The performance analysis of model parallelism in deep residual network

Data-driven rolling-bearing fault diagnosis methods are mostly based on deep-learning models, and their multilayer nonlinear mapping capability can improve the accuracy of intelligent fault diagnosis. However, problems such as gradient disappearance occur as the number of network layers increases. Moreover, directly taking the raw vibration signals of rolling bearings as the network input results in incomplete feature extraction. In order to efficiently represent the state characteristics of vibration signals in image form and improve the feature learning capability of the network, this paper proposes fault diagnosis model MTF-ResNet based on a Markov transition field and deep residual network. First, the data of raw vibration signals are augmented by using a sliding window. Then, vibration signal samples are converted into two-dimensional images by MTF, which retains the time dependence and frequency structure of time-series signals, and a deep residual neural network is established to perform feature extraction, and identify the severity and location of the bearing faults through image classification. Lastly, experiments were conducted on a bearing dataset to verify the effectiveness and superiority of the MTF-ResNet model. Features learned by the model are visualized by t-SNE, and experimental results indicate that MTF-ResNet showed better average accuracy compared with several widely used diagnostic methods.

Deep residual networks (ResNets) can learn deep feature representation from hyperspectral images (HSIs), and therefore have been widely used for HSI classification. Despite their high accuracies, there still exist a lot of challenging cases, such as open world recognition, limited-sample learning and visualization of learned classification features, which cannot be well addressed. Most of the challenges in HSI classification can be attributed to the dependence on softmax based loss function and classifiers, which cause the lack of robustness for deep learning models and the hardness to visualize the learned classification features. To improve the robustness and achieve the visualization of learned classification features, we propose a novel learning framework called Residual Prototype Learning Network, a combination of residual network and prototypes learning mechanism. Under the framework, a prototype learning based loss function is proposed to enhance intra-class compactness and the inter-class separation of these feature representations; in addition, a prototype learning based classifier is simultaneously proposed to achieve the 2D or 3D visualization of the classification features. The effectiveness of our proposed learning framework is evaluated on several publicly available HSI benchmarks, and the experimental results show that our approach achieve better results than traditional softmax based ResNets.

Offset computation for expanding a polyhedral object by an offset radius is a fundamental geometric process frequently used in manufacturing applications. This process combined with the triple-dexel representation solid model has become popular because of its robustness and compatibility with parallel processing using a graphics processing unit (GPU). In parallel geometric processing, conflicts between threads must be avoided. Thus, we propose a novel parallel offsetting algorithm free from conflicts between threads. The triple-dexel model is a combination of x-, y-, and z-axis-aligned dexel models. Each dexel model is defined based on an orthogonal grid given on a coordinate plane. We subdivide the grid into several sub-grids of a fixed size in advance. For each sub-grid, a block of GPU threads is assigned. As each GPU thread always processes different dexel elements from the other threads in this method, no conflict occurs. Our research group has previously presented a parallel offset computation algorithm for a polyhedral solid model that also uses a triple-dexel representation model and a GPU. In the previous algorithm, the surface polygons of the model are classified into several groups in advance. The parallel offset computation of multiple polygon groups is realized by selecting groups of polygon in which the offset processing of the polygons does not affect one another. This selection process is time-consuming. Computational experiments were performed to analyze the performance difference between the current algorithm and our previous algorithm. In our experiments, the current algorithm achieved speedups of 1.4 times to 3.2 times compared to our previous offsetting algorithm.

Performance Analysis and Optimization of Graphics Processing Unit

Surface reflectance (SR) is among the most basic parameters needed for deriving higher- level land products, such as crop classification and burned area, from satellite remote sensors. SR is traditionally derived from Top Of Atmosphere (TOA) reflectance observed by polar-orbiting satellites using atmospheric parameters that are difficult to measure and unstable across locations. While historically used for weather observation, contemporary geostationary satellites, such as GOES-R and Himawari-8, carry sensors similar to polar satellites like NASA's Moderate Resolution Imaging Spectroradiometer (MODIS). Here, we examine the hypothesis that deep learning will contribute to the efficient processing of remote sensing data. A deep learning methodology is proposed for predicting MODIS-like SR from geostationary satellite data using deep residual networks. Missing data in irregularly shaped areas due to clouds, aerosol and water bodies is one of the key challenges in processing remote sensing data. An image inpainting technique using partial convolution is applied to handle missing pixels. We demonstrate that deep learning methods (residual network and convolutional neural network) produce low error predictions for MODIS-like SR (RMSE = 0.02). Further, Bayesian uncertainty analyses are adapted to quantify and characterize predictive uncertainty.

