Search

Pyroelectric infrared (PIR) sensors are much promising for device-free localization (DFL) due to their advantages of lower cost, low power consumption, and privacy protection. Most PIR-based localization methods usually assume some geometric models according to the detection principle of PIR sensors, which are however not accurate or robust due to the various cases of infrared radiation from human body, especially the case of multiple persons. Recently, deep learning is utilized in the PIR-based localization method (i.e. PIRNet Yang et al.) and well handles the complex infrared radiation even in the multi-person case. However, this method requires a high training cost, and has very weak generalization ability as it assumes the PIR sensors’ deployment in the testing environment is same to the deployment in training environment. To reduce the training cost and achieve high generalization ability, in this paper, we propose a robust method DeepPIRATES, which can be directly utilized in various deployment scenarios without retraining. DeepPIRATES combines deep learning and a geometric model. Specifically, DeepPIRATES divides the localization task into two steps. The first step utilizes a neural network to estimate the azimuth changes of multiple persons to a PIR sensor. Then, DeepPIRATES utilizes the persons’ azimuth changes to infer their locations based on a geometric model. Extensive experimental results show that DeepPIRATES can achieve similar localization accuracy as PIRNet, while does not require to be retrained when the sensor deployment changes.

Accurate and reliable prediction of air-conditioning load plays a significant role in prosumer energy management system (EMS), because air-conditioning load accounts for a large proportion of the building's total energy consumption. This paper proposes a new meta ensemble learning method to realize short-term prediction of air-conditioning load for prosumers. This method is a hybrid of meta ensemble learning and stacked auto-encoder (SAE). First, we design multiple different forecasting structures based on SAE to achieve point prediction of air-conditioning loads. SAE is used to learn the deep features in air-conditioning load data. Second, a new meta ensemble learning prediction model is proposed. Meta ensemble learning is used to learn the nonlinear features and invariant structures in data, and determine the coefficients of each SAE-based point forecaster. Finally, the prediction results of each point forecaster are aggregated and integrated to estimate the final air-conditioning load prediction result. Air-conditioning load data from a commercial building in Singapore are used to validate the feasibility and effectiveness of the proposed method, demonstrating that the proposed meta ensemble learning method is attractive in prosumer energy management.

Human papillomavirus (HPV) remains a leading cause of virus-induced cancers and has a typical size of 52 to 55nm in diameter. Hence conventional light microscopy that usually sustains a resolution at ~100nm per pixel falls short of detecting it. This study explores four state of the art generative adversarial networks (GANs) for visualising HPV. The evaluation is achieved by counting the HPV clusters that are corrected identified as well as drug treated cultured cells, i.e. no HPVs. The average sensitivity and specificity are 78.81%, 76.37%, 76.62% and 84.71% for CycleGAN, Pix2pix, ESRGAN and Pix2pixHD respectively. For ESRGAN, the training takes place by matching pairs between low and high resolution (x4) images. For the other three networks, the translation is performed from original raw images to their coloured maps that have undertaken Gaussian filtering in order to discern HPV clusters visually. Pix2pixHD appears to perform the best.

This paper focuses on analyzing and studying the application of Recurrent Neural Network (RNN), a deep learning method, in news text classification. Aiming at the problem that the RCNN model cannot fully extract the text context feature information in text classification, and the training complexity of the LSTM method is high, a DGCNN model text classification method based on deep learning is proposed. The DGCNN model adopts dual-channel forward and backward bidirectional threshold recurrent unit (D-GRU) to fully obtain the text context feature information. Tests on the Sogou news corpus show that the improved model has a good classification effect, improving the accuracy, precision, recall and F 1 value of the classification.

Because of the important diagnostic value, computed tomography (CT) has been widely used in clinic. However, there are still some limitations in CT examination of small intestinal lesions. At present, the diagnosis of these lesions mainly depends on doctors' experience in CT image so that there would be some problems such as high labor cost, low efficiency and misdiagnosis risk. Therefore, a segmentation and identification method for small intestinal lesions based on U-Net with batch normalization (BN) and virtual sample, named as U-NetBV, is proposed in this paper. BN can help to improve the performance of the network and virtual sample can increase the number of samples which is usually small in CT image. Through U-NetBV, these lesions can be identified and segmented automatically. The experimental results show that the proposed U-NetBV has a promising prospect in the field of segmentation and identification of small intestinal lesions.

