In this paper, the dyadic discrete wavelet transform is proposed for feature extraction from a high-dimensional data space. The wavelet's inherent multiresolutional properties are discussed in terms related to multispectral and hyperspectral remote sensing. Furthermore, various wavelet-based features applied problem of automatic classification specific ground vegetations signatures. evaluated using an automated statistical classifier. system tested agricultural applications. experimental...

Our generation has seen the boom and ubiquitous advent of Internet connectivity. Adversaries have been exploiting this omnipresent connectivity as an opportunity to launch cyber attacks. As a consequence, researchers around globe devoted big attention data mining machine learning with emphasis on improving accuracy intrusion detection system (IDS). In paper, we present few-shot deep approach for improved detection. We first trained convolutional neural network (CNN) then extracted outputs...

The traditional intelligent diagnosis methods of rotating machinery generally require feature extraction the raw signals in advance. However, it is a very time-consuming and laborious process for extracting sensitive information to improve classification performance. Deep learning method, as novel machine approach, can simultaneously achieve pattern classification. With characteristics Belief Network (DBN) one-dimensional Convolutional Neural (1D-CNN) (e.g. complex nonlinear, sparse...

In this paper, we successfully demonstrate an AlGaN HFET with a high breakdown voltage of over 1.8 kV on 4 inch Si substrates. order to obtain the and improve crystalline quality GaN layers, thick epitaxial layer including buffer total thickness 6 mum was grown. The maximum drain current were achieved be 120 A, respectively. Furthermore, suppression collapse phenomenon is examined. on-resistance not so significantly increased up off-bias-stress 1.0 kV.

Specific emitter identification (SEI) is a state-of-the-art in electronic warfare. The conventional methods for SEI hardly satisfy the modern reconnaissance. In this study, first authors study quadratic time–frequency distributions and its slice features noise analysis. Based on time-frequency features, two sequential recognition based probabilistic support vector machine (SVM) iterative least-square estimation are studied, respectively. proposed able to reject interference pulses update...

We present an efficient feature selection algorithm for the general regression problem, which utilizes a piecewise linear orthonormal least squares (OLS) procedure. The 1) determines appropriate network (PLN) model given data set, 2) applies OLS procedure to PLN model, and 3) searches useful subsets using floating search algorithm. prevents "nesting effect." proposed is computationally very because only one pass required. Several examples are demonstrate effectiveness of

This paper presents a new deep learning based approach for soil detection using high resolution multispectral satellite images with of 0.31 m. In particular, convolutional neural network (CNN) is proposed to identify potential tunnel digging activities. Spatial and spectral information in the image cube has been incorporated into CNN. We also propose novel method handle imbalance context CNN model training. Experimental results on Worldview-2 (WV-2) captured at border between USA Mexico...

In practical engineering, data often lack labels, resulting in difficulty fault diagnosis. Because stack-denoising autoencoders possess robust feature extraction capabilities and resistance to interference, an automatic unsupervised bearing diagnosis method based on the autoencoder without output layer was proposed this study. As stacked denoising is algorithm, approach can reduce reliance manually labeled labels. Therefore, study a new for First, features of rolling were extracted using...

This paper proposes a novel locally linear backpropagation based contribution (LLBBC) for nonlinear process fault diagnosis. As method on the deep learning model of auto-encoder (AE), LLBBC can deal with diagnosis problem through extracting features. When on-line task is in progress, firstly built at current sample. According to basic idea reconstruction (RBC), propagation information described by using back-propagation (BP) algorithm. Then, index established measure correlation between...

Abstract Recent advancements in deep learning techniques have significantly improved methods for diagnosing bearing faults. However, most of these poor generalization performance when there are insufficient labeled samples and therefore cannot identify the root causes failures, especially noisy environments. Hence, study proposes a novel feature-enhanced few-shot method fault diagnosis based on dynamic sparse attention, which prioritizes feature extraction noise resilience over complex model...

Feature extraction plays an important role in fault diagnosis. It is critical to extract the representative features for improving classification performance. An intelligent diagnosis method based on Marginal Fisher analysis (MFA) put forward and applied rolling bearings. The high-dimensional time-domain, frequency-domain wavelet-domain are extracted from raw vibration signals obtain rich faulty information. Subsequently, MFA excavates underlying low-dimensional characteristics embedded...

Previous work on process monitoring has shown that chemical processes can be modeled by data-based models such as principal component analysis (PCA) and neural network models. However, it is difficult to train a model good generalization capabilities in fault detection, especially for nonlinear processes. On the basis of idea making trained robust with respect noisy training data, this paper intends develop unified method PCA autoencoder model. A called self-supervised first proposed. Then...