Various microfossils from the early Cambrian provide crucial clues for understanding the Cambrian explosion and the origin of animal phyla. However, specimens with important anatomical structures are extremely rare and the efficiency of retrieving such fossils by traditional manual selection under a microscope is quite low. Such a contradiction has hindered breakthroughs in micropaleontology for a long time. Here, we propose a solution for identifying specific taxa of Cambrian microfossils using only a few available specimens by transferring a model pre-trained on natural image datasets to the field of paleontological artificial intelligence. The method employs a 34-layer deep residual neural network as the underlying framework, migrates the ImageNet pre-trained model, freezes the low-layer network parameters and retrains the high-layer parameters to build a microfossil image recognition model. We built training sets with randomly selected images of varied number for each taxon. Our experiments show that the average recognition accuracy for specific taxa of Cambrian microfossils (50 images for each taxon) is higher than 0.97 and it can reach 0.85 with only three training samples per taxon. Comparative analyses indicate that our results are much better than those of various prevalent methods, such as the transpose convolutional neural network (TCNN). This demonstrates the feasibility of using natural images (ImageNet) for the training of microfossil recognition models and provides a promising tool for the discovery of rare fossils.

Aiming at the problem that the difference in fault data distribution caused by the variation of gear and bearing conditions during the operation of rotating machinery affects the accuracy of fault diagnosis, a multi-condition fault diagnosis approach based on whale optimization algorithm (WOA), variational mode decomposition (VMD), and deep residual shrinkage network (DRSN) is proposed. Firstly, WOA is employed to optimize the parameters of the decomposition level and the penalty factor value in VMD, enhancing its noise resistance. Subsequently, VMD processing yields several modal components, each characterized by a permutation entropy feature vector, ensuring optimal feature extraction. DRSN is then utilized, employing an attention mechanism to adaptively set soft threshold functions, thereby improving gear fault diagnosis accuracy under multiple operating conditions. Finally, the proposed method’s performance was validated through the collection of gear running signals under various conditions, and the fault diagnosis accuracy was used as the model evaluation index to conduct comparative experiments and ablation experiments. The experimental results indicate that the WOA-VMD feature extraction and DRSN fault diagnosis techniques outperform traditional methods in multi-condition gear fault diagnosis. This study provides a new effective approach for multi-condition diagnosis of rotating machinery.

Gait Recognition using Deep Residual Networks and Conditional Generative Adversarial Networks

This Human Skin Colour detection is analyzed to be a most suitable and strong cue for face detection and face recognition which finds various application in domain like surveillance, face biometrics, gesture recognition, interactive game application and many other. These applications require fast image processing in real time and demands enormously high performance with respect to time and processing speed. Earlier work done on sequential architecture does not provide required capability and can be achieved by parallel programming. As Image processing applications shows high degree of parallelism, they prove excellent source for multi-core platform. We can use GPU multithreaded parallel computing techniques to improve the speed of image processing. The paper proposes a data parallelism programming model for Human Skin Colour Detection algorithm. The objective is to increase the computational speed of the algorithm through data parallelism using CUDA framework. The framework is supported by OpenCV libraries and implemented by GPU (Graphics Processing Unit). The evaluation is done on basis of comparative analysis of serial and parallel programming computing. The pixel based Explicitly Defined Region skin classifier is used and RGB colour space is chosen due to its wider use in storing and processing of digital image data. To gain substantial acceleration in image computations, skin classifier code is off-loaded to GPU having compute capacity of 3.5 and rest of code is executed on CPU. We have tested module on static images, as well as on live camera captured images. SFA standard dataset is utilized to evaluate performance. Speedup of 23.45% was achieved using parallel programming.

