Cyber-attacks and network intrusion have surfaced as major concerns for modern days applications of the Internet of Things (IoT). The existing intrusion detection and prevention techniques have a wide range of limitations and thus are unable to precisely detect any type of attack or anomaly within the network traffic. Many machine learning-based algorithms have also been presented by the researchers, which lack performance in terms of classification accuracy, or in terms of multi-class classification. This research presents a data-driven approach for intrusion and anomaly detection, where the data is processed and filtered using different algorithms. The quality of the training dataset is improved by using Synthetic Minority Oversampling Technique (SMOTE) algorithm and mutual information. Automated machine learning is also used to detect the algorithm with auto-tuned hyper-parameters that best suit to classify the data. This technique not only saves the computational cost to test the data at run-time but also provides an optimal algorithm without the need to run calculations to tune hyper-parameters, manually. The resultant algorithm solves a multi-class classification problem with an accuracy of 99.7%, outperforming the existing algorithms by a decent margin.

Deep learning-based algorithms are considered an efficient solution to carry out the insulator defect diagnosis task based on the aerial images captured by Unmanned Aerial Vehicles (UAVs) for electric power systems. However, the sufficient and accurate annotations of image samples required by deep learning-based models can be costly or not feasible in practice. This paper proposed an active learning-based solution for insulator defect diagnosis of electric transmission networks. The proposed solution aims to identify the most valuable samples and assign ground-true labels to significantly reduce the labeling effort. Specifically, Batch Active learning by Diverse Gradient Embedding (BADGE) strategy is adopted for sampling and GradCAM++ is used to extract the key regions of image samples iteratively. Then, a region-sample pair construction method is proposed during the labeling stage enabling the model to focus on the most discriminative regions based on a well-designed loss function. The proposed solution is extensively assessed through experiments and the results demonstrate that F1 scores of four popular CNN models trained with 1/3 of the total samples can be increased by up to 2.0% compared to the fully-labeled baseline solutions.

Face clustering can provide pseudo-labels to the massive unlabeled face data and improve the performance of different face recognition models. The existing clustering methods generally aggregate the features within subgraphs that are often implemented based on a uniform threshold or a learned cutoff position. This may reduce the recall of subgraphs and hence degrade the clustering performance. This work proposed an efficient neighborhood-aware subgraph adjustment method that can significantly reduce the noise and improve the recall of the subgraphs, and hence can drive the distant nodes to converge towards the same centers. More specifically, the proposed method consists of two components, i.e. face embeddings enhancement using the embeddings from neighbors, and enclosed subgraph construction of node pairs for structural information extraction. The embeddings are combined to predict the linkage probabilities for all node pairs to replace the cosine similarities to produce new subgraphs that can be further used for aggregation of GCNs or other clustering methods. The proposed method is validated through extensive experiments against a range of clustering solutions using three benchmark datasets and numerical results confirm that it outperforms the SOTA solutions in terms of generalization capability.

This article has been retracted. A retraction notice can be found at https://doi.org/10.3233/JIFS-219433.

Neural network detectors have been widely used to detect the insulators' self-blast defect from aerial images. However, detector performance can be significantly limited by insufficient defect samples. To this end, this article proposes a novel detection method that can promote defect feature learning with normal samples and their bounding box annotation, so that detectors can be improved without any additional manual burden. First, an efficient contour-based segmentation network named CapNet is proposed to extract insulator caps from box annotation. Then, the memory mechanism and polar alignment loss (PAL) are designed to promote training convergence and segmentation quality, respectively. Finally, a lightweight algorithm named MirrorFill is proposed to erase randomly selected caps, thus multiple defect samples can be generated from a single normal insulator. To verify the effectiveness of the proposed method, 21 492 images with 46 387 normal and 745 self-blast insulators are collected for dataset establishment. The experimental results show that MirrorFill can effectively improve the detector performance of multiple architectures. Also, the proposed method is capable of unsupervised (Uns.) anomaly detection (AD), which can achieve 37% defect sample F1-score even training with defect-agnostic annotation.

