Layers Of Kernels Research Articles

A hybrid dynamic adversarial domain adaptation network with multi-channel attention mechanism for rotating machinery unsupervised fault diagnosis under varying operating conditions

Data-driven intelligent fault diagnosis methods have been extensively researched and applied in rotating machinery. In practical application scenarios, factors such as variable operating conditions and scarcity of labeled samples in rotating machinery hinder the engineering application and promotion of diagnostic models. To address these challenges, this paper proposes an unsupervised domain adaptation network called the Multi-scale Hybrid Domain Adaptation with Attention (MHDAA). Firstly, a multi-scale convolutional module was developed to extract fault features at different scales. Secondly, a multi-channel attention mechanism was proposed to enable the convolution layers of different convolution kernels fully extract feature information. Finally, a hybrid domain adaptation was constructed to dynamically extract invariant features from both the source and target domains. The method was evaluated in multiple transfer scenarios of planetary gearboxes and bearings. Experimental results demonstrate that the proposed method can effectively utilize fault features with high correlation from multiple source domains to complete fault diagnosis with unknown data labels in the target domain. Moreover, the proposed method exhibits superior diagnostic performance.

Biofortification of iron content by regulating a NAC transcription factor in maize.

Iron (Fe) deficiency remains widespread among people in developing countries. To help solve this problem, breeders have been attempting to develop maize cultivars with high yields and high Fe concentrations in the kernels. We conducted a genome-wide association study and identified a gene, ZmNAC78 (NAM/ATAF/CUC DOMAIN TRANSCRIPTION FACTOR 78), that regulates Fe concentrations in maize kernels. We cultivated maize varieties with both high yield and high Fe concentrations in their kernels by using a molecular marker developed from a 42-base pair insertion or deletion (indel) in the promoter of ZmNAC78. ZmNAC78 expression is enriched in the basal endosperm transfer layer of kernels, and the ZmNAC78 protein directly regulates messenger RNA abundance of Fe transporters. Our results thus provide an approach to develop maize varieties with Fe-enriched kernels.

Aberrant Behavior of Deoxynivalenol Production in the Malting of Fusarium Head Blight Infected Barley: Mycotoxin Accumulation and Hyphal Localization in Single Kernels

The malting industry faces a dilemma in which barley, initially containing low deoxynivalenol levels, occasionally exhibits an unexpected increase in deoxynivalenol levels during malting. In the current study, we investigated mycotoxin accumulation and hyphal localization in single kernels from Fusarium head blight infected barley and malt samples which showed an aberrant behavior of deoxynivalenol increases during malting. While deoxynivalenol levels in these bulk barley samples already ranged from <0.20 to 1.27 µg/g, they further increased on average by approximately 500% after malting, along with Fusarium growth. In addition, deoxynivalenol-3-glucoside and 3-acetyl deoxynivalenol levels increased from <0.20 µg/g to the mean levels of 2.47 µg/g and 1.03 µg/g in the bulk malt, respectively. In the investigation of single kernels, results showed that only 7% of barley single kernels (n = 385) had deoxynivalenol >1.00 μg/g, with the maximum level of 35.97 μg/g. However, 31% of malt single kernels contained deoxynivalenol >1.00 μg/g and up to 255.43 μg/g. Deoxynivalenol-3-glucoside and 3-acetyl deoxynivalenol were found in 40% of malt single kernels. Fungal hyphae were observed in the aleurone layer and embryo of barley kernels with deoxynivalenol >10.00 μg/g, in addition to the husk and vascular bundles of kernels with deoxynivalenol <1.00 μg/g. Hyphae spread largely in these tissues following malting, and even into the endosperm and pericarp cavities of extremely high deoxynivalenol kernels (i.e., >100.00 μg/g).

Insights into the Functional Components in Wheat Grain: Spatial Pattern, Underlying Mechanism and Cultivation Regulation.

