Filter Layer Research Articles

A Novel Deep Learning Approach for Tracking Regions of Interest in Ultrasound Images.

Due to their great success in learning a universal object similarity metric, Siamese Trackers have been adopted for motion tracking a Region of Interest (ROI) in Ultrasound (US) image sequences. However, these Fully Convolutional Siamese networks (SiamFC) offer no online adaptation of the network and fail to take cues from the input sequence. The more recent Correlation Filter Networks (CFNet) solve this problem by learning the reference template online using a Correlation Filter layer. In this work, we use the CFNet as our backbone model and propose an advanced tracking algorithm (Seq-CFNet) for tracking an ROI in US sequences by constructing a sequential cascade of two identical CFNet. The cascade with CFNet is novel and offers practical benefits in tracking accuracy. Our method is evaluated on 10 different sequences of a Carotid Artery (CA) dataset to track the transverse section of the carotid artery. Results show that Seq-CFNet obtains better Root Mean Square Error (RMSE) values than the baseline CFNet as well as SiamFC, without significantly compromising the speed.

Removal Efficiency and Mechanism of Ammonia Nitrogen in a Low Temperature Groundwater Purification Process

To explore the mechanism and efficiency of ammonia nitrogen removal, a pilot-scale biofilter for the simultaneous removal of high concentrations of iron, manganese, and ammonia nitrogen[Fe(Ⅱ) 11.9-14.8 mg·L-1, Mn(Ⅱ) 1.1-1.5mg·L-1, and NH4+-N 1.1-3.2 mg·L-1] from low temperature(5-6℃) groundwater was operated in a water supply plant in Northeast China. Results indicated excellent performance for ammonia nitrogen removal during the initial start-up stage. According to theoretical analysis and experimental verification, TNloss was driven by the adsorption of ammonia nitrogen by iron oxides, and the conversion of ammonia nitrogen into nitrate nitrogen occurred via biological nitrification. When the concentration of ammonia nitrogen increased, due to limited adsorption sites, the adsorption capacity of iron oxides remained stable at approximately 1 mg·L-1. For the same period, the amount of ammonia nitrogen removal via oxidation continued to increase, with higher quantities removed in the upper filter layer than in the lower filter layer. Dissolved oxygen(DO) is the limiting factor in the further increase in the removal of ammonia nitrogen by oxidation. With an increase in the filtration rate, the adsorption time of ammonia nitrogen by iron oxides was shortened, and the adsorption amount was reduced. Meanwhile, the shortening of EBCT reduced the ammonia nitrogen removed by nitrification under the action of nitrifying bacteria in the unit volume of the filter material. Based on these findings, it is recommended that the thickness of the filter layer should be increased to improve ammonia nitrogen removal performance.

Formula for the maximum reference pressure at the interface of the breakwater core and filter layer

ABSTRACT A formula is proposed to determine the maximum reference pressure at the interface of the breakwater core and filter layer, based on small-scale physical model tests. Regular and irregular waves have been considered. It is found that the maximum reference pressure at the interface of the breakwater core and filter layer depends on the level of the water surface along the breakwater seaward slope. The proposed formula is verified using large-scale model results and real-scale field observations in non-overtopping and non-breaking wave conditions; it contains a nondimensional parameter that is a function of the wavelength and width of the core at a given water depth.

GaAs on Si substrate with dislocation filter layers for wafer‐scale integration

GaAs on Si grown via metalorganic chemical vapor deposition is demonstrated using various Si substrate thicknesses and three types of dislocation filter layers (DFLs). The bowing was used to measure wafer-scale characteristics. The surface morphology and electron channeling contrast imaging (ECCI) were used to analyze the material quality of GaAs films. Only 3-μm bowing was observed using the 725-μm-thick Si substrate. The bowing shows similar levels among the samples with DFLs, indicating that the Si substrate thickness mostly determines the bowing. According to the surface morphology and ECCI results, the compressive strained indium gallium arsenide/GaAs DFLs show an atomically flat surface with a root mean square value of 1.288 nm and minimum threading dislocation density (TDD) value of 2.4 × 107 cm−2. For lattice-matched DFLs, the indium gallium phosphide/GaAs DFLs are more effective in reducing the TDD than aluminum gallium arsenide/GaAs DFLs. Finally, we found that the strained DFLs can block propagate TDD effectively. The strained DFLs on the 725-μm-thick Si substrate can be used for the large-scale integration of GaAs on Si with less bowing and low TDD.

