Search

The recent rapid development of deep learning has laid a milestone in industrial image anomaly detection (IAD). In this paper, we provide a comprehensive review of deep learning-based image anomaly detection techniques, from the perspectives of neural network architectures, levels of supervision, loss functions, metrics and datasets. In addition, we extract the promising setting from industrial manufacturing and review the current IAD approaches under our proposed setting. Moreover, we highlight several opening challenges for image anomaly detection. The merits and downsides of representative network architectures under varying supervision are discussed. Finally, we summarize the research findings and point out future research directions. More resources are available at https://github.com/M-3LAB/awesome-industrial-anomaly-detection.

Frequent itemset mining (FIM) and high-utility itemset mining (HUIM) are two important branches of itemset mining which is a key technology of knowledge discovery in many applications. Nowadays, there have been extensive algorithms on FIM and HUIM, but few studies consider frequency and utility together, so skyline frequent-utility itemset mining (SFUIM) is proposed to find useful itemsets with both frequency and utility measurements. Nevertheless, SFUIM is more challenging than FIM and HUIM since the search space is large and the calculation cost is expensive without any threshold, especially on large-scale databases. To address it, this paper proposes a novel prefix-based algorithm PSI* to mine skyline frequent-utility itemsets on massive data. PSI* divides the huge database by prefix-based partitioning, so that the calculation of itemsets with a specific prefix-item only involves a partition instead of the database. A multilevel-index based list is presented to compactly maintain the maximal utility under the frequency constraint, and a novel grid-based structure is devised to organize partitions or items by a designed order. Moreover, four efficient pruning strategies are proposed to prune itemsets as early as possible. Substantial experiments show that the PSI* algorithm has better performance than the state-of-the-art algorithms, obviously on large-scale databases.

Image anomaly detection (IAD) is an emerging and vital computer vision task in industrial manufacturing (IM). Recently, many advanced algorithms have been reported, but their performance deviates considerably with various IM settings. We realize that the lack of a uniform IM benchmark is hindering the development and usage of IAD methods in real-world applications. In addition, it is difficult for researchers to analyze IAD algorithms without a uniform benchmark. To solve this problem, we propose a uniform IM benchmark, for the first time, to assess how well these algorithms perform, which includes various levels of supervision (unsupervised versus fully supervised), learning paradigms (few-shot, continual and noisy label), and efficiency (memory usage and inference speed). Then, we construct a comprehensive IAD benchmark (IM-IAD), which includes 19 algorithms on seven major datasets with a uniform setting. Extensive experiments (17 017 total) on IM-IAD provide in-depth insights into IAD algorithm redesign or selection. Moreover, the proposed IM-IAD benchmark challenges existing algorithms and suggests future research directions. For reproducibility and accessibility, the source code is uploaded to the website: https://github.com/M-3LAB/open-iad

Functional dependencies are the most common constraints in the design theory for relational databases, which have very important practical applications in many areas. Different kinds of algorithms are proposed to discover the functional dependencies. However, it is found in this paper that the existing algorithms cannot deal well with massive data which cannot be held entirely in memory due to high memory consumption and high computation cost. In this paper, a novel algorithm FSC is presented to compute functional dependencies on massive data. The two-step execution of FSC relies on a pre-computed update-friendly assistant structure of comparable pairs which reflect the identifier pairs for tuples with at least one equal attribute. In step 1, FSC determines the violated functional dependencies and introduces the selective comparison by comparable pairs to reduce the required pairwise comparison significantly. The direct value-combination compression strategy is devised to process attributes of small cardinality. In step 2, FSC induces the required functional dependencies by the results in step 1. The extensive experimental results, conducted on synthetic and real-life data sets, show that FSC can discover functional dependencies on massive data efficiently.