The automatic detection of insulator defects with UAV and CNN-based detectors has become a popular paradigm in recent years. However, insufficient insulator data has always been the bottleneck of detector performance. The existing augmentation method either performs a whole image transformation that lacks new semantic features or generates samples with massive manual annotation. Therefore, this paper proposes an automatic augmentation method called Weakly-Supervised Segmentation Mix (WSSM), where a Foreground Segmentation Network (FSN) is trained under the supervision of the bounding box label to extract the insulators for new sample synthesis. In the FSN training process, UnionMix is designed to generate hard samples based on the pseudo-label, thus facilitating the FSN segmentation ability for insignificant insulator boundaries. Oriented Muti-Instance Loss (OMIL) is proposed to extract supervision from the oriented bounding box so that FSN can be fully trained to handle the diverse angle distribution of insulators. The experiments conducted on the Aerial Insulator Dataset (AID) indicate that the synthesized images of WSSM can achieve stable improvement on multiple mainstream detectors. Both in-domain and cross-domain backgrounds can be used in WSSM to promote the network. For example, the AP of YOLOv5-m can be improved from 68.14 to 69.73 with 2600 COCO images, which exceeds the AP of bassline YOLOv5-l (69.69). To further verify the foreground extraction capability, this paper takes the FSN result as the pseudo-label and trains the instance segmentation network on iSAID. The comparison with existing SOTA methods proves the superior quality of the proposed method.

Background Migrant and left-behind families are vulnerable in health services utilization, but little is known about their disparities in immunization of non-National Immunization Program (NIP) vaccines. This study aims to evaluate the immunization coverage, knowledge, satisfaction, and associated factors of non-NIP vaccines among local and migrant families in the urban areas and non-left-behind and left-behind families in the rural areas of China. Methods A cross-sectional survey was conducted in urban areas of Zhejiang and rural areas of Henan in China. A total of 1648 caregivers of children aged 1–6 years were interviewed face-to-face by a pre-designed online questionnaire, and their families were grouped into four types: local urban, migrant, non-left-behind, and left-behind. Non-NIP vaccines included Hemophilus influenza b (Hib) vaccine, varicella vaccine, rotavirus vaccine, enterovirus 71 vaccine (EV71) and 13-valent pneumonia vaccine (PCV13). Log-binomial regression models were used to calculate prevalence ratios ( PR s) and 95% confidence intervals ( CI s) for the difference on immunization coverage of children, and knowledge and satisfaction of caregivers among families. The network models were conducted to explore the interplay of immunization coverage, knowledge, and satisfaction. Logistic regression models with odds ratios ( OR s) and 95% CI s were used to estimate the associated factors of non-NIP vaccination. Results The immunization coverage of all non-NIP vaccines and knowledge of all items of local urban families was the highest, followed by migrant, non-left-behind and left-behind families. Compared with local urban children, the PR s (95% CI s) for getting all vaccinated were 0.65 (0.52–0.81), 0.29 (0.22–0.37) and 0.14 (0.09–0.21) among migrant children, non-left-behind children and left-behind children, respectively. The coverage-knowledge-satisfaction network model showed the core node was the satisfaction of vaccination schedule. Non-NIP vaccination was associated with characteristics of both children and caregivers, including age of children (> 2 years- OR : 1.69, 95% CI : 1.07–2.68 for local urban children; 2.67, 1.39–5.13 for migrant children; 3.09, 1.23–7.76 for non-left-behind children); and below caregivers’ characteristics: family role (parents: 0.37, 0.14–0.99 for non-left-behind children), age (≤ 35 years: 7.27, 1.39–37.94 for non-left-behind children), sex (female: 0.49, 0.30–0.81 for local urban children; 0.31, 0.15–0.62 for non-left-behind children), physical health (more than average: 1.58, 1.07–2.35 for local urban children) and non-NIP vaccines knowledge (good: 0.45, 0.30–0.68 for local urban children; 7.54, 2.64–21.50 for left-behind children). Conclusions There were immunization disparities in non-NIP vaccines among migrant and left-behind families compared with their local counterparts. Non-NIP vaccination promotion strategies, including education on caregivers, and optimization of the immunization information system, should be delivered particularly among left-behind and migrant families. Graphical Abstract

Insulator is one of the most critical components of power transmission lines and its timely and accurate defect detection is considered important to ensure reliable and safe operation of transmission grids. This paper proposes an efficient augmentation method of aerial images captured using the unmanned aerial vehicle for the accurate detection of insulators with self-detonation defects. Through the adoption of the improved Resnet-18 model with the insulator edge features and the Grad-CAM based saliency map generation, the proposed solution can well maintain the vital regions for fine-grained classification in the augmentation process. The proposed solution is extensively assessed in comparison with the CNN-based benchmark methods through experiments. The numerical results indicate that the improved ResNet-18 model with the augmented images outperform the existing solutions and can identify the self-detonation defects with an accuracy of 95.1%.