ENHANCED PIXEL BASED URBAN AREA CLASSIFICATION OF SATELLITE IMAGES USING CONVOLUTIONAL NEURAL NETWORK

Recent years have witnessed a great development in the use of deep learning in the applied fields in general, including the improvement of remote sensing. Satellite imagery classification has played a prominent role in various development processes. This paper presents a new improvement in automatic urban classification using One Dimension Convolutional Neural Network (1DCNN) architecture. The suggested approach has three enhancement processes. First, select training boxes for different classes and create many pixels with variable class signatures. This makes the training process dependent on the broadband of signature for the classes. Second, modified 1D convolution was used to re-encode pixel values to increase distinguish power. Third, adding a new median filter layer at the end of network architecture to remove pixels like noise to make the resulting map smoother. An image of Greater Cairo is used and the different urban classes are defined within it. The proposed method was compared to other methods based on pixels. The proposed method proved to be numerically and visually superior.

Two-dimensional metal−porphyrin framework membranes for efficient molecular sieving

Two-dimensional (2D) membranes were constructed using porphyrin-based metal organic frameworks (MOFs), i.e., copper-tetrakis(4-carboxyphenyl) porphyrin (Cu-TCPP), for efficient molecular sieving of dye/salt mixture solutions. These novel membranes contained a regular Cu-TCPP nanosheet filter layer fabricated by the vacuum filtration method to retain the oriented interlayer structure of the MOF material, resulting in a robust 2D membrane. The hydrated Cu-TCPP membrane with a thickness of 300 nm displayed high retention rates for dyes (90.1% for crystal violet (CV), 93.3% for Chrome black T (CB-T), 99.7% for Methyl blue (MB) and 99.8% for Congo red (CR); CV, Mw= 407.9 g/mol; CB-T, Mw=416.4 g/mol) with diameters approximately greater than 1.3 nm. A water permeance of 840.1 L·m−2·h−1·bar−1 and a CR (Mw= 696.7 g/mol) rejection rate of 99.9% were obtained with the optimal metal−porphyrin framework membrane (small lamellar dimension nanosheets) undergoing the cross-flow mode filtration of dye/salt mixture solutions for 24 h. More importantly, due to the abundant interlayers and regular surface nanochannels, the Cu-TCPP membrane exhibited a low salt rejection rate (0.4% for NaCl) and an outstanding CR/NaCl selectivity (as high as 249.5), which was more superior compared with that of other reported two-dimensional membranes. Additionally, the nanochannels of the membrane presented strong rigidity under pressure load as high as 0.4 MPa, endowing the membrane with high tolerance to shear stress and good antifouling performance after several fouling-cleaning cycles.

Sharp U-Net: Depthwise convolutional network for biomedical image segmentation.

The U-Net architecture, built upon the fully convolutional network, has proven to be effective in biomedical image segmentation. However, U-Net applies skip connections to merge semantically different low- and high-level convolutional features, resulting in not only blurred feature maps, but also over- and under-segmented target regions. To address these limitations, we propose a simple, yet effective end-to-end depthwise encoder-decoder fully convolutional network architecture, called Sharp U-Net, for binary and multi-class biomedical image segmentation. The key rationale of Sharp U-Net is that instead of applying a plain skip connection, a depthwise convolution of the encoder feature map with a sharpening kernel filter is employed prior to merging the encoder and decoder features, thereby producing a sharpened intermediate feature map of the same size as the encoder map. Using this sharpening filter layer, we are able to not only fuse semantically less dissimilar features, but also to smooth out artifacts throughout the network layers during the early stages of training. Our extensive experiments on six datasets show that the proposed Sharp U-Net model consistently outperforms or matches the recent state-of-the-art baselines in both binary and multi-class segmentation tasks, while adding no extra learnable parameters. Furthermore, Sharp U-Net outperforms baselines that have more than three times the number of learnable parameters.

