State Key Laboratory of Subtropical Building Science

Unsupervised anomaly detection aims to identify data samples that have low probability density from a set of input samples, and only the normal samples are provided for model training. The inference of abnormal regions on the input image requires an understanding of the surrounding semantic context. This work presents a Semantic Context based Anomaly Detection Network, SCADN, for unsupervised anomaly detection by learning the semantic context from the normal samples. To achieve this, we first generate multi-scale striped masks to remove a part of regions from the normal samples, and then train a generative adversarial network to reconstruct the unseen regions. Note that the masks are designed in multiple scales and stripe directions, and various training examples are generated to obtain the rich semantic context . In testing, we obtain an error map by computing the difference between the reconstructed image and the input image for all samples, and infer the abnormal samples based on the error maps. Finally, we perform various experiments on three public benchmark datasets and a new dataset LaceAD collected by us, and show that our method clearly outperforms the current state-of-the-art methods.

Surrogate-assisted evolutionary algorithms (SAEAs) have become one popular method to solve complex and computationally expensive optimization problems. However, most existing SAEAs suffer from performance degradation with the dimensionality increasing. To solve this issue, this article proposes a classifier-assisted level-based learning swarm optimizer on the basis of the level-based learning swarm optimizer (LLSO) and the gradient boosting classifier (GBC) to improve the robustness and scalability of SAEAs. Particularly, the level-based learning strategy in LLSO has a tight correspondence with the classification characteristic by setting the number of levels in LLSO to be the same as the number of classes in GBC. Together, the classification results feedback the distribution of promising candidates to accelerate the evolution of the optimizer, while the evolved population helps to improve the accuracy of the classifier. To select informative and valuable candidates for real evaluations, we devise an <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">${L}1$ </tex-math></inline-formula> -exploitation strategy to extensively exploit promising areas. Then, the candidate selection is conducted between the predicted <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">${L}1$ </tex-math></inline-formula> offspring and the already real-evaluated <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">${L}1$ </tex-math></inline-formula> individuals based on their Euclidean distances. Extensive experiments on commonly used benchmark functions demonstrate that the proposed optimizer can achieve competitive or better performance with a very small training dataset compared with three state-of-the-art SAEAs.

This paper presents a deep multi-model fusion network to attentively integrate multiple models to separate layers and boost the performance in single-image dehazing. To do so, we first formulate the attentional feature integration module to maximize the integration of the convolutional neural network (CNN) features at different CNN layers and generate the attentional multi-level integrated features (AMLIF). Then, from the AMLIF, we further predict a haze-free result for an atmospheric scattering model, as well as for four haze-layer separation models, and then fuse the results together to produce the final haze-free image. To evaluate the effectiveness of our method, we compare our network with several state-of-the-art methods on two widely-used dehazing benchmark datasets, as well as on two sets of real-world hazy images. Experimental results demonstrate clear quantitative and qualitative improvements of our method over the state-of-the-arts.

Here, we reported a strategy for channel methylation to construct a robust ultramicroporous metal–organic framework (MOF) Ni(TMBDC)(DABCO)0.5 through hydrothermal synthesis method and investigated its adsorption performance for recovering ethane (C2) and propane (C3) from natural gas. The as-synthesized Ni(TMBDC)(DABCO)0.5 featured ultramicroporosity with a uniform pore size of 0.5 nm. The resulting sample showed a strong adsorption interaction with C3H8 and C2H6, and its C3H8 adsorption capacity at a low pressure of 1 kPa was up to 2.80 mmol/g and its C2H6 adsorption capacity at a low pressure of 10 kPa reached as high as 2.93 mmol/g, exhibiting strong binding affinity for ethane and propane. The enhanced adsorption can be attributed to the presence of the dense and accessible methyl and methylene groups in the channels of the sample. Grand Canonical Monte Carlo (GCMC) simulations also confirmed that the methylene groups from the DABCO pillar and the methyl groups from the TMBDC ligand play an important role in enhancing the adsorption of ethane and propane. Its ideal adsorbed solution theory (IAST)-predicted selectivity of C2H6/CH4 reached unprecedentedly 29, much higher than most of the reported data for MOFs. The stability test confirmed that the crystal structure of Ni(TMBDC)(DABCO)0.5 still remained intact after it was exposed to moist air with a relative humidity of 100% for days. The breakthrough experiment demonstrated that the CH4/C2H6/C3H8 ternary mixture was completely separated using a fixed bed of Ni(TMBDC)(DABCO)0.5 at ambient temperature, showing a great potential for recovering the low content of ethane and propane from natural gas.