Federated Learning (FL) is a promising distributed learning paradigm and has gained recent attention from both academia and industry. One challenge in FL is that when local data across different devices are not independent and identically distributed (non-IID), models trained using FL generally have degraded performance. To address the problem, one natural approach is clustering: clients with similar data distributions are grouped into the same clusters and each cluster trains a specialized model. However, features utilized for clustering generally rely on a single global model trained during FL, whose convergence usually incurs high communication cost. In this paper, we propose CAFL, an energy-efficient clustering method in FL. In CAFL, clustering features of a client are not based on a collaboratively trained global model by FL, but a tensor of gradient vectors computed on local data. With this approach, the communication overhead for clustering is greatly reduced. We validated CAFL on simulated datasets include Fashion-MNIST and CIFAR-10, and the results show that compared with existing clustering methods in FL, CAFL has much lower communication cost while still ensuring a high clustering accuracy.

Quick Response (QR) code is one of the most worldwide used two-dimensional codes. Traditional QR codes appear as random collections of black-and-white modules that lack visual semantics and aesthetic elements, which inspires the recent works to beautify the appearances of QR codes. However, these works adopt fixed generation algorithms and therefore can only generate QR codes with a pre-defined style. In this paper, combining the Neural Style Transfer technique, we propose a novel end-to-end method, named ArtCoder, to generate the stylized QR codes that are personalized, diverse, attractive, and scanning-robust. To guarantee that the generated stylized QR codes are still scanning-robust, we propose a Sampling-Simulation layer, a module-based code loss, and a competition mechanism. The experimental results show that our stylized QR codes have high-quality in both the visual effect and the scanning-robustness, and they are able to support the real-world application.

The polarization states of scattered photons can be used to map or image the anisotropic features of a nanostructure. However, the scattering strength depends heavily on the refractivity contrast in the near field under measurement, which limits the imaging sensitivity for viral particles which have little refractivity contrast with their nano-ambientes. In this paper, we show the photon scattering signal strength can be magnified by introducing a more abrupt change of refractivity at the virus particle using antibody-conjugated gold nanoparticles (AuNPs), allowing the presence of such viruses to be detected. Using two different deep learning methods to minimize scattering noise, the photon states scattering signal of a AuNPs ligated virus is enhanced significantly compared to that of a bare virus particle. This is confirmed by Finite Difference Time Domain (FDTD) numerical simulations. The sensitivity of the polarization state scattering spectra from a virus-gold particle doublet is 5.4 times higher than that of a conventional microscope image.

Target detection and classification in the shadow area of hyperspectral images (HSIs) has always been an important problem in the field of HSI data processing. However, there are few methods to detect or classify targets in the shadow area effectively because of occlusion of objects or oblique solar radiation. Dynamic stochastic resonance (DSR) theory shows that under the synergistic action of weak input signal, noise and non-linear system, the energy of noise can be transferred into signal partially. It breaks the idea that signal can be enhanced only by eliminating noise and has been proved to be effective in many fields. In this paper, DSR is introduced into the shadow area enhancement of HSI from both spatial and spectral dimensions. Then, the enhanced HSI data can be obtained by fusing the processed shadow with the original HSI data. Finally, 2D convolutional neural networks (2D-CNN) is used to classify the enhanced HSI. To evaluate the proposed method, a real-world HSI has been processed in the experiment and the results show that DSR can improve the contrast and the spectral intensity of HSI shadow area. Compared to other state-of-the-art methods, the classification accuracy is better at containing different targets with small samples, especially in the spectral dimension.

In this paper we demonstrate a novel methodology to electrically test and characterize resistive random-access memory (RRAM) single bit devices for deep learning application. We extract critical device performance metrics for validating and optimizing fabrication processes which feed into yield learning. We adopt the algorithm-based bias condition search methodology and extract forming and switching voltage parameters without overdriving the devices. This test methodology can be used for Technology Development Learning Cycle in a research and development environment.

The existing deep CNN models on single image super-resolution processing are computationally-intensive in terms of memory usage and training time. In resources-limited platforms, it is desirable to consider developing light-weight models for super-resolution tasks. This paper proposes a parallel-group convolution, which uses 25% computation of the standard convolutions. With parallel-group convolutions, we develop an efficient light-weight convolutional neural network named MobileSR for super-resolution. Experimental results show that our proposed method achieves appreciable improvements over the state-of-the-art models with approximately 75% size reduction. The source code is available at https://github.com/DestinyK/MobileSR.