Treatment of organic peroxide containing wastewater and water recovery by fenton-adsorption and fenton-nanofiltration processes

Treatment of organic peroxide-containing chemical industry wastewater by oxidation methods and recovery of water by the adsorption and nanofiltration (NP010) methods were investigated in this study. The COD and TOC removal rates were obtained as 72.8% and 58.0% in Fenton oxidation and were improved to 78.8% and 59.2% using photo-Fenton oxidation in the same conditions after 5 h of oxidation, respectively. The Fenton-treated wastewater was passed through nanofiltration to remove the organics and recover of the wastewater. The maximum COD removal efficiency of the NP010 membrane varied between 25% and 30% at all pH values. At low and high pH values (pH: 2.5 and pH: 11), as the filtration time increased, the COD removal efficiencies increased, and the highest COD removal efficiencies were obtained in the 180th and 210th minutes. The increase in the COD removal over time at low and high pH was related to the thickness of the filter layer and surface load balance accumulated on the filter surface. By Fenton oxidation coupling with adsorption, 81% of COD (decreased from 10,055 mg/L to 1906 mg/L) and 75.2% of TOC (decreased from 2597 mg/L to 645.4 mg/L) removal could be obtained, while 83.3% of COD (decreased from 9978 mg/L to 1664 mg/L) and 71.1% of TOC (decreased from 2597 mg/L to 750 mg/L) removal could be achieved using Fenton oxidation coupling with nanofiltration (P: 4 bar, pH: 11). In nanofiltration, the filtrate amounts were measured as 41.11 L/m2.h and 38.33 L/m2.h, respectively, at 4 bar and 6 bar of filter pressure and 30 min of filtration time. The increase in filtration time and filter pressure caused a decrease in the amount of the filtrate due to the rapid clogging of the filter pores.

InAs/InGaAs Quantum Dot Lasers on Multi-Functional Metamorphic Buffer Layers.

With the development of dry fiber over the past two decades, the E-band has become a new telecommunication wavelength. However, owing to material constraints, an effective high-performance semiconductor light source has not yet been realized. InAs quantum dot (QD) lasers on GaAs substrates are in the spotlight as O-band light sources because of their excellent thermal properties and high efficiency. The introduction of a very thick InGaAs metamorphic buffer layer is essential for realizing an E-band InAs QD laser, but it can cause degradation in laser performance. In this study, we fabricate an E-band InAs/GaAs QD laser on a GaAs substrate with an AlInGaAs multifunctional metamorphic buffer layer that realizes the function of the bottom cladding layer of normal thickness in addition to the functions of a metamorphic buffer layer and a dislocation filter layer. The lasing oscillation at a wavelength of 1428 nm is demonstrated at room temperature under continuous-wave operation. This result paves the way toward the realization of highly efficient light sources suitable for E-band telecommunications.

Nested Gaussian filters for recursive Bayesian inference and nonlinear tracking in state space models

We introduce a new sequential methodology to calibrate the fixed parameters and track the stochastic dynamical variables of a state-space system. The proposed method is based on the nested hybrid filtering (NHF) framework of [1], that combines two layers of filters, one inside the other, to compute the joint posterior probability distribution of the static parameters and the state variables. In particular, we explore the use of deterministic sampling techniques for Gaussian approximation in the first layer of the algorithm, instead of the Monte Carlo methods employed in the original procedure. The resulting scheme reduces the computational cost and so makes the algorithms potentially better-suited for high-dimensional state and parameter spaces. We describe a specific instance of the new method and then study its performance and efficiency of the resulting algorithms for a stochastic Lorenz 63 model and for a stochastic volatility model with real data.