Clarifying the ecosystem service (ES) trade-off/synergy is a prerequisite for scientific implementation and optimization of integrated ecological system management strategies, especially in highly urbanized areas with declining eco-environmental carrying capacity. This study took China's Yangtze River Delta urban agglomeration (YRDUA) as an example. First, we used a pixel-by-pixel analysis based on dynamic spatial correlation to quantify the trade-off/synergy among the 6 primary ESs' change amounts from 2005 to 2020 (habitat quality, soil retention, carbon sequestration, water yield, food production, leisure and recreation). Then, we integrated the multiple ES trade-off/synergy and explored their driving factors by spatial cluster analysis and geographically and temporally weighted regression (GTWR). The results showed synergy between supporting and regulating services predominated while regulating services were mostly trade-offs with regulating and cultural services. The trade-off/synergy of the 15 ES functional pairs within YRDUA can be downscaled into 4 ES trade-off and synergy bundles. Furthermore, the climate drivers (annual highest temperature (TH), annual total precipitation (PRE), annual total solar radiation (SRT)) significantly influenced the spatial heterogeneity of the 6 ESs in the YRDUA. The socio-economic drivers (night light index (NLI), gross domestic product (GDP)) and land use drivers (proportion of arable land (ALP), proportion of construction land (CLP), proportion of woodland (WLP)) had a more remarkable power of explanation for the ES temporal variation and ES trade-off/synergy. YRDUA and other urban agglomerations can use the study's findings to develop differentiated integrated ecological management strategies.

This paper presents a novel deep learning model to aggregate the attentional dilated features for salient object detection by exploring the complementary information between the global and local context in a convolutional neural network. There are two technical contributions to our network design. First, we develop an attentional dense atrous (dilated) spatial pyramid pooling (AD-ASPP) module to selectively use the local saliency cues captured by dilated convolutions with a small rate and the global saliency cues captured by dilated convolutions with a large rate. Second, taking the feature pyramid network as the backbone, we develop an aggregation network to integrate the refined features by formulating two consecutive chains of residual learning based modules: one chain from deep to shallow layers while another chain from shallow to deep layers. We evaluate our network on seven widely-used saliency detection benchmarks by comparing it against 21 state-of-the-art methods. Experimental results show that our network outperforms others on all the seven benchmark datasets.

In this paper, some simple stress-strain relationships of the concrete material recommended in relevant Codes are appropriately simplified, then the damage factors of the simplified plastic damage constitutive model is determined based on Sidiroff’s energy equivalence principle. Mechanical characteristics of the concrete material under the simple tension or compression are analyzed by Finite element method. Through the comparison of numerical analysis results and Code constitutive relations, the damage factors of the simplified plastic damage constitutive model is verified. The unreinforced concrete beam static tests by Petersson is simulation analyzed by the nonlinear finite element method and plastic damage constitutive model. The effect of the unit size and the different linear softening constitutive relation on the analysis results are considered. The results show that there is no obvious size effect on the plastic damage analysis results based on fracture cracking criterion, the results of the bilinear softening constitutive analysis have good accuracy, and the form of softening constitutive relation has a great influence on the result.

This work presents a gated non-local deep residual learning framework for image deraining. It can avoid the over-deraining or under-deraining caused by the global residual learning in existing deraining networks, since the learned soft gate in our method adaptively adjusts the amount of global residual to be passed for generating the final derained result. To generate feature maps for global residual prediction, we develop a non-local guided attention module (NLAM), which first obtains non-local features by exploiting spatial inter-dependencies among all the feature positions of local features produced by convolutional neural network (CNN), and then leverages the attention mechanism to merge the local and non-local features based on their complementary relation. Moreover, we develop a channel-wise gated prediction module to learn a soft gate on the global residual by explicitly modelling channel inter-dependencies of the feature maps obtained from NLAM. Experiments on four deraining benchmark datasets and real-world rainy images show that our network has a quantitative and qualitative improvement over state-of-the-arts.

