Decomposition and replacement: Spatial knowledge distillation for monocular depth estimation

Knowledge distillation has become a key technique for making smart and light-weight networks through model compression and transfer learning. Unlike previous methods that applied knowledge distillation to the classification task, we propose to exploit the decomposition-and-replacement based distillation scheme for depth estimation from a single RGB color image. To do this, Laplacian pyramid-based knowledge distillation is firstly presented in this paper. The key idea of the proposed method is to transfer the rich knowledge of the scene depth, which is well encoded through the teacher network, to the student network in a structured way by decomposing it into the global context and local details. This is fairly desirable for the student network to restore the depth layout more accurately with limited resources. Moreover, we also propose a new guidance concept for knowledge distillation, so-called ReplaceBlock, which replaces blocks randomly selected in the decoded feature of the student network with those of the teacher network. Our ReplaceBlock gives a smoothing effect in learning the feature distribution of the teacher network by considering the spatial contiguity in the feature space. This process is also helpful to clearly restore the depth layout without the significant computational cost. Based on various experimental results on benchmark datasets, the effectiveness of our distillation scheme for monocular depth estimation is demonstrated in details. The code and model are publicly available at : https://github.com/tjqansthd/Lap_Rep_KD_Depth.     

A new signal processing-based islanding detection method using pyramidal algorithm with undecimated wavelet transform for distributed generators of hydrogen energy

Machine learning-based fault detection methods are frequently combined with wavelet transform (WT) to detect an unintentional islanding condition. In contrast to this condition, these methods have long detection and computation time. Thus, selecting a useful signal processing-based approach is required for reliable islanding detection, especially in real-time applications. This paper presents a new modified signal processing-based islanding detection method (IDM) for real-time applications of hydrogen energy-based distributed generators. In the study, a new IDM using a modified pyramidal algorithm approach with an undecimated wavelet transform (UWT) is presented. The proposed method is performed with different grid conditions with the presence of electric noise in real-time. Experimental results show that oscillations in the acquired signal can be reduced by the UWT, and noise sensitivity is lower than other WT-based methods. The non-detection zone is zero and the maximum detection and computational time is also 75 ms at a close power match.     

No-Reference Video Quality Assessment using novel hybrid features and two-stage hybrid regression for score level fusion

This paper presents a novel No-Reference Video Quality Assessment (NR-VQA) model that utilizes proposed 3D steerable wavelet transform-based Natural Video Statistics (NVS) features as well as human perceptual features. Additionally, we proposed a novel two-stage regression scheme that significantly improves the overall performance of quality estimation. In the first stage, transform-based NVS and human perceptual features are separately passed through the proposed hybrid regression scheme: Support Vector Regression (SVR) followed by Polynomial curve fitting. The two visual quality scores predicted from the first stage are then used as features for the similar second stage. This predicts the final quality scores of distorted videos by achieving score level fusion. Extensive experiments were conducted using five authentic and four synthetic distortion databases. Experimental results demonstrate that the proposed method outperforms other published state-of-the-art benchmark methods on synthetic distortion databases and is among the top performers on authentic distortion databases. The source code is available at https://github.com/anishVNIT/two-stage-vqa.     

Growth mechanism and gas-sensing characteristics of organic-additive-free zinc oxide of various shapes

Zinc oxide is widely used in gas sensors, solar cells, and photocatalysts because of its wide bandgap and exciton binding energy of 60 meV in various metal oxides. To use ZnO as a gas sensor, it is necessary to synthesize it with surface defects and a large specific surface area. In this study, hydrothermal synthesis without surfactants was employed to obtain organic-additive-free ZnO. For morphology control, we varied the ratio of the hydroxide ion concentration to the zinc ion concentration. To confirm the growth mechanism of ZnO, we performed X-ray diffraction, scanning electron microscopy, and transmission electron microscopy analyses. Raman spectroscopy and photoluminescence measurements were performed to analyze the surface properties. The Brunauer–Emmett–Teller method and probe stations were used to measure the specific surface area and sensitivity of the gas sensor, respectively. The results confirmed that flower-shaped ZnO is the most suitable gas-sensing material.     