Among existing device-free localization (DFL) methods, the methods based on pyroelectric infrared (PIR) sensor networks are much promising due to their advantages of low cost and privacy protection. Recently, we proposed a deep-learning-based method PIRNet which much decreases the deployment density of PIR-based DFL methods in multi-person scenarios. However, since PIRNet utilizes an end-to-end neural network that receives all the deployed PIR sensors' signals as input for localization, it has a defect of deployment-dependence: it assumes the PIR sensors' deployment in the testing environment is same to the training environment. Otherwise, it requires to be retrained. To address this problem, in this paper, we propose a deployment-independent method DeepPIRATes, which can be applied in environments of any deployments without retraining. DeepPIRATES has the character of deployment-independence because it divides the localization task into two steps and only utilizes deep learning in the first step. Especially, the first step aims at estimating the information about the persons' relative locations to a PIR sensor. Therefore, the utilized neural network only needs to receive a single PIR sensor's signal as input and is independent to the sensors' deployment. In the second step, DeepPIRATES further infers the persons' absolute locations by a particle filter which fuses the predicted information about the persons' relative locations to each sensor and does not require training data. Through DeepPIRATes, we achieve average localization errors of 0.55m, 0.73m, and 0.88m in scenarios of 1-person, 2-persons, and 3-persons with a deployment density of 0.08 sensors/m2.

Fisheye lens is widely equipped on remote operated vehicle (ROV) due to the wide field of view (FOV) and the calibration of fisheye image is one of the main topics. However, the existing calibration algorithms are not ideal in the edge of the fisheye image and are not suitable for the visual characteristics of human eyes, whose interest in areas include the horizontal angle is 20 degrees and the vertical angle is 15 degrees. Therefore, an improved method has been proposed in this paper, which can adapt to the perspective of the human eye by rotating the spherical model. Firstly, a two-dimensional plane can be mapped to a spherical coordinate system. Secondly, every area information in the distorted image include edge information can be got by rotating the spherical model. Finally, three-dimensional points that are rotated are remapped to the corrected image. The results have shown that distorted chessboard and underwater scene were tested using this proposed method in the experiment, the results meet the eye to the basic requirements.

A high-speed infrared imaging data acquisition and transmission system based on FPGA and USB3.0 technology is designed. FPGA is used to control the uncooled infrared sensor and image acquisition, and the imaging data is transmitted to PC in real-time through the USB3.0 interface. Cyclone IV series chips from Altera are used as the main controlling system. Cypress’s USB3.0 full protocol chip CYUSB3014 serves as the high-speed communication bridge between FPGA and PC.

In this study, we build an integrated microscope with multi-spectrum polarization imaging (MSPI) technology. Polarization imaging is achieved by rotating the polarizer and quarter wave plate, which can obtain parametric images of Stokes parameters, the polarization phase retardance, and the polarization ellipticity orientation angle. Hyperspectral imaging is achieved by using an electronically controlled liquid crystal wavelength filter in front of the broad-band visible light source. By calculating and integrating the intensity and parametric images under different illumination spectrum, the limitation of current polarization imaging due to the wavelength-dependent absorption of the sample was avoided. A significant improvement in resolving power and contrast was found in the imaging results of oocyte cells.

In optical imaging of biological tissues, surface reflection is a serious issue in many cases, while near-infrared (NIR) light can be used for deep imaging of biological tissues due to its strong tissue penetration ability and weaker light scattering in tissues. In this article, a method based on parametric imaging of the polarization status of reflected or transmitted light from biological tissues in the NIR spectrum was proposed. By modulating the polarization states of the NIR illumination source, recording the images of the samples, the far-field variation of the pixel intensity of recorded images was fitted to a theoretical model, and a series of polarization parametric images were obtained. These images describe the relation between the polarization status of the reflected or transmitted light and the properties of the samples. A tissue phantom with deep buried test target was imaged successfully. This method can provide a useful detection method for the deep imaging of biomedical tissues.