Temporal repetition counting aims to estimate the number of cycles of a given repetitive action. Existing deep learning methods assume repetitive actions are performed in a fixed time-scale, which is invalid for the complex repetitive actions in real life. In this paper, we tailor a context-aware and scale-insensitive framework, to tackle the challenges in repetition counting caused by the unknown and diverse cycle-lengths. Our approach combines two key insights: (1) Cycle lengths from different actions are unpredictable that require large-scale searching, but, once a coarse cycle length is determined, the variety between repetitions can be overcome by regression. (2) Determining the cycle length cannot only rely on a short fragment of video but a contextual understanding. The first point is implemented by a coarse-to-fine cycle refinement method. It avoids the heavy computation of exhaustively searching all the cycle lengths in the video, and, instead, it propagates the coarse prediction for further refinement in a hierarchical manner. We secondly propose a bidirectional cycle length estimation method for a context-aware prediction. It is a regression network that takes two consecutive coarse cycles as input, and predicts the locations of the previous and next repetitive cycles. To benefit the training and evaluation of temporal repetition counting area, we construct a new and largest benchmark, which contains 526 videos with diverse repetitive actions. Extensive experiments show that the proposed network trained on a single dataset outperforms state-of-the-art methods on several benchmarks, indicating that the proposed framework is general enough to capture repetition patterns across domains. Code and data are available in https://github.com/Xiaodomgdomg/Deep-Temporal-Repetition-Counting.

With rapid urbanization, inundation-induced property losses have become more and more severe. Urban inundation modeling is an effective way to reduce these losses. This paper introduces a simplified urban stormwater inundation simulation model based on the United States Environmental Protection Agency Storm Water Management Model (SWMM) and a geographic information system (GIS)-based diffusive overland-flow model. SWMM is applied for computation of flows in storm sewer systems and flooding flows at junctions, while the GIS-based diffusive overland-flow model simulates surface runoff and inundation. One observed rainfall scenario on Haidian Island, Hainan Province, China was chosen to calibrate the model and the other two were used for validation. Comparisons of the model results with field-surveyed data and InfoWorks ICM (Integrated Catchment Modeling) modeled results indicated the inundation model in this paper can provide inundation extents and reasonable inundation depths even in a large study area.

Superhydrophobic surfaces have attracted much attention in environmental control because of their excellent water-repellent properties. A successful design of superhydrophobic surfaces requires a correct understanding of the influences of surface roughness on water-repellent behaviors. Here, a new approach, a mesoscale lattice Boltzmann simulation approach, is proposed and used to model the dynamic behavior of droplets impacting on surfaces with randomly distributed rough microstructures. The fast Fourier transformation method is used to generate non-Gaussian randomly distributed rough surfaces, with the skewness and kurtosis obtained from real surfaces. Then, droplets impacting on the rough surfaces are modeled. It is found that the shape of droplet spreading is obviously affected by the distributions of surface asperity. Decreasing the skewness and keeping the kurtosis around 3 is an effective method to enhance the ability of droplet rebound. The new approach gives more detailed insights into the design of superhydrophobic surfaces.

Fire hazard in public buildings may result in serious casualties due to the difficulty of evacuation caused by intricate interior space and unpredictable development of fire situations. It is essential to provide safe and reliable indoor navigation for people trapped in the fire. Distinguished from the global shortest rescue route planning, a framework focusing on the local safety performance is proposed for emergency evacuation navigation. Sufficiently utilizing the information from Building Information Modeling (BIM), this framework automatically constructs geometry network model (GNM) through Industry Foundation Classes (IFC) and integrates computer vision for indoor positioning. Considering the available local egress time (ALET), a back propagation (BP) neural network is applied for adjusting the rescue route according to the fire situation, improving the local safety performance of evacuation. A campus building is taken as an example for proving the feasibility of the framework proposed. The result indicates that the rescue route generated by proposed framework is secure and reasonable. The proposed framework provides an idea for using real-time images only to implement the automatic generation of rescue route when a fire hazard occurs, which is passive, cheap, and convenient.