Wheat is a staple crop; its production must achieve both high yield and good quality due to worldwide demands for food security and better quality of life. It has been found that the grain qualities vary greatly within the different layers of wheat kernels. In this paper, the spatial distributions of protein and its components, starch, dietary fiber, and microelements are summarized in detail. The underlying mechanisms regarding the formation of protein and starch, as well as spatial distribution, are discussed from the views of substrate supply and the protein and starch synthesis capacity. The regulating effects of cultivation practices on gradients in composition are identified. Finally, breakthrough solutions for exploring the underlying mechanisms of the spatial gradients of functional components are presented. This paper will provide research perspectives for producing wheat that is both high in yield and of good quality.

A Double Stream Person Re-Identification Method Based on Attention Mechanism and Multi-Scale Feature Fusion

Person re-identification is the key technology for designing a intelligent video surveillance system to retrieve and recongise a specific person in a series of videos. It mainly plays a role in multiple non-overlapping camera monitoring environments to determine whether the target person of interest that has appeared under a camera appears again under others. However, the image data, in the real scene, taken by the surveillance camera may be occluded or blurred, which increased the difficulty of identifying the pedestrian posturn and then lead to dramatically decrease the accuracy of recognition. To solve the above problems, we propose a dualbranch multi-scale feature fusion network, which improves the expression ability of pedestrian features under partial occlusion by learning discriminative pedestrian features. The feature erasure mechanism is introduced into the branch network to effectively enhance the fine-grained feature learning ability of the network. By embedding the lightweight attention module into Resnet-50, the image sequence features of the channel dimension will be extracted, and the interference caused by the cluttered back ground information will be suppressed. In the training phase, the average pooling layer and the maximum pooling layer of different kernels and strides are utilized for different residual stages, and the mixed pooling strategy and mixed loss function of different kernels are designed. Finally, a novel hybrid loss function which combines triplet loss and ID loss is presented to train the network model. By comparison with the existing representative methods on Market1501, DukeMTMC reID and MTMS17 datasets, the experimental results show that the features extracted by the proposed method are more discriminative and with high recognition accuracy.

Comparative characteristics of two methods for popping popcorn

Topicality. The comparative study on effectiveness of several methods of popping popcorn from Zea Mays L. everta Sturt. is important and relevant. Technological indicators of popcorn kernels depend on many factors, including the method of popping. Purpose. To compare the technological characteristics for popcorn kernels obtained by two methods of popping the grain: frying in a frying pan and heating with air-popping appliances. Methods. The initial material for research was kernels of popcorn lines bred by the Laboratory of Maize Breeding for Food Use at SE Institute of Grain Crops of NAAS. According to the first method of popping, one layer of popcorn kernels were placed on an aluminum pan heated to 175 ºC and fried without oil, shaking the kernels every 5–6 seconds. For the second popping method, Clatronic PM 3635 electric popcorn maker having a capacity of 1200 W was used. Results. In the first method of producing popcorn, the dominant type of popcorn flakes was unilaterally expanded; the average percentage of unpopped kernels was 3.5 %; the coefficient of kernel volume expansion varied from 16.4 to 29.7; and eating qualities were from 3 to 4.5 points. In the second method, the dominant type of popcorn flakes was bilateral expanded; the average percentage of unpopped kernels was 18.2 %; coefficient of kernel volume expansion ranged from 23.4 to 45.2; and eating qualities were from 3.5 to 5 points. For the first popping method, the coefficient of kernel volume expansion had an inverse average correlation with the percentage of unpopped kernels; for the second popping method, it had a direct average correlation with the thousand kernel weight. Conclusions. The method of popping popcorn affects the technological indicators of popcorn such as the number, weight and percentage of unpopped kernels; volume and weight of popcorn flakes; coefficient of kernel volume expansion; type of popped flakes and flavor of popcorn. Lines RB 20, RK 1, RK 44, RK 6 with kernel volume expansion coefficient over 40 and popcorn taste qualities of 5 points were selected. Key words: breeding, line, technological indicators, coefficient of kernel volume expansion, unpopped kernels, type of popped flakes, correlation

