Multiwalled carbon nanotubes (MWCNTs) were used to convert radome materials microwave absorbing materials. Dense MWCNT-fused silica composites prepared by hot-pressing technique. The exhibit high complex permittivities at X-band frequencies, depending on the content of MWCNTs. value loss tangent increases three orders over pure fused only incorporating 2.5vol% MWCNTs into composites. average magnitude transmission reaches −33dB 11–12GHz in 10vol% composites, which indicates have excellent...

This paper proposes a joint acoustic echo cancellation (AEC) and speech enhancement method with adaptive filter deep neural network (DNN) model. A partitioned block is adopted for linear AEC followed by convolutional transformer based model to suppress the residual echo, noise, reverberation. The DNN has three modules: encoder, dual-path (DPT) decoder. encoder explore potential relationships of far-end near-end signals attention mechanism transformer. DPT module further used context...

In order to improve the accuracy of sub-pixel grayscale centroid extraction for noise-containing spot images, an image denoising method based on low-rank and sparse decomposition (LRSD) was proposed in this paper. Relative total variation (RTV) introduced into basis weighted nuclear norm minimization (WNNM) model construct a new RTV-WNNM model, so as enhance detail-preserving capability LRSD method. Alternating direction multiplier (ADMM) used solve convex problem iteratively. Moreover, view...

The deep neural network (DNN) based speech enhancement approaches have achieved promising performance. However, the number of parameters involved in these methods is usually enormous for real applications on device with limited resources. This seriously restricts applications. To deal this issue, model compression techniques are being widely studied. In paper, we propose a method matrix product operators (MPO) to substantially reduce DNN models enhancement. method, weight matrices linear...

The studies of binaural hearing indicated considerable benefits the spatial information sound sources in speech understanding noise. In this paper, we propose a enhancement approach based on deep neural network. approach, signals at left and right channels are regarded as real imaginary parts monaural complex signal, ideal ratio mask is accordingly introduced then further estimated using network, followed by applying to signal. Experimental results showed that suggested able effectively...

The minimum variance distortionless response (MV-DR) beamformer is a widely used beamforming technique that extracts sound components coming from direction specified by steering vector. In this paper, we present four different vector estimation methods and analyze their influence on the MVDR in speech recognition. first one based of arrival under plane wave propagation assumption with prior knowledge microphone array geometry. other three are decomposition observed covariance matrix,...

The performance of the traditional direction-of-arrival (DOA) estimation algorithms greatly degrades in noisy and reverberant environments. Recently, deep learning has been applied to sound source localization provided substantial improvement robustness for DOA estimation. In this paper, we propose a approach using learning-based steering vector phase difference enhancement. vectors their reliability functions (ERFs) are first estimated under guidance time-frequency masks that predicted...

Abstract This paper summarizes the background of construction campus video security monitoring system, points out that deep learning has an important impact on development and puts forward design principles ideas system. On whole, architecture system is constructed. The application algorithm to will realize precision management, improve ability emergency command decision, intelligence operation, efficiency work, level ensure more safe stable.

Long Short Term Memory(LSTM) models are the building blocks of many state-of-the-art natural language processing(NLP) and speech enhancement(SE) algorithms. However, there a large number parameters in an LSTM model. This usually consumes resources to train Also, suffer from computational inefficiency inference phase. Existing model compression methods (e.g., pruning) can only discriminate based on magnitude parameters, ignoring issue importance distribution information. Here we introduce MPO...

Abstract In the current machine learning methods, deep is focus of attention. Deep technology has achieved rapid development in various related fields, especially field face recognition applications. to imitate mechanism human neural perception system by layer-by-layer autonomous obtain high-level abstract features, which can solve distribution facial changes, with fast speed and high accuracy. This article introduces advantages core technologies learning, studies analyzes application...

Magnetic hysteresis and eddy-current heating are the most significant disadvantages in giant magnetostrictive materials (GMM). And these properties have an effect on displacement precision other working performances of actuator (GMA). To know solve above problems fundamentally, key point is establishing a mathematical model which can describe magnetisation process energy loss characteristics for GMM analysing essential reason causes phenomenon. In this paper, basis Armstrong theory through...