The data collected in Internet of Thing (IoT) systems (IoT data) have stimulated dramatic extension to the boundary of commercialized data statistic analysis, owing to the pervasive availability of low-cost wireless network access and off-the-shelf mobile devices. In such cases, many data consumers post their queries for urban statistic analysis in the system, like the scales of traffics, and then data contributors in IoT networks upload their contents, which can be evaluated by data brokers and responded to data consumers. However, huge volumes of devices bring large scales of data, constituting heavy burdens for data exchange. Even worse, contents in IoT systems are also sensitive as they are usually linked to private physical status of data contributors. The previous studies for IoT data trading fail to provide comprehensive estimation and pricing towards these difficulties. Therefore, this paper proposes a novel framework for the range counting trading over IoT networks by jointly considering data utility, bandwidth consumption, and privacy preservation. The range counting accumulates the number of data items falling in a concerned range of value, providing important information on the underlying data distribution. This paper first proposes a novel sampling-based method with histogram sketching for range counting estimation. The estimator is proved to be unbiased and achieves advanced performance on variance. Then the framework adopts a perturbation mechanism that can further preserve the results under differential privacy. The theoretical analysis shows that the mechanism can guarantee the privacy preservation under a given size of samples and the accuracy requirement of results. Finally, two types of pricing strategies for range counting trading are introduced for different circumstances, providing holistic consideration on how the parameters given in the estimator should be used for data trading. The framework is evaluated by estimating the air pollution levels and the traffic levels with different ranges on the 2014 CityPulse Smart City datasets. The evaluation results demonstrate that our framework can provide more accurate and reliable statistical information, with reduced bandwidth consumption and strengthened privacy preservation.

Age of Information (AoI) has been proposed to measure the freshness of the sensory data for IoT applications. In Battery-free WSNs (BF-WSNs), The AoI minimization data collection problem has been extensively studied. Apart from data collection, target monitoring is also an important application for BF-WSNs. To capture the freshness of the sensory data, the AoI related to targets (The AoI of targets) needs to be considered. To guarantee the performance of the target monitoring system, the maximum AoI of all targets should be minimized. However, existing works mainly investigated the AoI related to nodes (The AoI of nodes), which is different from the AoI of targets. Also, existing energy models are not practical enough for battery-free nodes. To deal with those problems, in this paper, we first propose a more practical energy model. Then the problem of Maximum AoI minimization for Target monitoring in Battery-free WSNs (MTB) is formally defined based on the proposed energy model. A two-stage algorithm is proposed to solve MTB optimally, in which all nodes in the network are scheduled collaboratively to monitor all the targets in the monitoring field. Extensive simulations and real-world experiments verify the high performance of our algorithm and energy model.

In recent years, the problem of traffic congestion has become a hot topic. Accurate traffic flow prediction methods have received extensive attention from many researchers all over the world. Although many methods proposed at present have achieved good results in the field of traffic flow prediction, most of them only consider the static characteristic of traffic data, but do not consider the dynamic characteristic of traffic data. The factors that affect traffic flow prediction are changeable, and they will change over time. In response to this dynamic characteristic, the authors propose a model fusion mechanism based on transformer (TransFusion). The authors adopt two basic forecasting models (TCN and LSTM) as the underlying architectures. In view of the performance of different models on the traffic data at different times, the authors design a model fusion mechanism to assign dynamic weights to basic models at different times. Experiments on three datasets have proved that TransFusion has a significant improvement compared with basic models.

The widely employed tiny neural networks (TNNs) in mobile devices are vulnerable to adversarial attacks. However, more advanced research on the robustness of TNNs is highly in demand. This work focuses on improving the robustness of TNNs without sacrificing the model’s accuracy. To find the optimal trade-off networks in terms of the adversarial accuracy, clean accuracy, and model size, we present TAM-NAS, a tiny adversarial multi-objective one-shot network architecture search method. First, we build a novel search space comprised of new tiny blocks and channels to establish a balance between the model size and adversarial performance. Then, we demonstrate how the supernet facilitates the acquisition of the optimal subnet under white-box adversarial attacks, provided that the supernet significantly impacts the subnet’s performance. Concretely, we investigate a new adversarial training paradigm by evaluating the adversarial transferability, the width of the supernet, and the distinction between training subnets from scratch and fine-tuning. Finally, we undertake statistical analysis for the layer-wise combination of specific blocks and channels on the first non-dominated front, which can be utilized as a design guideline for the design of TNNs.