Rethinking Local and Global Feature Representation for Dense Prediction

Although fully convolution networks (FCNs) have dominated dense prediction tasks (e.g., semantic segmentation, depth estimation and object detection) for decades, they are inherently limited in capturing long-range structured relationship with the layers of local kernels. While recent Transformer-based models have proven extremely successful in computer vision tasks by capturing global representation, they would deteriorate dense prediction results by over-smoothing the regions containing fine details (e.g., boundaries and small objects). To this end, we aim to provide an alternative perspective by rethinking local and global feature representation for the dense prediction task. Specifically, we deploy a Dual-Stream Convolution-Transformer architecture, called DSCT, by taking advantage of both the convolution and Transformer to learn a rich feature representation, combining with a task decoder to provide a powerful dense prediction model. DSCT extracts high resolution local feature representation from convolution layers and global feature representation from Transformer layers. With the local and global context modeled explicitly in every layer, the two streams can be combined with a decoder to perform task of semantic segmentation, monocular depth estimation or object detection. Extensive experiments show that DSCT can achieve superior performance on the three tasks above. For semantic segmentation, DSCT builds a new state of the art on Cityscapes validation set (83.31% mIoU) with only 80,000 training iterations and appealing performance (49.27% mIoU) on ADE20K validation set, outperforming most of the alternatives. For monocular depth estimation, our model achieves 2.423 RMSE on KITTI Eigen split, superior to most of the convolution or Transformer counterparts. For object detection, without using FPN, we can achieve 44.5% APb on COCO dataset when using Faster R-CNN, which is higher than Conformer.

Bearings Intelligent Fault Diagnosis by 1-D Adder Neural Networks

Integrated with sensors, processors and RF communication modules, intelligent bearing could achieve the autonomous perception and autonomous decision-making, guarantying the safety and reliability during their use. However, because of the resource limitations of the end device, processors in the intelligent bearing are unable to carry the computational load of deep learning models like convolutional neural network (CNN), which involves a great amount of multiplicative operations. To minimize the computation cost of the conventional CNN, based on the idea of AdderNet, a 1-D adder neural network with a wide first-layer kernel (WAddNN) suitable for bearing fault diagnosis is proposed in this paper. The proposed method uses the l1-norm distance between filters and input features as the output response, thus making the whole network almost free of multiplicative operations. The whole model takes the original signal as the input, uses a wide kernel in the first adder layer to extract features and suppress the high frequency noise, then uses two layers of small kernels for nonlinear mapping. Through experimental comparison with CNN models of the same structure, WAddNN is able to achieve a similar accuracy as CNN models with significantly reduced computational cost. The proposed model provides a new fault diagnosis method for intelligent bearings with limited resources.

Cutting tool wear monitoring based on a smart toolholder with embedded force and vibration sensors and an improved residual network

Current milling tool condition monitoring (TCM) methods mainly rely on commercially available sensors with cutting force and vibration being the most widely used. However, they suffer from inconvenient installation or invasive machining when the installation position is close to the cutting area. Aiming at the problem of TCM, this paper proposes a TCM system based on a self-developed smart toolholder which is capable of sensing triaxial cutting force, torque and triaxial vibration simultaneously for the first time. The results of modal test and circuit system test demonstrate that the smart toolholder has a dynamic natural frequency of up to 1 kHz. The milling test results show that vibration sensed by the smart toolholder are much larger than the accelerometers installed on the spindle, workpiece and workbench in terms of maximum value, root mean square and kurtosis characteristics. Subsequently, a milling cutter with two physical vapor deposition (PVD) coated inserts was subjected to a full life cycle milling test on a 304 stainless steel workpiece. To build the TCM model, an improved residual network is proposed, which adopts the structure of the first layer of multi-scale large convolution kernels and the subsequent residual network, which can take the raw cutting force and vibration signals as input and automatically extract features. The results show that the established TCM model can identify the degree of tool wear up to 97.5% with lower fluctuations, which provides a practical solution for factory automated milling tool change.