Regional land-use change is the leading cause of ecosystem carbon stock change; it is essential to investigate the response of LUCC to carbon stock to achieve the strategic goal of “double carbon” in a region. This paper proposes a residual network algorithm, the Residual Multi-module Fusion Network (R-MFNet), to address the problems of blurred feature boundary information, low classification accuracy, and high noise, which are often encountered in traditional classification methods. The network algorithm uses an R-ASPP module to expand the receptive field of the feature map to extract sufficient and multi-scale target features; it uses the attention mechanism to assign weights to the multi-scale information of each channel and space. It can fully preserve the remote sensing image features extracted by the convolutional layer through the residual connection. Using this classification network method, the classification of three Landsat-TM/OLI images of Zhengzhou City (the capital of Henan Province) from 2001 to 2020 was realized (the years that the three images were taken are 2001, 2009, and 2020). Compared with SVM, 2D-CNN, and deep residual networks (ResNet), the overall accuracy of the test dataset is increased by 10.07%, 3.96%, and 1.33%, respectively. The classification achieved using this method is closer to the real land surface, and its accuracy is higher than that of the finished product data obtained using the traditional classification method, providing high-precision land-use classification data for the subsequent carbon storage estimation research. Based on the land-use classification data and the carbon density data corrected by meteorological data (temperature and precipitation data), the InVEST model is used to analyze the land-use change and its impact on carbon storage in the region. The results showed that, from 2001 to 2020, the carbon stock in the study area showed a downward trend, with a total decrease of 1.48 × 107 t. Over the course of this 19-year period, the farmland area in Zhengzhou decreased by 1101.72 km2, and the built land area increased sharply by 936.16 km2. The area of land transfer accounted for 29.26% of the total area of Zhengzhou City from 2001 to 2009, and 31.20% from 2009 to 2020. The conversion of farmland to built land is the primary type of land transfer and the most important reason for decreasing carbon stock. The research results can provide support, in the form of scientific data, for land-use management decisions and carbon storage function protections in Zhengzhou and other cities around the world undergoing rapid urbanization.

The aim of this paper is to provide a deep learning based method that can solve high-dimensional Fredholm integral equations. A deep residual neural network is constructed at a fixed number of collocation points selected randomly in the integration domain. The loss function of the deep residual neural network is defined as a linear least-square problem using the integral equation at the collocation points in the training set. The training iteration is done for the same set of parameters for different training sets. The numerical experiments show that the deep learning method is efficient with a moderate generalization error at all points. And the computational cost does not suffer from “curse of dimensionality” problem.

The development of the Internet-of-Things (IoT) security is comparatively slower than the pace of the IoT innovations. The seamless IoT network operates in an untrusted environment and is exposed to many malicious active attacks. As the process of identifying the modulation format of signals is corrupted by noise and fading, automatic modulation classification (AMC) can be viewed as an effective approach to counter physical-layer threats for IoT as it can detect and identify the pilot jamming, deceptive jamming, and Sybil attacks. Nowadays, data-driven deep learning (DL) techniques, which are capable of extracting discriminative features and perform better robustness to channel and noise conditions, have drawn widespread attention. The deep residual network (ResNet) has a strong representative ability, which can learn latent information repeatedly from the received signals and improve the classification accuracy. Meanwhile, the gated recurrent unit (GRU), which is capable of exploiting temporal information of the received signal can expand the dimension of the signal features for satisfactory classification performance. Considering the advantages of the above networks, this article proposes a novel gated recurrent residual neural network (GrrNet) for feature-based AMC, where the amplitude and phase of the received signal are utilized as the inputs of GrrNet. In GrrNet, a ResNet extractor module is first designed to extract the highly representative features and then temporal information is obtained by the subsequent GRU module which is capable of processing the representative features with the arbitrary length for modulation classification. Moreover, extensive simulations are conducted to verify the classification performance and robustness of the proposed GrrNet and it is shown that GrrNet outperforms other recent DL-based AMC methods. Moreover, the influence of the network parameters, symbol length, and frequency offset on performance is also explored.