Fast and non-destructive ultrasonic test for face masks

As a consequence of the large demand of face masks due to the COVID19 pandemic, cheap, fast and non-destructive tests that can verify in-line the variability of the filtration capacities, prove the potential disinfection and/or evaluate the performance of new filtering materials are needed. Using two different approaches based on air-coupled ultrasounds (0.15–1.6 MHz) with equivalent results, this work shows that each face mask presents a distinctive ultrasonic signature that enables the classification and the evaluation of their performance. Moreover, it is shown that the ultrasonic propagation through the face masks and the main filter layers takes place through the pore space and that low frequency response of the attenuation and the velocity is highly dispersive and is dominated by the interaction between the air in the pores and the fibers in the filters. Hence, the parameters that describe ultrasonic velocity, attenuation and dispersion can be related with their filtration efficiency and breathability. These techniques are fully contactless, non-invasive and fast.

Adaptive multi-layer filters incorporated with Volterra filters for impairment compensation including transmitter and receiver nonlinearity.

We propose a receiver-side signal processing to compensate for nonlinearity that occurs in transmitter (Tx) and receiver (Rx) components of coherent optical fiber transmission systems. Nonlinear effects in transmission systems are not mutually commutative with any linear effects in general. Considering the order in which all the relevant impairments occur, we adopt a multi-layer (ML) filter architecture. The ML filters consist of strictly-linear and widely-linear filter layers to compensate for relevant linear impairments that occur in a transmission system and two Volterra filter layers to compensate for Rx and Tx nonlinearity. The coefficients of the ML filters including Volterra filter layers are adaptively controlled by using a gradient calculation with back propagation, which is similar to that used in the learning of neural networks, from the last layer and stochastic gradient descent to minimize a loss function that is composed of the last layer outputs. We evaluated the compensation performance of Tx and Rx nonlinearity using the proposed adaptive ML filters including Volterra filter layers both in simulations and experiments of the transmission of a 23 Gbaud polarization-division-multiplexed 64-quadrature amplitude modulation signal over a 100-km single-mode-fiber span. The results demonstrated that the Volterra filter layers in the ML filter architecture could compensate for the nonlinearity that occurs in Tx and Rx simultaneously and effectively even when other impairments such as chromatic dispersion coexist.

P‐4.7: Fundamental Research on High Resolution Full Color Micro‐LED Display with Quantum Dot Color Conversion by Lithography

This study is based on the blue micro‐LED backlight and photoluminescent quantum dot color conversion layer (QDCC) as the light source of the full color micro‐LED display structure. In order to obtain a high resolution Micro‐LED display panel with color convertion, we firstly studied the the influence of black matrix (BM) layer and color filter (CF) layer. The CF layer is used to reduce the blue light leakage. It should be noted that thickness and absorption spectrum of the CF layer will have a significant impact on the color gamut. The BM layer is to prevent backlight leakage from the sidewalls and crosstalk. The successfully prepared of the 5µm pattern has greatly promoted our subsequent process, and the photolithography of quantum dot layer will be further studied in our subsequent work.

Specificity of UV-C LED disinfection efficacy for three N95 respirators

The recent surge in the use of UV technology for personal protective equipment (PPE) has created a unique learning opportunity for the UV industry to deepen surface disinfection knowledge, especially on surfaces with complex geometries, such as the N95 filter facepiece respirators (FFR). The work outlined in this study addresses the interconnectedness of independent variables (e.g., UV Fluence, respirator material) that require consideration when assessing UV light efficacy for disinfecting respirators. Through electron microscopy and Fourier-transform infrared (FTIR) spectroscopy, we characterized respirator filter layers and revealed that polymer type affects disinfection efficacy. Specifically, FFR layers made from polypropylene (PP) (hydrophobic in nature) resulted in higher disinfection efficiency than layers composed of polyethylene terephthalate (PET-P) (hygroscopic in nature). An analysis of elastic band materials on the respirators indicated that silicone rubber-based bands achieved higher disinfection efficiency than PET-P bands and have a woven, fabric-like texture. While there is a strong desire to repurpose respirators, through this work we demonstrated that the design of an appropriate UV system is essential and that only respirators meeting specific design criteria may be reasonable for repurposing via UV disinfection.