Greenways are linear green spaces that are widely incorporated as policy instruments to address various urban issues. Heterogeneity is observed among the forms, functions, and activities of greenways. However, a number of studies have viewed urban greenways as homogeneous landscape features despite the increasing heterogeneity of urban greenways caused by transportation development. Taking the “three-legged stool” concept as a theoretical starting point, this article develops a conceptual framework for understanding the heterogeneous landscapes of urban greenways. The framework is then applied to empirical work in Shenzhen. This study shows that traffic impact, corridor width and land use are crucial factors in determining the heterogeneity of urban greenways and resolving the conflicts that result from the overemphasis on the transportation function of greenways. These factors also determine the primary benefits of greenways and differentiate various types of greenways. Based on field observations and empirical data, we identify four types of greenways in Shenzhen: transport greenways, forest greenways, park greenways and rural greenways. Greenways in Shenzhen have apparent heterogeneity in recreational attractiveness due to the surrounding landscape and external interference. Furthermore, the majority of Shenzhen greenways are nonmotorized transportation infrastructure with narrow corridors of street greenery. The composition and heterogeneity of greenways in Shenzhen are the result of the “one-size-fits-all” approach to greenway typologies and planning activities, which has become a challenge for multipurpose greenway planning in urban environments. Future efforts should place more emphasis on the heterogeneous landscapes of urban greenways in order to develop improvement strategies associated with specific policy goals.

Deep learning has been recently demonstrated as an effective tool for raster-based sketch simplification. Nevertheless, it remains challenging to simplify extremely rough sketches. We found that a simplification network trained with a simple loss, such as pixel loss or discriminator loss, may fail to retain the semantically meaningful details when simplifying a very sketchy and complicated drawing. In this paper, we show that, with a well-designed multi-layer perceptual loss, we are able to obtain aesthetic and neat simplification results preserving semantically important global structures as well as fine details without blurriness and excessive emphasis on local structures. To do so, we design a multi-layer discriminator by fusing all VGG feature layers to differentiate sketches and clean lines. The weights used in layer fusing are automatically learned via an intelligent adjustment mechanism. Furthermore, to evaluate our method, we compare our method to state-of-the-art methods through multiple experiments, including visual comparison and intensive user study.

The fluid flow and conjugate heat and mass transfer across a hollow fiber membrane tube bundle used for liquid desiccant air dehumidification are investigated. In this process, humid air flows across the fiber bank and salt solution flows inside the fibers packed in a shell. They exchange heat and moisture through the membranes. To overcome the difficulties in the direct modeling of the whole tube bundle, a representative cell, which comprises of a single fiber, a solution stream inside the fiber, and an air stream flowing across the fiber, is selected as the calculation domain. The liquid flow inside the fibers is assumed to be laminar due to the low Reynolds numbers, while the air flow across the bank is considered to be turbulent as a result from the disturbances from the numerous fibers. The governing equations for fluid flow and heat and mass transfer in the two flows and in the membrane are coupled together and solved numerically with a self-built code. Experimental work on hollow fiber membrane-based liquid desiccant air dehumidification is performed to validate the model. The fundamental data on friction factor, Nusselt and Sherwood numbers on both the shell and the tube sides are then obtained for Re = 300–600.

Given the greater risk of flooding in cities due to climate change, spatial planning systems are increasingly expected to contribute to flood resilience. However, incorporating expanded adaption measures in conventional planning practices remains a major challenge due to institutional barriers. Based on the theories of historical institutionalism in relation to path divergence, this paper aims to understand the factors which determine the fate of innovations and departures from established practice. Using Guangzhou as a case study, the paper traces the history of the city's struggle against flooding from the 1920s onwards, building on documentary analysis, mapping and interviews. The findings highlight a deeply rooted attachment to engineering-based solutions to tackle flood risk. It also indicates that departing from an established path to embed nature-based and non-structural solutions in the planning system is more likely to take place in response to changing socio-economic needs and strong institutional support for changes, rather than in response to major flooding events. These findings provide lessons for policymakers and urban planners seeking to enact new policies to enhance flood resilience in spatial planning.