Although metallic chalcogenides are deemed as attractive sodium anode materials recently, the electrochemical performance is severely confined by the liability of structural collapse and sluggish ion diffusion kinetics. Herein a composite of carbon-encapsulated bimetallic selenides MoSe2–Sb2Se3was prepared by a hydrothermal method on the basis of abundant reaction sites, high activity, an extra built-in electric field generated from heterointerfaces, and synergistic effects between the different components. Equally important, the carbon coating is effective to support the structural stability by restraining the vast volumetric variation to achieve the purpose of improving the cycling performance. The density functional theory calculation results indicate that the band gap is narrowed and that the work function is decreased on the interface of the MoSe2–Sb2Se3heterojunction, leading to an additional driving force stemming from the introduction of the built-in electric field and the formation of the Sb–Se (Se from MoSe2) bond. Therefore, the resultant composite presents increased reaction kinetics and good electrochemical properties by acquiring a capacity of 376.0 mA h g–1over 580 cycles at 2.0 A g–1for the half-cell and 276 mA h g–1over 750 cycles at 2 A g–1for the full-cell. This work highlights bimetallic selenides with facilitated ion transferability with high performance.

Bismuth-based materials have attracted increasing attention in the research field of sodium/potassium-ion batteries owing to the high theoretical capacity. Unfortunately, the large volume variation and poor electrical conductivity limit their electrochemical performance and applications. Herein, we report a composite of heterostructured Bi2S3/MoS2encapsulated in nitrogen-doped carbon shell (BMS@NC) obtained by a solvothermal reaction as a novel anode material for sodium/potassium-ion batteries. The coating of the carbon layer could effectively relieve structural strains stemmed from the large volume change and improve electrical conductivity. More importantly, by skillfully constructing the heterostructure, an internal electric field formed on the heterointerface provides a rapid diffusion of ion and charge. As a consequence, the BMS@NC composite showed an excellent electrochemical performance for both sodium-ion batteries (a capacity of 381.5 mA h g–1achieved at a current density of 5.0 A g–1and 412 mA h g–1at 0.5 A g–1after 400 cycles) and potassium-ion batteries (a high specific capacity of 382.8 mA h g–1achieved after 100 cycles at 0.1 A g–1). The design of the Bi2S3/MoS2heterostructure provides an effective strategy to develop energy storage materials with good electrochemical properties.

Abstract

Indoor object detection is a very demanding and important task for robot applications. Object knowledge, such as two-dimensional (2D) shape and depth information, may be helpful for detection. In this article, we focus on region-based convolutional neural network (CNN) detector and propose a geometric property-based Faster R-CNN method (GP-Faster) for indoor object detection. GP-Faster incorporates geometric property in Faster R-CNN to improve the detection performance. In detail, we first use mesh grids that are the intersections of direct and inverse proportion functions to generate appropriate anchors for indoor objects. After the anchors are regressed to the regions of interest produced by a region proposal network (RPN-RoIs), we then use 2D geometric constraints to refine the RPN-RoIs, in which the 2D constraint of every classification is a convex hull region enclosing the width and height coordinates of the ground-truth boxes on the training set. Comparison experiments are implemented on two indoor datasets SUN2012 and NYUv2. Since the depth information is available in NYUv2, we involve depth constraints in GP-Faster and propose 3D geometric property-based Faster R-CNN (DGP-Faster) on NYUv2. The experimental results show that both GP-Faster and DGP-Faster increase the performance of the mean average precision.