Earthing, Inspiration, Humanity (on some tendencies in civilisatiins development in the XXI century)

This article is devoted to topical trends in the evolution of civilizations in XXI century. It is shown that the network processes of versatile and multi-layered anthropic exchange lay at the basis of the civilizations phenomenon which form both the world civilization and its local manifestations. It is also demonstrated that social cultural matrix (invariant, homeoresis), comprising in its turn several encoding layers and elements, is acting as the central factor in this process. Decoding of the encoding layers of different civilizational kernels allowed us to create a picture of the world evolutional wave movement in XXI century, which is shown in the article. This article provides the viewing of anthropic vector of this wave , when we have the cyclical turn from the West to the North, with the regulating civilizational chain of the Middle. The data of these chart in combination with the other sections of the article may be useful in the research of the current Big anthropological transition in XX-XXI centuries.

SleepFCN: A Fully Convolutional Deep Learning Framework for Sleep Stage Classification Using Single-Channel Electroencephalograms.

Sleep is a vital process of our daily life as we roughly spend one-third of our lives asleep. In order to evaluate sleep quality and potential sleep disorders, sleep stage classification is a gold standard method. In this paper, we introduce a novel fully convolutional neural network architecture (SleepFCN) to classify sleep stages into five classes using single-channel electroencephalograms (EEGs). The framework of SleepFCN includes two major parts for feature extraction and temporal sequence encoding namely multi-scale feature extraction (MSFE) and residual dilated causal convolutions (ResDC), respectively. These are then followed by convolutional layers of 1-sized kernels instead of dense layers to build the fully convolutional neural network. Due to the imbalance in the distribution of sleep stages, we incorporate a weight corresponding to the number of samples of each class in our loss function. We evaluated the performance of SleepFCN using the Sleep-EDF and SHHS datasets. Our experimental results show that the proposed method outperforms state-of-the-art works in both classification correctness and learning speed.

Dynamic Graph Message Passing Networks.

Modelling long-range dependencies is critical for scene understanding tasks in computer vision. Although convolution neural networks (CNNs) have excelled in many vision tasks, they are still limited in capturing long-range structured relationships as they typically consist of layers of local kernels. A fully-connected graph, such as the self-attention operation in Transformers, is beneficial for such modelling, however, its computational overhead is prohibitive. In this paper, we propose a dynamic graph message passing network, that significantly reduces the computational complexity compared to related works modelling a fully-connected graph. This is achieved by adaptively sampling nodes in the graph, conditioned on the input, for message passing. Based on the sampled nodes, we dynamically predict node-dependent filter weights and the affinity matrix for propagating information between them. This formulation allows us to design a self-attention module, and more importantly a new Transformer-based backbone network, that we use for both image classification pretraining, and for addressing various downstream tasks (e.g. object detection, instance and semantic segmentation). Using this model, we show significant improvements with respect to strong, state-of-the-art baselines on four different tasks. Our approach also outperforms fully-connected graphs while using substantially fewer floating-point operations and parameters. Code and models will be made publicly available at https://github.com/fudan-zvg/DGMN2.

Infrared object classification with a hybrid optical convolution neural network.

Recent advancements in machine vision have enabled a great range of applications from image classification to autonomous driving. However, there is still a dilemma between the pursuit of higher-resolution training images that require a detector array with more pixels on the front end, and the demands on acquisition for embedded systems restrained by power, transmission bandwidth, and storage. In this paper, a multi-pixel hybrid optical convolutional neural network machine vision system was designed and validated to perform high-speed infrared object detection. The proposed system replicates the front convolution layer in a convolutional neural network utilizing a high-speed digital micro-mirror device to display the first layer of kernels at a resolution greater than the subsequent detector. After this, further convolutions are carried out in software to perform the object recognition. An infrared vehicle dataset was used to validate the performance of the hybrid system through simulation. We also tested this in hardware by performing infrared classification on toy vehicles to showcase the feasibility of such a design.