Abstract Drought is one of the major natural hazards with a possibly devastating impact on the regional environment, agriculture, and water resources. Previous studies have assessed the historic changes in meteorological drought over various regional scales but have rarely considered hydrological drought due to limited hydrological observations. Here, we use long-term (1960–2012) hydro-meteorological data to analyze the meteorological and hydrological drought comparatively in the Pearl River basin (PRB) in southern China using the standardized precipitation index (SPI) and the standardized runoff index (SRI). The results indicate a strong positive correlation between the SPI and SRI, and the correlation tends to be stronger at the longer timescale. The SPI is reliable to substitute for the SRI to represent the hydrological drought at the long-term scale (e.g., 12 months or longer). Trend analysis reveals a noticeably wetting trend mainly in the eastern regions and a significant drying trend mainly in the western regions and the downstream area of the PRB. The drought frequency is spatially heterogeneous and varies slightly at the interannual scale. Overall, the drought is dominated by noticeable cycles of shorter periodicity (0.75–1.8 years), and periodic cycles in the meteorological drought are mainly responsible for those in the hydrological drought.

A simply supported reinforced concrete (RC) beam only experiences sagging moment under static loads while it might experience both sagging and hogging moments under impact loads due to the inertial effect. In order to investigate inertial effect on the impact behavior of RC beam, a numerical model is developed by using the finite element code LS-DYNA. The strain rate effect of the material is considered in the numerical model. The numerical model is calibrated with the testing results of drop weight impact on RC beams available in the literature. The numerical results show that the prediction is better than some other researchers’ predictions in terms of peak impact force and peak deformation. In addition, inertial effect is quantitatively evaluated by the peak impact force and the peak hogging moment. The relationship between the peak hogging moment and the peak impact force of the beam is investigated by conducting parametric studies with regard to various net spans, impact masses and impact velocities. The empirical formulae are then proposed to predict the peak impact force and the peak hogging moment. The predications by the proposed empirical formulae are compared with the testing results and the predicted results by other formulae available in the literatures.

The rapid development of information and communication technology has led to the Internet and social media becoming a vital platform for public participation in China. The present research sought to understand the complexity of participation in the network society by taking the cancellation of the number 55 bus route in Shanghai as a case study. Both qualitative and quantitative research methods were used to analyze data from a leading social networking site in China. An analysis of participation patterns led to an understanding of the main characteristics of public participation in the network society, and a statistical analysis of the network revealed the features of elite participants in the planning adjustment. A qualitative approach was also used to explore the communication process, which was influenced by Chinese social capital— guanxi. The case study revealed an uneven pattern of public participation in the network society, and suggestions are provided to enhance fairness and effectiveness in this process.

Existing GAN inversion methods are stuck in a paradox that the inverted codes can either achieve high-fidelity reconstruction, or retain the editing capability. Having only one of them clearly cannot realize real image editing. In this paper, we resolve this paradox by introducing consecutive images (e.g., video frames or the same person with different poses) into the inversion process. The rationale behind our solution is that the continuity of consecutive images leads to inherent editable directions. This inborn property is used for two unique purposes: 1) regularizing the joint inversion process, such that each of the inverted codes is semantically accessible from one of the other and fastened in an editable domain; 2) enforcing inter-image coherence, such that the fidelity of each inverted code can be maximized with the complement of other images. Extensive experiments demonstrate that our alternative significantly outperforms state-of-the-art methods in terms of reconstruction fidelity and editability on both the real image dataset and synthesis dataset. Furthermore, our method provides the first support of video-based GAN inversion and an interesting application of unsupervised semantic transfer from consecutive images. Source code can be found at: https://github.com/cnnlstm/InvertingGANs_with_ConsecutiveImgs.