Search

Generally, the performance of deep learning-based classification models is highly related to the captured features of training samples. When a sample is not clear or contains a similar number of features of many objects, we cannot easily classify what it is. Actually, human beings classify objects by not only the features but also some information such as the probability of these objects in an environment. For example, when we know further information such as one object has a higher probability in the environment than the others, we can easily give the answer about what is in the sample. We call this kind of probability as local probability as this is related to the local environment. In this paper, we carried out a new framework that is named L-PDL to improve the performance of deep learning based on the analysis of this kind of local probability. Firstly, our method trains the deep learning model on the training set. Then, we can get the probability of objects on each sample by this trained model. Secondly, we get the posterior local probability of objects on the validation set. Finally, this probability conditionally cooperates with the probability of objects on testing samples. We select three popular deep learning models on three real datasets for the evaluation. The experimental results show that our method can obviously improve the performance on the real datasets, which is better than the state-of-the-art methods.

Currently, deep learning methods have been widely applied to many fields like classification. Generally, these methods use the technology like transferring to make a model work well on different do...

Currently, image denoising is a challenge in many applications of computer vision. The existing denoising methods depend on the information of noise types or levels, which are generally classified by experts. These methods have not applied computational methods to pre-classify the image noise types. Furthermore, some methods assume that the noise type of the image is a certain one like Gaussian noise, which limits the ability of the denoising in real applications. Different from the existing methods, this paper introduces a new method that can classify and denoise not only a certain type noise but also mixed types of noises for real demand. Our method utilizes two types of deep learning networks. One is used to classify the noise type of the images and the other one performs denoising based on the classification result of the first one. Our framework can automatically denoise single or mixed types of noises with these efforts. Our experimental results show that our classification network achieves higher accuracy, and our denoising network can ensure higher PSNR and SSIM values than the existing methods.

3D object detection is widely used in the field of autonomous driving. The embedded devices have few resources and low power consumption, so the existing 3D object detection algorithms are difficult to complete the detection task in the required time under some conditions. In this paper, the Complex-YOLOv4-Trans model is proposed to use point cloud data for 3D object detection. Firstly, a transformer encoder block is introduced to improve the structure of Backbone, which makes full use of its context information to improve detection accuracy. Secondly, the general convolution in Backbone is replaced with a depth-wise separable convolution, which effectively reduces the computational load of the model. Finally, Complete-IoU (CIoU) is used as the loss function, and then 8-bit model quantization is performed to speed up the inference. We select the KITTI dataset and deploy the model on NVIDIA Jetson Xavier NX by using TensorRT acceleration tools to test the methods. The mAP of Bird’s Eye View for Car is 71.79%, the mAP for 3D Detection is 15.82%, and the FPS on the NX device is 42 frames. Experimental results show that our Complex-YOLOv4-Trans can perform real-time 3D object detection tasks on low-energy embedded devices.

Recently, deep learning has developed rapidly and contributed in many fields like the classification in radar and sonar applications. In some special fields like the underwater acoustic signals, th...

In real-world applications, different problems can adopt different models. Most of the existing denoising methods use the framework of deep learning, and the most commonly used denoised algorithm evaluation indicators, such as PSNR, MSE, etc., all without exception, require pictures’ ground truth which is needed as a reference. However, there are few real and noise-free pictures in the field of image denoising, only the noise reduction map can be compared with the noise map, which seems to be less persuasive. Therefore, this paper proposes a new criterion for judging the denoising model. The most important thing is that this method does not require noiseless images compared to PSNR when testing. Moreover, we improved the denoising model and verified the reliability of the criterion. At the same time, we conduct statistics on the recognition rate of different types of targets, and analyze the trend of misjudgment. In this paper, the synthetic aperture radar (SAR) image dataset is used as an experimental sample, and different noise parameters are used to obtain denoising data sets with different noise levels. Then we use different denoising models such as DN-CNN to process the data set. Finally, the CNN classification model is used for screening comparison. In this paper, the experimental results show that it is feasible to use classification to judge denoising, so based on this feasibility, this paper modified the denoising network and used classification to judge. The results show that the denoising effect is better and the classification accuracy is higher, that is, the denoising and classification are a complementary relationship.

Objective Systematic reviews and meta-analyses have revealed the associations between H. pylori infection and various health outcomes. We aimed to evaluate the strength and breadth of evidence on the associations. Design Umbrella review of systematic reviews and meta-analyses. Setting No settings. Participants No patients involved. Data sources Embase, PubMed, Web of Science, Cochrane Library Databases, CNKI, VIP database and Wangfang database from inception to February 1, 2019. Outcomes measures Diverse diseases (such as cancer and ischaemic heart disease). Results Sixty articles reporting 88 unique outcomes met the eligible criteria. 74 unique outcomes had nominal significance (p75%). Moreover, 20% exhibited publication bias (p

Generally, on the same data set, different deep learning classification models will achieve different performances. The existing weighted voting method can combine the results of models, which can improve the performance of classification. On the other side, its classification accuracy is affected by the accuracy of all models. In this paper, we proposed a dynamic weighted voting method. Our method dynamically selects models on different data sets, and integrates them according to their weights, thereby improving the classification accuracy. We evaluated the methods on three data sets of CIFAR10, CIFAR100 and Existing, which increased the accuracy about 0.65%, 0.91%, and 0.78% respectively compared with the existing weighted voting method.

