from the state-of-the-art multi-channel model. There is an improvement of 2% compared to the baseline model. The experimental results show that knowledge from multiple domains (different datasets) and multiple channels (e.g. EMG, EOG) could be transferred to single-channel sleep staging.

This paper proposed a Multi-Channel Multi-Domain (MCMD) based knowledge distillation algorithm for sleep staging using single-channel EEG. Both knowledge from different domains and different channels are learnt in the proposed algorithm, simultaneously. A multi-channel pre-training and single-channel fine-tuning scheme is used in the proposed work. The knowledge from different channels in the source domain is transferred to the single-channel model in the target domain. A pre-trained teacher-student model scheme is used to distill knowledge from the multi-channel teacher model to the single-channel student model combining with output transfer and intermediate feature transfer in the target domain. The proposed algorithm achieves a state-of-the-art single-channel sleep staging accuracy of 86.5%, with only 0.6% deterioration from the state-of-the-art multi-channel model. There is an improvement of 2% compared to the baseline model. The experimental results show that knowledge from multiple domains (different datasets) and multiple channels (e.g. EMG, EOG) could be transferred to single-channel sleep staging.

For high-resolution side scan sonar images, accurate and fast segmentation of sonar images is crucial for underwater target detection and recognition. However, due to the characteristics of low signal-to-noise ratio (SNR) and complex environmental noise of sonar, the existing methods with high accuracy and good robustness are mostly iterative methods with high complexity and poor real-time performance. For this purpose, a region growing based segmentation using the likelihood ratio testing method (RGLT) is proposed. This method obtains the seed points in the highlight and the shadow regions by likelihood ratio testing based on the statistical probability distribution and then grows them according to the similarity criterion. The growth avoids the processing of the seabed reverberation regions, which account for the largest proportion of sonar images, thus greatly reducing segmentation time and improving segmentation accuracy. In addition, a pre-processing filtering method called standard deviation filtering (STDF) is proposed to improve the SNR and remove the speckle noise. Experiments were conducted on three sonar databases, which showed that RGLT has significantly improved quantitative metrics such as accuracy, speed, and segmentation visual effects. The average accuracy and running times of the proposed segmentation method for 100 × 400 images are separately 95.90% and 0.44 s.

Finding the correspondence between the captured and projected uncoded stripe pattern, which is called the indexing problem, has proved to be difficult to overcome reliably. Due to the interference of the non‐ideal phenomena, the extracted stripe might be lost, increased or shifted undetectably. This letter proposes a graph‐matching based method to address this problem. The captured stripes are categorized into two parts according to whether they should be offset after projection. For the non‐offset stripes, they can be accurately indexed by the predetermined projection pattern. Based on the structural similarity that exists on the two kinds of stripes, we establish two graphs and match them to index the offset stripes fast and effectively. Thus, the iterative process in the previous study is avoided and the troublesome issues are solved. The proposed method is applied to the virtual touchpad scheme to make an arbitrary surface touch‐sensitiveness. Experimental results show that the ameliorated virtual touchpad scheme yields an average touch detection accuracy of 96.07%, and the indexing accuracy evaluated by randomly selected patterns is 100%.

This brief proposed LightSleepNet - a light-weight, 1-d Convolutional Neural Network (CNN) based personalized architecture for real-time sleep staging, which can be implemented on various mobile platforms with limited hardware resources. The proposed architecture only requires an input of 30s single-channel EEG signal for the classification. Two residual blocks consisting of group 1-d convolution are used instead of the traditional convolution layers to remove the redundancy in the CNN. Channel shuffles are inserted into each convolution layer to improve the accuracy. In order to avoid over-fitting to the training set, a Global Average Pooling (GAP) layer is used to replace the fully connected layer, which further reduces the total number of the model parameters significantly. A personalized algorithm combining Adaptive Batch Normalization (AdaBN) and gradient re-weighting is proposed for unsupervised domain adaptation. A higher priority is given to examples that are easy to transfer to the new subject, and the algorithm could be personalized for new subjects without re-training. Experimental results show a state-of-the-art overall accuracy of 83.8% with only 45.76 Million Floating-point Operations per Second (MFLOPs) computation and 43.08 K parameters.

This paper proposes a novel 3D discrete cosine transform (DCT) based image compression method for medical endoscopic applications. Due to the high correlation among color components of wireless capsule endoscopy (WCE) images, the original 2D Bayer data pattern is reconstructed into a new 3D data pattern, and 3D DCT is adopted to compress the 3D data for high compression ratio and high quality. For the low computational complexity of 3D-DCT, an optimized 4-point DCT butterfly structure without multiplication operation is proposed. Due to the unique characteristics of the 3D data pattern, the quantization and zigzag scan are ameliorated. To further improve the visual quality of decompressed images, a frequency-domain filter is proposed to eliminate the blocking artifacts adaptively. Experiments show that our method attains an average compression ratio (CR) of 22.94:1 with the peak signal to noise ratio (PSNR) of 40.73 dB, which outperforms state-of-the-art methods.

It is essential but quite challenging to alleviate speech information loss and distortion while developing the speech processing algorithms in hearing aids. Recently, many speech enhancement methods based on deep learning are proven effective. However, most of the algorithms fail to achieve real-time processing, which is significant for hearing aids, especially for a smartphone-centered binaural hearing aid system. A supervised speech enhancement method based on an RNN structure is proposed to address the real-time problem. The problem is explored as a resource-constrained speech intelligibility improvement problem with the target of improving speech intelligibility at low SNR situations. Both the objective and subjective experimental results, using the standard evaluation metrics and the experiments on volunteers, respectively, have verified the superiority of the proposed method.

The quality of images captured by wireless capsule endoscopy (WCE) is key for doctors to diagnose diseases of gastrointestinal (GI) tract. However, there exist many low-quality endoscopic images due to the limited illumination and complex environment in GI tract. After an enhancement process, the severe noise become an unacceptable problem. The noise varies with different cameras, GI tract environments and image enhancement. And the noise model is hard to be obtained. This paper proposes a convolutional blind denoising network for endoscopic images. We apply Deep Image Prior (DIP) method to reconstruct a clean image iteratively using a noisy image without a specific noise model and ground truth. Then we design a blind image quality assessment network based on MobileNet to estimate the quality of the reconstructed images. The estimated quality is used to stop the iterative operation in DIP method. The number of iterations is reduced about 36% by using transfer learning in our DIP process. Experimental results on endoscopic images and real-world noisy images demonstrate the superiority of our proposed method over the state-of-the-art methods in terms of visual quality and quantitative metrics.

This paper proposed a wearable smart sEMG recorder integrated gradient boosting decision tree (GBDT) based hand gesture recognition. A hydrogen-silica gel based flexible surface electrode band is used as the tissue interface. The sEMG signal is collected using a neural signal acquisition analog front end (AFE) chip. A quantitative analysis method is proposed to balance the algorithm complexity and recognition accuracy. A parallel GBDT implementation is proposed featuring a low latency. The proposed GBDT based neural signal processing unit (NSPU) is implemented on an FPGA near the AFE. A RF module is used for wireless communication. A hand gesture set including 12 gestures is designed for human-computer interaction. Experimental results show an overall hand gesture recognition accuracy of 91%.