In laser additive manufacturing, the optical absorption characteristics of particles play a fundamental role in its application. As a non-destructive, high resolution, high contrast test method, photoacoustic imaging uses the endogenous optical absorption characteristics of the sample for imaging and obtains the structural and functional information of the sample. In this paper, photoacoustic imaging is utilized to conduct an experimental study on the absorption properties of the micro-sized copper balls. The experimental results show that this method can effectively map the absorption distribution characteristics of micro-particles, which is of great importance for quality control in laser additive manufacturing.

Nearest neighbor search, a classic way of identifying similar data, can be applied to various areas, including database, machine learning, natural language processing, software engineering, etc. Secure nearest neighbor search aims to find nearest neighbors to a given query point over encrypted data without accessing data in plaintext. It provides privacy protection to datasets when nearest neighbor queries need to be operated by an untrusted party (e.g., a public server). While different solutions have been proposed to support nearest neighbor queries on encrypted data, these existing solutions still encounter critical drawbacks either in efficiency or privacy. In light of the limitations in the current literature, we propose a novel approximate nearest neighbor search solution, referred to as Sparser, by leveraging a combination of space-filling curves, perturbation, and Order-Preserving Encryption. The advantages of Sparser are twofold, strengthening privacy and improving efficiency. Specifically, Sparser pre-processes plaintext data with space-filling curves and perturbation, such that data is sparse, which mitigates leakage abuse attacks and renders stronger privacy. In addition to privacy enhancement, Sparser can efficiently find approximate nearest neighbors over encrypted data with logarithmic time. Through extensive experiments over real-world datasets, we demonstrate that Sparser can achieve strong privacy protection under leakage abuse attacks and minimize search time.

Online bank system has been widely deployed to provide financial service worldwide. Big data mining brings serious privacy issues to consumers because consumption records may disclose their activities. Existing authentication and encryption algorithms in privacy-preserving deposit transaction schemes are mostly limited to the case that each consumption has to be reported to online banks directly and an insider adversary can monitor a consumer's bank account during the financial transaction period. Motivated by the above problems, we propose a deposit transaction scheme based on the differential privacy. Considering a mobile payment system as a noise generator, this scheme aims to reduce the relevancy between real consumption and deposit transaction. In addition, we provide in-depth theoretical analysis that our scheme can satisfy the definition of the differential privacy. Experimental results illustrate the consumption pattern can be protected with the differential privacy.

Screen lock is a critical security feature for smart-phones to prevent unauthorized access. Although various screen unlocking technologies including fingerprint and facial recognition have been widely adopted, they still have some limitations. For example, fingerprints can be stolen by special material stickers and facial recognition systems can be cheated by 3D-printed head models. In this paper, we propose EmgAuth, a novel electromyography(EMG)-based smartphone unlocking system based on the Siamese network. EmgAuth leverages the Myo armband to collect the EMG data of smartphone users and enables users to unlock their smartphones when picking up and watching their smartphones. In particular, when training the Siamese network, we design a special data augmentation technique to make the system resilient to the rotation of the armband. We conduct experiments including 40 participants and the evaluation results show that EmgAuth can effectively authenticate users with an average true acceptance rate of 91.81% while keeping the average false acceptance rate of 7.43%. In addition, we also demonstrate that EmgAuth can work well for smartphones with different sizes and at different locations, and is applicable for users with different postures. EmgAuth bears great promise to serve as a good supplement for existing screen unlocking systems to improve the safety of smartphones.

Gesture recognition systems based on myoelectric signal have raised more and more attention from researchers. Traditional gesture recognition methods are susceptible to multiple types of noise and require a large number of features, which increase overhead and decrease recognition efficiency. Fully utilizing the characteristics of the signal to recognize gestures is a big challenge. This paper proposes an improved empirical mode decomposition method (XB-EMD) based on autocorrelation function to denoise myoelectric signal. In addition, a novel deep neural network (CRNet) which combines the CNN and RNN layers together is trained for classifying the gestures based on denoised myoelectric signal. Experimental results show that the proposed gesture recognition system can improve recognition effectiveness at 97.4\% with 10 typical gestures, and 72.55\% with 16 complicated gestures.