Recently, deep learning has developed rapidly, which has made significant progress in tasks such as target detection and classification. Compared with traditional methods, using deep learning techniques contribute to achieve higher detection accuracy, recognition rate, and other better performance with big data set. In the fields of radar and sonar especially like underwater acoustic signals, training samples are scarce due to the difficulty of the collection or security reason, which leads to poor performance of the classification models, as those need big training samples. In this paper, we present a novel framework based on Generative Adversarial Networks (GAN) to resolve the problem of insufficient samples for the underwater acoustic signals. Our method preprocesses the audio samples to the gray-scale spectrum images, so that, those can fit the GAN to captures the features and reduce the complexity at the same time. Then our method utilizes an independent classification network outside the GAN to evaluate the generated samples by GAN. The experimental results show that the samples generated by our approach outperform existing methods with higher quality, which can significantly improve the prediction accuracy of the classification model.

ObjectiveSystematic reviews and meta-analyses have revealed the associations betweenH. pyloriinfection and various health outcomes. We aimed to evaluate the strength and breadth of evidence on the associations.DesignUmbrella review of systematic reviews and meta-analyses.SettingNo settings.ParticipantsNo patients involved.Data sourcesEmbase, PubMed, Web of Science, Cochrane Library Databases, CNKI, VIP database and Wangfang database from inception to February 1, 2019.Outcomes measuresDiverse diseases (such as cancer and ischaemic heart disease).ResultsSixty articles reporting 88 unique outcomes met the eligible criteria. 74 unique outcomes had nominal significance (p2=50%–75%) and 24 had high heterogeneity (I2>75%). Moreover, 20% exhibited publication bias (pH. pyloriinfection may be associated with an increased risk of five diseases and a decreased risk of irritable bowel syndrome.ConclusionAlthough 60 meta-analyses explored 88 unique outcomes, moderate quality evidence only existed for six outcomes with statistical significance.H. pyloriinfection may be associated with a decreased risk of irritable bowel syndrome and an increased risk of hypertriglyceridemia, chronic cholecystitis and cholelithiasis, gestational diabetes mellitus, gastric cancer and systemic sclerosis.Trial registrationCRD42019124680.

Lung cancer is the most common cancer that cannot be ignored and cause death with late health care. Currently, CT can be used to help doctors detect the lung cancer in the early stages. In many cases, the diagnosis of identifying the lung cancer depends on the experience of doctors, which may ignore some patients and cause some problems. Deep learning has been proved as a popular and powerful method in many medical imaging diagnosis areas. In this paper, three types of deep neural networks (e.g., CNN, DNN, and SAE) are designed for lung cancer calcification. Those networks are applied to the CT image classification task with some modification for the benign and malignant lung nodules. Those networks were evaluated on the LIDC-IDRI database. The experimental results show that the CNN network archived the best performance with an accuracy of 84.15%, sensitivity of 83.96%, and specificity of 84.32%, which has the best result among the three networks.

Understanding and predicting electromagnetic behavior is very important in electrical engineering field. The finite element method (FEM) is a powerful tool for the electromagnetic phenomena. However, the FEM in electromagnetic problems is computationally intensive, which can run for a few hours to several days in lager scale problems. Thus, accelerating FEM is one of the most significant works. The Jacobi iterative method is the technique widely used for solving linear equations that is the most time-consuming step in FEM. In this paper, a hardware Jacobi solver is designed and implemented using a pipelined architecture and fine-grained parallelism. The hardware solver is applied to the electromagnetic boundary value problem (BVP) to verify the implementation and explore the hardware speed and cost.

Multiple vector reduction is one of the most fundamental and widely used operations in scientific computation. It is an attractive option to accelerate this computation program using FPGA as hardware accelerators. However, core operator of the vector reduction is usually floating point units which are often deeply pipelined, improper control may destroy the benefit from pipelining or impose unrealistic buffer requirements. In this paper, we propose a new approach on FPGA for multiple vector reduction. The design contributes an interleave-based vector reduction approach with high pipeline utilization. The proposed approach significantly simplifies the control units of the vector reduction operations. The sparse matrix vector multiplication is implemented as an application example in this paper.

A fast and reliable signal reconstruction algorithm is the core part of the compressed sensing (CS) theory. As a reconstruction algorithm, interior point method has a high precision but it is time-consuming and needs large computation, so it is difficult to meet the actual needs. In view of the above problems, in this paper we propose a design for interior point method that is based on the Field Programmable Gate Array (FPGA )hardware platform, which is about the solution of linear equations that have the largest amount of computation. The array structure of the conjugate gradients (CG) coprocessor completes the main operations, and the parallel and pipeline coprocessor effectively take advantage of the inherent parallelism of the algorithm and the parallel structure of FPGA. Thus, it has greatly improved the processing speed.

Vector-reduction operation is the basis of many scientific computations. FPGA-based vector-reduction circuit must use deeply pipelined floating-point IP cores to gain a performance advantage over general-purpose processor (GPP). Improper design of reduction circuit will destroy the benefit from pipelining or impose unrealistic buffer requirements. In this paper, a high-performance improved reduction method is proposed and analyzed for FPGA platform. This design runs in optimal time while requires only four buffers of fixed size and a single pipelined floating-point unit. Using ALTERA Cyclone II EP2C70F896C6 as the target device, we implement vector summation which is most common example of vector-reduction using improved reduction method.