Application of multi-layer denoising based on ensemble empirical mode decomposition in extraction of fault feature of rotating machinery

Aiming at the problem that the weak features of non-stationary vibration signals are difficult to extract under strong background noise, a multi-layer noise reduction method based on ensemble empirical mode decomposition (EEMD) is proposed. First, the original vibration signal is decomposed by EEMD, and the main intrinsic modal components (IMF) are selected using comprehensive evaluation indicators; the second layer of filtering uses wavelet threshold denoising (WTD) to process the main IMF components. Finally, the virtual noise channel is introduced, and FastICA is used to de-noise and unmix the IMF components processed by the WTD. Next, perform spectral analysis on the separated useful signals to highlight the fault frequency. The feasibility of the proposed method is verified by simulation, and it is applied to the extraction of weak signals of faulty bearings and worn polycrystalline diamond compact bits. The analysis of vibration signals shows that this method can efficiently extract weak fault characteristic information of rotating machinery.

Construction of reliable image captioning system for web camera based traffic analysis on road transport application

The automated captioning of natural images with appropriate descriptions is an intriguing and complicated task in the field of image processing. On the other hand, Deep learning, which combines computer vision with natural language, has emerged in recent years. Image emphasization is a record file representation that allows a computer to understand the visual information of an image in one or more words. When it comes to connecting high-quality images, the expressive process not only requires the credentials of the primary item and scene but also the ability to analyse the status, physical characteristics, and connections. Many traditional algorithms substitute the image to the front image. The image characteristics are dynamic depending on the ambient condition of natural photographs. Image processing techniques fail to extract several characteristics from the specified image. Nonetheless, four properties from the images are accurately described by using our proposed technique. Based on the various filtering layers in the convolutional neural network (CNN), it is an advantage to extract different characteristics. The caption for the image is based on long short term memory (LSTM), which comes under recurrent neural network. In addition, the precise subtitling is compared to current conventional techniques of image processing and different deep learning models. The proposed method is performing well in natural images and web camera based images for traffic analysis. Besides, the proposed algorithm leverages good accuracy and reliable image captioning.

Use of PET Granules for Improving a Surface Runoff Treatment

Pollution of surface and ground waters by surface (rain and melt) waste water is a big problem for many countries of the world. In addition to anthropogenic factors leading to the pollution of surface runoff, one should add runoff, which is formed after extinguishing fires of various scales and territories. As a result of extinguishing fires, a large amount of pollutants enters surface wastewater, which creates an additional load on soils, groundwater aquifers and water bodies. The paper proposes a wastewater treatment method using PET granules and polyurethane foam granules as a filtering layer. The use of PET packaging waste for the purification of rain and melt wastewater allows to reduce the anthropogenic load on the environment and to increase the use of plastic waste.

Research on Generation Method of Grasp Strategy Based on DeepLab V3+ for Three-Finger Gripper

A reasonable grasping strategy is a prerequisite for the successful grasping of a target, and it is also a basic condition for the wide application of robots. Presently, mainstream grippers on the market are divided into two-finger grippers and three-finger grippers. According to human grasping experience, the stability of three-finger grippers is much better than that of two-finger grippers. Therefore, this paper’s focus is on the three-finger grasping strategy generation method based on the DeepLab V3+ algorithm. DeepLab V3+ uses the atrous convolution kernel and the atrous spatial pyramid pooling (ASPP) architecture based on atrous convolution. The atrous convolution kernel can adjust the field-of-view of the filter layer by changing the convolution rate. In addition, ASPP can effectively capture multi-scale information, based on the parallel connection of multiple convolution rates of atrous convolutional layers, so that the model performs better on multi-scale objects. The article innovatively uses the DeepLab V3+ algorithm to generate the grasp strategy of a target and optimizes the atrous convolution parameter values of ASPP. This study used the Cornell Grasp dataset to train and verify the model. At the same time, a smaller and more complex dataset of 60 was produced according to the actual situation. Upon testing, good experimental results were obtained.

