Insulation equipment plays an important role in mechanical support and electrical insulation in the power grid. When there are defects in the insulation equipment, the safe operation of the power grid will be seriousl...
详细信息
Insulation equipment plays an important role in mechanical support and electrical insulation in the power grid. When there are defects in the insulation equipment, the safe operation of the power grid will be seriously threatened. Non-destructive testing (NDT) is an important means to timely find hidden dangers. In view of the low reliability of defect recognition in the case of insufficient sample marks, based on autoencoder feature extraction and semisupervised networks, combined with a terahertz (THz) wave detection device, this article studies the nondestructive detection method of insulator internal defects. First, the spectrum signal of the THz wave is obtained by continuous wavelet transform. Then, for THz time-domain and frequency-domain data, autoencoders incorporating a soft attention mechanism and a channel-spatial attention mechanism are used to automatically extract features, and time-frequency domain cognition is spliced to form fusion features. Finally, a semisupervised ladder network classification model is constructed to train the algorithm efficiently and classify reliably when it is difficult to obtain labels of defective samples. Compared with other networks oriented to 1-D and 2-D data that are trained in the common supervised way, the method in this article has a better performance in classification accuracy and recall rate, which is helpful to improve the detection effect of internal defects of insulation equipment based on the THz wave.
Traditional event detection from video frames are based on a batch or offline based algorithms: it is assumed that a single event is present within each video, and videos are processed, typically via a pre-processing ...
详细信息
ISBN:
(数字)9781510621725
ISBN:
(纸本)9781510621725
Traditional event detection from video frames are based on a batch or offline based algorithms: it is assumed that a single event is present within each video, and videos are processed, typically via a pre-processing algorithm which requires enormous amounts of computation and takes lots of CPU time to complete the task. While this can be suitable for tasks which have specified training and testing phases where time is not critical, it is entirely unacceptable for some real-world applications which require a prompt, real-time event interpretation on time. With the recent success of using multiple models for learning features such as generative adversarial autoencoder (GANS), we propose a two-model approach for real-time detection. Like GANs which learns the generative model of the dataset and further optimizes by using the discriminator which learn per sample difference between generated images. The proposed architecture uses a pre-trained model with a large dataset which is used to boost weekly labeled instances in parallel with deep-layers for the small aerial targets with a fraction of the computation time for training and detection with high accuracy. We emphasize previous work on unsupervised learning due to overheads in training labeled data in the sensor domain.
Alzheimer's disease (AD) is a typical chronic neurodegenerative disease. Mild cognitive impairment (MCI) is a transitional stage between health and AD. Early diagnosis and early treatment can significantly prolong...
详细信息
ISBN:
(纸本)9781538669563
Alzheimer's disease (AD) is a typical chronic neurodegenerative disease. Mild cognitive impairment (MCI) is a transitional stage between health and AD. Early diagnosis and early treatment can significantly prolong the survival of patients with Alzheimer's disease. By mining gene expression data and extracting the expression pattern of AD/MCI related genes, it is of great significance for early detection of AD. Here we applied a three-layer stacked denoising autoencoder (SDAE), which is a promising approach extract useful features, to the gene expression data of AD patients in the database of Alzheimer's Disease Neuroimaging Initiative (ADNI). By optimizing numbers of hidden layer nodes and corruption levels, we constructed a model with low loss. Using the features generated from SDAE, an SVM classifier was constructed to classify health status and AD/MCI samples. Results showed that the characteristics of SDAE were significantly better for the classification of AD than the raw gene expression data. Different hidden layer of SDAE was able to extract different dimensional features. Integrating all the three-layer features, the classification accuracy of 744 samples in the ADNI dataset reached 100% in 10-fold cross validation.
In motor-imagery-based brain-computer interfaces, the frequency, and spatial information of electroencephalography signals can be used to improve the performance of motor imagery classification. However, the problem o...
详细信息
In motor-imagery-based brain-computer interfaces, the frequency, and spatial information of electroencephalography signals can be used to improve the performance of motor imagery classification. However, the problem of subject-specific frequency band selection occurs frequently in spatial feature extraction. In this study, to enhance the frequency information in a spatial filter, we design an upper triangle filter bank to determine discriminative frequency components and apply the common spatial pattern to extract spatial features from subbands. Furthermore, an autoencoder neural network is constructed to reduce the high dimensionality of spatial features. The classification performance of the proposed method is experimentally evaluated on motor imagery datasets. The proposed method provides more discriminative features and higher classification performance in comparison with competing algorithms. This proposed filter bank method can be used to extend the other spatial and spectral processing method for motor imagery classification.
暂无评论