Dynamic Wide and Deep Neural Network for Hyperspectral Image Classification

Recently, deep learning has been successfully and widely used in hyperspectral image (HSI) classification. Considering the difficulty of acquiring HSIs, there are usually a small number of pixels used as the training instances. Therefore, it is hard to fully use the advantages of deep learning networks; for example, the very deep layers with a large number of parameters lead to overfitting. This paper proposed a dynamic wide and deep neural network (DWDNN) for HSI classification, which includes multiple efficient wide sliding window and subsampling (EWSWS) networks and can grow dynamically according to the complexity of the problems. The EWSWS network in the DWDNN was designed both in the wide and deep direction with transform kernels as hidden units. These multiple layers of kernels can extract features from the low to high level, and because they are extended in the wide direction, they can learn features more steadily and smoothly. The sliding windows with the stride and subsampling were designed to reduce the dimension of the features for each layer; therefore, the computational load was reduced. Finally, all the weights were only from the fully connected layer, and the iterative least squares method was used to compute them easily. The proposed DWDNN was tested with several HSI data including the Botswana, Pavia University, and Salinas remote sensing datasets with different numbers of instances (from small to big). The experimental results showed that the proposed method had the highest test accuracies compared to both the typical machine learning methods such as support vector machine (SVM), multilayer perceptron (MLP), radial basis function (RBF), and the recently proposed deep learning methods including the 2D convolutional neural network (CNN) and the 3D CNN designed for HSI classification.

Assessment of unsteady flow predictions using hybrid deep learning based reduced-order models

In this paper, we present two deep learning-based hybrid data-driven reduced-order models for prediction of unsteady fluid flows. These hybrid models rely on recurrent neural networks (RNNs) to evolve low-dimensional states of unsteady fluid flow. The first model projects the high-fidelity time series data from a finite element Navier–Stokes solver to a low-dimensional subspace via proper orthogonal decomposition (POD). The time-dependent coefficients in the POD subspace are propagated by the recurrent net (closed-loop encoder–decoder updates) and mapped to a high-dimensional state via the mean flow field and the POD basis vectors. This model is referred to as POD-RNN. The second model, referred to as the convolution recurrent autoencoder network (CRAN), employs convolutional neural networks (instead of POD) as layers of linear kernels with nonlinear activations, to extract low-dimensional features from flow field snapshots. The flattened features are advanced using a recurrent (closed-loop manner) net and up-sampled (transpose convoluted) gradually to high-dimensional snapshots. Two benchmark problems of the flow past a cylinder and the flow past side-by-side cylinders are selected as the unsteady flow problems to assess the efficacy of these models. For the problem of the flow past a single cylinder, the performance of both the models is satisfactory and the CRAN model is found to be overkill. However, the CRAN model completely outperforms the POD-RNN model for a more complicated problem of the flow past side-by-side cylinders involving the complex effects of vortex-to-vortex and gap flow interactions. Owing to the scalability of the CRAN model, we introduce an observer-corrector method for calculation of integrated pressure force coefficients on the fluid–solid boundary on a reference grid. This reference grid, typically a structured and uniform grid, is used to interpolate scattered high-dimensional field data as snapshot images. These input images are convenient in training the CRAN model, which motivates us to further explore the application of the CRAN-based models for prediction of fluid flows.

Deep kernel learning in extreme learning machines

Emergence of extreme learning machine as a breakneck learning algorithm has marked its prominence in solitary hidden layer feed-forward networks. Kernel-based extreme learning machine (KELM) reflected its efficiency in diverse applications where feature mapping functions of hidden nodes are concealed from users. The conventional KELM algorithms involve only solitary layer of kernels, thereby emulating shallow learning architectures for its feature transformation. Trend in migrating shallow-based learning models into deep learning architectures opens up a new outlook for machine learning domains. This paper attempts to bestow deep kernel learning approach in a conventional shallow architecture. The emerging arc-cosine kernels possess the potential to mimic the prevailing deep layered frameworks to a greater extent. Unlike other kernels such as linear, polynomial and Gaussian, arc-cosine kernels have a recursive nature by itself and have the potential to express multilayer computation in learning models. This paper explores the possibility of building a new deep kernel machine with extreme learning machine and multilayer arc-cosine kernels. This framework outperforms conventional KELM and deep support vector machine in terms of training time and accuracy.