Text clustering has been widely used in many Natural Language Processing (NLP) applications such as text summarization and news recommendation. However, most of the current algorithms need to predefine a clustering number, which is difficult to obtain. Moreover, the mutli-label clustering is useful in multiple clustering tasks in many applications, but related works are rarely available. Although several studies have attempted to solve above two problems, there is a need for methods that can solve the two issues simultaneously. Therefore, we propose a new text clustering algorithm called Word2Cluster. Word2Cluster can automatically generate an adaptive number of clusters and support multi-label clustering. To test the performance of Wrod2Cluster, we build a Chinese text dataset, Hotline, according to real world applications. To evaluate the clustering results better, we propose an improved evaluation method based on basic accuracy, precision and recall for multi-label text clustering. Experimental results on a Chinese text dataset (Hotline) and a public English text dataset (Reuters) demonstrate that our algorithm can achieve better F1-measure and runs faster than the state-of- the-art baselines.

Split air-conditioner is an important part in smart building, and its operation energy efficiency analysis plays a critical role in energy management and real-time operation of a smart building. In recent years, the energy efficiency of split air conditioners has usually been evaluated using a linear regression method. However, linear regression only studies the relationship between the independent variable and the conditional expectation of the dependent variable, it is impossible to comprehensively analyze the energy efficiency of real split air-conditioner. In this paper, a new method based on robust quantile regression is proposed to evaluate the energy efficiency of split air-conditioner. This method analyzes how the independent variables affect the median and multiple quantiles of the dependent variable. In addition, the proposed method does not require that the resultant statistical errors follow any Gaussian distribution. The simulation results show that the proposed method has stability and robustness in the energy efficiency assessment of split air-conditioners, so it has high application potential in smart buildings.

The widespread recognition of the blockchain technology that decoupled from the cryptocurrency has raised the interest of the industry. As the apparent drawbacks of the Proof of Work (PoW) protocol hinder the large deployment of PoW protocol for many distributed applications, the consensus mechanism evolved quickly to satisfy the business requirements. Since the consensus mechanism is the core of the blockchain technology, it is imperative to ensure that it performs correctly. Proof of Authority, a novel Byzantine fault-tolerant algorithm, is an appealing alternative to PoW algorithm for its high throughput rate and the security. However, the PoA protocol seems to do not convince the public about its security as it lacks adequate analysis. In this paper, we firstly explored the balance between security and scalability of PoA protocol, and subsequently, we proposed a flexible quantitative analysis framework based on Markov Decision Process. The framework can be used to indicate the security from the perspective of the adversarial optimal strategies. To capture the action strategies of an adversary, we also presented a formal model of PoA and its attack model. Since the modeling process for a blockchain protocol is not a trivial task, we implemented a case study to demonstrate the modeling process and discussed the security provisions of a specific PoA protocol.

In recent years, mobile social networks, which combine mobile communication and Internet technology, have developed rapidly and gradually become an indispensable part of people's daily life. In addition to sharing their own dynamics, users in mobile social networks also hope to find some new friends with the same interests through some attribute matching strategies. In view of this requirement and the implicit problem of protecting user privacy, this paper combines the idea of Matchmaking Encryption which is a new cryptographic primitive and of attribute-base encryption with outsourcing, and proposes an efficient and secure bidirectional friends matching scheme. Under the condition of protecting the privacy of users' secret attributes and matching strategies, the scheme enables both parties to make their own matching policies, thus achieving the goal that both parties can match to their expected new friends. In addition, the security and efficiency of the scheme are analyzed, and its feasibility is proved.

This paper provides a comprehensive review of the development of Collaborative Mobile Industrial Manipulator (CMIM), which is currently in high demand. Such a review is necessary to have an overall understanding about CMIM advanced technology. This is the first review to combine the system architecture and application which is necessary in order to gain a full understanding of the system. The classical framework of CMIM is firstly discussed, including hardware and software. Subsystems that are typically involved in hardware such as mobile platform, manipulator, end-effector and sensors are presented. With regards to software, planner, controller, perception, interaction and so on are also descripted. Following this, the common applications (logistics, manufacturing and assembly) in industry are surveyed. Finally, the trends are predicted and issues are indicated as references for CMIM researchers. Specifically, more research is needed in the areas of interaction, fully autonomous control, coordination and standards. Besides, experiments in real environment would be performed more and novel collaborative robotic systems would be proposed in future. Additionally, some advanced technology in other areas would also be applied into the system. In all, the system would become more intelligent, collaborative and autonomous.