Search

Images captured under poor lighting conditions often suffer from low brightness, low contrast, color distortion, and noise. The function of low-light image enhancement is to improve the visual effect of such images for subsequent processing. Recently, deep learning has been used more and more widely in image processing with the development of artificial intelligence technology, and we provide a comprehensive review of the field of low-light image enhancement in terms of network structure, training data, and evaluation metrics. In this paper, we systematically introduce low-light image enhancement based on deep learning in four aspects. First, we introduce the related methods of low-light image enhancement based on deep learning. We then describe the low-light image quality evaluation methods, organize the low-light image dataset, and finally compare and analyze the advantages and disadvantages of the related methods and give an outlook on the future development direction. [ABSTRACT FROM AUTHOR]

Gesture recognition plays an important role in smart homes, such as human–computer interaction, identity authentication, etc. Most of the existing WiFi signal-based approaches exploit a large number of channel state information (CSI) datasets to train a gestures classification model; however, these models require a large number of human participants to train, and are not robust to the recognition environment. To address this problem, we propose a WiFi signal-based gesture recognition system with matched averaging federated learning (WiMA). Since there are differences in the distribution of WiFi signal changes caused by the same gesture in different environments, the traditional federated parameter average algorithm seriously affects the recognition accuracy of the model. In WiMA, we exploit the neuron arrangement invariance of neural networks in parameter aggregation, which can improve the robustness of the gesture recognition model with heterogeneous CSI data of different training environments. We carried out experiments with seven participant users in a distributed gesture recognition environment. Experimental results show that the average accuracy of our proposed system is up to 90.4 % , which is very close to the accuracy of state-of-the-art approaches with centralized training models. [ABSTRACT FROM AUTHOR]

Visual tracking task is divided into classification and regression tasks, and manifold features are introduced to improve the performance of the tracker. Although the previous anchor-based tracker has achieved superior tracking performance, the anchor-based tracker not only needs to set parameters manually but also ignores the influence of the geometric characteristics of the object on the tracker performance. In this paper, we propose a novel Siamese network framework with ResNet50 as the backbone, which is an anchor-free tracker based on manifold features. The network design is simple and easy to understand, which not only considers the influence of geometric features on the target tracking performance but also reduces the calculation of parameters and improves the target tracking performance. In the experiment, we compared our tracker with the most advanced public benchmarks and obtained a state-of-the-art performance. [ABSTRACT FROM AUTHOR]

Road power generation technology is of significance for constructing smart roads. With a high electromechanical conversion rate and high bearing capacity, the stack piezoelectric transducer is one of the most used structures in road energy harvesting to convert mechanical energy into electrical energy. To further improve the energy generation efficiency of this type of piezoelectric energy harvester (PEH), this study theoretically and experimentally investigated the influences of connection mode, number of stack layers, ratio of height to cross-sectional area and number of units on the power generation performance. Two types of PEHs were designed and verified using a laboratory accelerated pavement testing system. The findings of this study can guide the structural optimization of PEHs to meet different purposes of sensing or energy harvesting. [ABSTRACT FROM AUTHOR]

This paper proposes a new damage identification approach for beam structures with stochastic parameters based on uncertain static measurement data. This new approach considers not only the static measurement errors, but also the modelling error of the initial beam structures as random quantities, and can also address static damage identification problems with relatively large uncertainties. First, the stochastic damage identification equations with respect to the damage indexes were established. On this basis, a new homotopy analysis algorithm was used to solve the stochastic damage identification equations. During the process of solution, a static condensation technique and a L1 regularization method were employed to address the limited measurement data and ill-posed problems, respectively. Furthermore, the definition of damage probability index is presented to evaluate the possibility of existing damages. The results of two numerical examples show that the accuracy and efficiency of the proposed damage identification approach are good. In comparison to the first-order perturbation method, the proposed method can ensure better accuracy in damage identification with relatively large measurement errors and modelling error. Finally, according to the static tests of a simply supported concrete beam, the proposed method successfully identified the damage of the beam. [ABSTRACT FROM AUTHOR]

Bionic electronic skin can accurately sense and locate surface pressure, which is widely demanded in many fields. Traditional electronic skin design usually relies on grid-architecture sensor arrays, requiring complex grid and interconnection arrangements as well as high cost. Grid-less planar sensors can solve the problem by using electrodes only at the edges, but they usually require the use of mapping software such as electrical impedance tomography to achieve high precision. In this work, a design method of high-precision grid-less planar pressure sensors based on the back-propagation (BP) neural network is proposed. The measurement precision of this method is demonstrated to be over two orders of magnitude higher than that of a grid-structure sensor array with the same electrode distribution density. Moreover, this method can be used for irregularly-shaped and non-uniform sensors, which further reduces the manufacturing difficulty and increases the application flexibility. [ABSTRACT FROM AUTHOR]

Detecting and classifying real-life small traffic signs from large input images is difficult due to their occupying fewer pixels relative to larger targets. To address this challenge, we proposed a deep-learning-based model (Dense-RefineDet) that applies a single-shot, object-detection framework (RefineDet) to maintain a suitable accuracy–speed trade-off. We constructed a dense connection-related transfer-connection block to combine high-level feature layers with low-level feature layers to optimize the use of the higher layers to obtain additional contextual information. Additionally, we presented an anchor-design method to provide suitable anchors for detecting small traffic signs. Experiments using the Tsinghua-Tencent 100K dataset demonstrated that Dense-RefineDet achieved competitive accuracy at high-speed detection (0.13 s/frame) of small-, medium-, and large-scale traffic signs (recall: 84.3%, 95.2%, and 92.6%; precision: 83.9%, 95.6%, and 94.0%). Moreover, experiments using the Caltech pedestrian dataset indicated that the miss rate of Dense-RefineDet was 54.03% (pedestrian height > 20 pixels), which outperformed other state-of-the-art methods. [ABSTRACT FROM AUTHOR]

Aiming to alleviate traffic congestion, many congestion avoidance and traffic optimization systems have been proposed recently. However, most of them suffer from three main problems. Firstly scalability: they rely on a centralized server, which has to perform intensive communication and computational tasks. Secondly unpredictability: they use smartphones and other sensors to detect the congested roads and warn upcoming vehicles accordingly. In other words, they are used to solve the problem rather than avoiding it. Lastly, infrastructure dependency: they assume the presence of pre-installed infrastructures such as roadside unit (RSU) or cellular 3G/4G networks. Motivated by the above-mentioned reasons, in this paper, we proposed a fully distributed and infrastructure-less congestion avoidance and traffic optimization system for VANET (Vehicular Ad-hoc Networks) in urban environments named DIFTOS (Distributed Infrastructure-Free Traffic Optimization System), in which the city map is divided into a hierarchy of servers. The vehicles that are located in the busy road intersections play the role of servers, thus DIFTOS does not rely on any centralized server and does not need internet connectivity or RSU or any kind of infrastructure. As far as we know, in the literature of congestion avoidance using VANET, DIFTOS is the first completely infrastructure-free congestion avoidance system. The effectiveness and scalability of DIFTOS have been proved by simulation under different traffic conditions. [ABSTRACT FROM AUTHOR]