This paper proposes a method to evaluate the degradation stages of stability of the screw with data fusion technology and deep residual neural network. Firstly, the data provided by multi-sensors are fused, and then the time domain images of the signals are input into the deep residual neural network for training and testing. The effectiveness of the proposed method is verified using data sets collected from the degradation test bench of ball screw. The data sets contain massive samples involving 3 degradation stages under 9 working conditions obtained by 3 accelerometers. From the results and comparison, it can be found that the deep residual neural network can automatically extract the features from the original data layer by layer and achieve a better evaluation rate when it is used to evaluate the performance stage of the screw, and the multi-sensor data fusion vibration data can better reflect the degradation stages of the screw's performance. It is showed that this method enhances the intelligence of the evaluation compared with the traditional method.

Abstract. Thanks to the nature of the graphics processing, the newly released products offer highly parallel processing units with high-memory bandwidth and computational power of more than teraflops per second. The modern GPUs are not only powerful graphic engines but also they are high level parallel programmable processors with very fast computing capabilities and high-memory bandwidth speed compared to central processing units (CPU). Data-parallel computations can be shortly described as mapping data elements to parallel processing threads. The rapid development of GPUs programmability and capabilities attracted the attentions of researchers dealing with complex problems which need high level calculations. This interest has revealed the concepts of “General Purpose Computation on Graphics Processing Units (GPGPU)” and “stream processing”. The graphic processors are powerful hardware which is really cheap and affordable. So the graphic processors became an alternative to computer processors. The graphic chips which were standard application hardware have been transformed into modern, powerful and programmable processors to meet the overall needs. Especially in recent years, the phenomenon of the usage of graphics processing units in general purpose computation has led the researchers and developers to this point. The biggest problem is that the graphics processing units use different programming models unlike current programming methods. Therefore, an efficient GPU programming requires re-coding of the current program algorithm by considering the limitations and the structure of the graphics hardware. Currently, multi-core processors can not be programmed by using traditional programming methods. Event procedure programming method can not be used for programming the multi-core processors. GPUs are especially effective in finding solution for repetition of the computing steps for many data elements when high accuracy is needed. Thus, it provides the computing process more quickly and accurately. Compared to the GPUs, CPUs which perform just one computing in a time according to the flow control are slower in performance. This structure can be evaluated for various applications of computer technology. In this study covers how general purpose parallel programming and computational power of the GPUs can be used in photogrammetric applications especially direct georeferencing. The direct georeferencing algorithm is coded by using GPGPU method and CUDA (Compute Unified Device Architecture) programming language. Results provided by this method were compared with the traditional CPU programming. In the other application the projective rectification is coded by using GPGPU method and CUDA programming language. Sample images of various sizes, as compared to the results of the program were evaluated. GPGPU method can be used especially in repetition of same computations on highly dense data, thus finding the solution quickly.

Automatic modulation classification (AMC) is a signal processing technology used to identify the modulation type of unknown signals without prior information such as modulation parameters for drone communications. In recent years, deep learning (DL) has been widely used in AMC methods due to its powerful feature extraction ability. The significant performance of DL-based AMC methods is highly dependent on large amount of data. However, with the increasingly complex signal environment and the emergence of new signals, several recognition tasks have difficulty obtaining sufficient high-quality signals. To address this problem, we propose an AMC method based on a deep residual neural network with masked modeling (DRMM). Specifically, masked modeling is adopted to improve the performance of a deep neural network with limited signal samples. Both complex-valued and real-valued residual neural networks (ResNet) play an important role in extracting signal features for identification. Several typical experiments are conducted to evaluate our proposed DRMM-based AMC method on the RadioML 2016.10A dataset and a simulated dataset, and comparison experiments with existing AMC methods are also conducted. The simulation results illustrate that our proposed DRMM-based AMC method achieves better performance in the case of limited signal samples with low signal-to-noise ratio (SNR) than other existing methods.