基于LP与小波包的复杂机电系统监测序列混沌降噪方法

针对流程工业生产系统监测数据存在强噪声和混沌性的特点,提出了一种局部投影方法(Local Projection Method)与小波包方法相结合的信号降噪方法。该方法先利用局部投影方法从动力学系统嵌入流形的角度进行多次迭代降噪,并根据关联维数来判定迭代终止;再利用小波包方法从频率的角度进行降噪,抑制高频噪声的干扰,取得了较好的降噪效果。用Lorenz时间序列进行仿真验证,对仿真时间序列加入不同程度的噪声,对比分析小波包、局部投影与该方法降噪后的相空间、SNR值和最大Lyapunov指数,证明了该方法对于中高强度噪声具有更好的降噪效果。并将该方法用于某压缩机组的实际监测数据降噪,评估三种方法的降噪效果,进一步验证了该方法的优越性。     

一种基于全变分技术和Lp伪范数的椒盐噪声去除方法

针对传统各向同性全变分（Isotropy total variation，ITV）去噪算法容易导致图像边缘模糊、不易保持图像细节信息等问题，提出一种基于L _p 伪范数和各向同性全变分的图像去噪方法。该方法将L _p 伪范数代替ITV模型中的L ₁范数，利用交替方向乘子算法（Alternating direction method of multipliers, ADMM）将能量泛函拆解成若干个子问题，并将差分算子视为卷积算子；然后引入卷积定理和快速傅里叶变换(Fast Fourier transform, FFT)提高算法运算效率；最后通过Matlab进行仿真实验，运用图像质量的客观和主观评价方法进行评价分析。结果表明，本文方法能够较好地保留图像的边缘特性，有效提升去噪效果。     

改进RetinaFace的自然场景口罩佩戴检测算法

新型冠状病毒可以通过空气中的飞沫、气溶胶等载体进行传播，在公共场所下正确佩戴口罩可以有效地防止病毒的传播。提出了一种自然场景下人脸口罩佩戴检测方法，对RetinaFace算法进行了改进，增加了人脸口罩佩戴检测任务，优化了损失函数。在特征金字塔网络中引入了一种改进的自注意力机制，增强了特征图的表达能力。建立了包含3 000张图片的数据集，并进行手工标注，用于网络训练。实验结果表明该算法可以有效进行口罩佩戴检测，在自然场景视频中也取得了不错的检测效果。     

The control of paracetamol particle size and surface morphology through crystallisation in a spray dryer

The parameters governing the crystallisation of paracetamol using various conventional techniques has been extensively studied, however the factors influencing the drug crystallisation using spray drying is not as well understood. The aim of this work was to investigate the crystallisation of an active pharmaceutical ingredient through evaporative crystallisation using a spray dryer to study the physicochemical properties of the drug and to use semi-empirical equations to gain insight into the morphology and particle size of the dried powder. Paracetamol solutions were spray dried at various inlet temperatures ranging from 60 °C to 120 °C and also from a series of inlet feed solvent compositions ranging from 50/50% v/v ethanol/water to 100% ethanol and solid-state characterisation was done. The size and morphology of the dried materials were altered with a change in spray drying parameters, with an increase in inlet temperature leading to an increase in particle Sauter mean diameter (from 3.0 to 4.4 µm) and a decrease in the particle size with an increase in ethanol concentration in the feed (from 4.6 to 4.4 µm) as a result of changes in particle density and atomised droplet size. The morphology of the dried particles consisted of agglomerates of individual crystallites bound together into larger semi-spherical agglomerates with a higher tendency for particles having crystalline ridges to form at higher ethanol concentrations of the feed.     

基于双树复小波的无参考立体图像质量评价

随着3D技术的不断发展，立体图像的使用领域越来越广泛，同时人们对图像的清晰度要求越来越高，因此，立体图像的质量评价成为关注点，基于此，提出了一种基于双树复小波变换的立体图像质量评价算法。使用双树复小波变换对立体图像的左、右视图进行处理，生成纹理结构图像，且根据最小能量误差的原理，获取左右视图的视差图；对纹理结构图像和视差图提取非对称广义高斯分布模型的参数、梯度幅值、相对梯度方向方差和奇异值曲线与坐标轴的面积等特征；使用AdaBoosting BP神经网络，进行训练和预测立体图像的质量得分。在LIVE立体图像数据库上的实验结果表明，新方法预测得分与主观得分有较好的一致性，获得了比较好的实验结果。