Label-Free Comparative Proteomic Analysis Combined with Laser-Capture Microdissection Suggests Important Roles of Stress Responses in the Black Layer of Maize Kernels

The black layer (BL) is traditionally used as an indicator for kernel harvesting in maize, as it turns visibly dark when the kernel reaches physiological maturity. However, the molecular roles of BL in kernel development have not been fully elucidated. In this work, microscopy images showed that BL began to appear at a growth stage earlier than 10 days after pollination (DAP), and its color gradually deepened to become dark as the development period progressed. Scanning electron microscopy observations revealed that BL is a tissue structure composed of several layers of cells that are gradually squeezed and compressed during kernel development. Laser-capture microdissection (LCM) was used to sample BL and its neighboring inner tissue, basal endosperm transfer layer (BETL), and outer tissue, inner epidermis (IEP), from 20 DAP of kernels. Matrix-assisted laser desorption/ionization time-of-flight mass spectrometry profiling (MALDI-TOF MS profiling) detected 41, 104, and 120 proteins from LCM-sampled BL, BETL, and IEP, respectively. Gene ontology (GO) analysis indicated that the 41 BL proteins were primarily involved in the response to stress and stimuli. Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis found that the BL proteins were enriched in several defense pathways, such as the ascorbate and aldarate metabolic pathways. Among the 41 BL proteins, six were BL-specific proteins that were only detected from BL. Annotations of five BL-specific proteins were related to stress responses. During kernel development, transcriptional expression of most BL proteins showed an increase, followed by a decrease, and reached a maximum zero to 20 DAP. These results suggest a role for BL in stress responses for protecting filial tissue against threats from maternal sides, which helps to elucidate the biological functions of BL.

Stress simulation and cracking prediction of corn kernels during hot-air drying

A moisture-heat transfer and stress model of corn kernels based on multi-component heterogeneity was developed and verified to predict the stress cracking during hot-air drying. Simulated results showed that the moisture gradient and von Mises stress of each corn component simulated by the verified model increased first and then decreased, where the von Mises stress in the outer layer of corn kernels was larger than that in the inner layer. The maximum von Mises stress of each corn component increased with the increase of drying temperature and the decrease of relative humidity. The stress cracks were mainly formed in hard endosperm and soft endosperm, because the maximum von Mises stress of hard endosperm was larger than the yield stress of the two components, which were agreement with statistical results of stress cracks. Therefore, a drying operation with lower temperature and higher relative humidity is beneficial for inhibiting the stress cracking in corn kernels.

Cultivar differences in lipoxygenase activity affect volatile compound formation in dough from wheat mill stream flour

Flavor, an important aspect of wheat (Triticum aestivum) dough products, forms during preparation and is affected by the generation of volatile compounds. These compounds are produced mainly by the oxidation of unsaturated fatty acids by lipoxygenase (LOX). Here, we analyzed the volatile compounds emitted from dough made from wheat mill stream flour to clarify the cultivar differences in LOX distribution and its effect on the formation of volatile compounds. LOX activity increased with ash content, and the rate of increase was significantly higher for Norin 61 than for other wheat cultivars. The volatile compound profiles were analyzed by principal component analysis, with almost all volatile compounds correlated to ash content. The volatile compounds in some Norin 61 streams, however, were more characteristic of the outer layer of wheat kernels than those in the other cultivars. Since the ash content among cultivars was similar, the characteristic volatile compound profiles of Norin 61 were presumed to be caused by higher LOX activity in this cultivar. In conclusion, the characteristic volatile compound profiles in each cultivar are due to the differences of LOX activity.

Multi-resolution networks for ship detection in infrared remote sensing images

Ship detection is a hot point of remote sensing image processing and analysis. Most current methods merely utilize a single resolution image and cannot take full advantage of registered multi-resolution images. In this paper, we proposed a novel ship detection method for multi-resolution infrared remote sensing images based on convolutional neural networks. With a specially designed layer of various size kernels, our net takes in registered multi-resolution images simultaneously, extracting more robust features and gives a more accurate detection result. Besides, as ship targets only occupy a few pixels in an infrared remote sensing image, to handle the extreme background-foreground imbalance problem in infrared remote sensing images, we introduce the focal loss to train our net. Compared with classical methods, experiments on visible bands and infrared bands of Landsat-8 satellite images demonstrate the effectiveness of our method.