Pressure drop in fibrous filter – Representative domain size and effect of fibre orientation

Clean air or gas is critical e.g. in many industrial or health applications. Various filtration processes are employed to ensure that liquid droplets and/or solid particles, i.e. so–called aerosols, are removed efficiently and at low costs from the respective gas streams. Fibrous filters are often used in this context, because of their low cost, their high capture efficiency, and their low pressure drop. Hence, the performance of fibrous filter materials is judged based on their pressure drop and capture efficiency. A generalized model to describe and predict these two parameters would be helpful in developing optimal filter material.In the current work, numerical investigations using the commercial CFD tool ANSYS CFX are engaged to predict the pressure drop caused by multiple randomly–oriented fibres in a 3D domain. The fibres are randomly generated until the desired filter solidity is met. Hereby, the generation algorithm ensures that no fibres intersect. The implementation is based on a fictitious–domain approach, which has the great advantage to perfectly decouple the mesh from the fibre generation, as fibres are implemented as volumes of infinitely–high flow resistance in the fixed mesh. Hence, there is no need for a case–specific and body–fitted mesh. The effect of the domain size is particularly investigated and we find that a domain size of more than 45 times the fibre diameter in all three directions, proves to give domain–size–independent results for the pressure drop. The boundary conditions are also carefully analyzed, and we find that particularly the symmetry planes parallel to the mean flow direction are not appropriate as the representative domain is cut out of the filter layer. Here, the introduction of a pseudo–plane proves to minimize non–physical flow fields. Moreover, using scanning microscopy, a sample fibrous filter is analysed to characterize the (random) fibre orientation. It is found, that within the filter plane an evenly–distributed fibre orientation angle α can be observed, while out of the filter plane the orientation angle β appears rather limited. This limitation is also implemented into the model. Finally, several simulations are performed for various operating conditions to infer an empirical correlation to predict the pressure drop. This correlation is validated using experimental data.

CONSTRUCTIVE DESIGN AND FEATURES OF THE PROCESSES IN THE ROASTING SHAFT FURNACES BY USING GASEOUS FUEL

The article analyzes shaft furnaces for lime production. According to the method of firing, shaft furnaces are of bulk type, semi-gas type, gaseous and liquid fuels. The disadvantages of the operation of the main modern kilns during lime burning have been established. The main aggregates for high-tonnage lime production are mine lime kilns, in which blast furnace coke or anthracite, or high-quality coals are used as fuel. Generally, lean fuels are used in industries where, together with the technological cycle, carbon dioxide is used, which are contained in flue gases with a concentration of 36−40%mass. The data on the operation of six furnaces on the territory of Ukraine at the enterprises showed that gas shaft furnaces of cylindrical and slotted type with a straight-lining profile were widely used. The parameters of such furnaces are productivity from 100 to 250 tons per day, shaft diameter from 3.2 to 5.4 m, and height 6−8 furnace diameters. Most often, they use heating systems with central and two rows of peripheral injection burners without protrusion. The most critical part of all furnaces is the distribution of gas flow rates over the cross-section of the furnace shaft and the depth of radial penetration of the gas flow into the material layer. Calculations of τ1 and τП were carried out in the following way: for τ1, the temperature range of the medium was chosen from 1300 to 250−300 °С, τox − the temperature range for the material was chosen from 1000 to 100 °С and the density of the product was taken into account at the degree of decomposition of limestone 98% (=1700 kg/m3). In the preheating zone, the limiting stage is the supply of heat from the gas to the lump material, and at the same time can be taken equal to ⁓400−450 kJ/(m2∙h∙К) and τ1 is estimated as 0.8÷1 hour. The total residence time of the material in the furnace (in the heating zone and in the firing zone) is estimated at 1.4÷1.5 hours. The average speed of material movement in the cylindrical part of the furnace is estimated as 0.004 m/s, which makes it possible to consider the filtering layer conditionally stationary. These characteristics affect the quality of limestone and the uniformity of limestone firing that was the basis for the design of roasting shaft kilns using gaseous fuel.