This article proposes a lightweight YOLO-ACG detection algorithm that balances accuracy and speed, which improves on the classification errors and missed detections present in existing steel plate defect detection algorithms. To highlight the key elements of the desired area of surface flaws in steel plates, a void space convolutional pyramid pooling model is applied to the backbone network. This model improves the fusion of high- and low-level semantic information by designing feature pyramid networks with embedded spatial attention. According to the experimental findings, the suggested detection algorithm enhances the mapped value by about 4% once compared to the YOLOv4-Ghost detection algorithm on the homemade data set. Additionally, the real-time detection speed reaches about 103FPS, which is about 7FPS faster than the YOLOv4-Ghost detection algorithm, and the detection capability of steel surface defects is significantly enhanced to meet the needs of real-time detection of realistic scenes in the mobile terminal.

Dynamic voltage and frequency scaling (DVFS) is an essential approach to optimize the performance and energy tradeoff. In this article, learning models predict the workload and then estimate the corresponding power and thermal dissipation. The proposed framework utilizes a deep reinforcement learning (DRL)-based controller. —Ulf Schlichtmann, Technical University of Munich

For the problems of inaccurate recognition and the high missed detection rate of existing mask detection algorithms in actual scenes, a novel mask detection algorithm based on the YOLO-GBC network is proposed. Specifically, in the backbone network part, the global attention mechanism (GAM) is integrated to improve the ability to extract key information through cross-latitude information interaction. The cross-layer cascade method is adopted to improve the feature pyramid structure to achieve effective bidirectional cross-scale connection and weighted feature fusion. The sampling method of content-aware reassembly of features (CARAFE) is integrated into the feature pyramid network to fully retain the semantic information and global features of the feature map. NMS is replaced with Soft-NMS to improve model prediction frame accuracy by confidence decay method. The experimental results show that the average accuracy (mAP) of the YOLO-GBC reached 91.2% in the mask detection data set, which is 2.3% higher than the baseline YOLOv5, and the detection speed reached 64FPS. The accuracy and recall have also been improved to varying degrees, increasing the detection task of correctly wearing masks.

With the development of the automatic inspection of unmanned aerial vehicles (UAVs), improving the detection accuracy of insulators will not only help further insulator state detection and fault diagnosis but also contribute to the early landing of the UAVs’ automatic inspection system. In this paper, we propose a data augmentation method based on the random crop to improve the detection accuracy of insulators. Firstly, it ensures the validity of the label by generating a patch that contains the centers of all ground truth boxes. Secondly, it achieves a balance between protecting the ground truth and random cropping by limiting the area ratio of each ground truth box before and after random cropping. We find that these two steps increase the attention of the model to the insulator. On the self-made insulator dataset, the solution achieves 91.2% and 89.3% mAP in YOLOv3 and RetinaNet respectively, which is 3% and 1.5% better than the random crop.

Insulators on the electric power transmission lines are prone to defects due to outdoor exposure that can degrade the system performance or even lead to power outage. The existing solutions of defect diagnosis based on vision-based supervised deep learning algorithms suffer from the lack of defect samples. This work develops an efficient data augmentation method for aerial insulator images to promote the detection model performance. The insulator and background are separated firstly and the perspective transformation or random flip is adopted for the foreground and random erasing is used for the background in the augmentation process, respectively. In addition, the noises and weather features are incorporated for the overall image to simulate different photography scenarios. The proposed data augmentation solution is assessed through experiments using the CNN-based training models. The numerical results confirm that the proposed solution enables performance improvements for a range of models in terms of accuracy, recall and F2 score. Besides, the Grad-CAM visualization method is also used and demonstrates the superiority of models trained based on the augmented dataset for insulator defect diagnosis tasks.

Network pruning is considered efficient for sparsification and acceleration of Convolutional Neural Network (CNN) based models that can be adopted in re-source-constrained environments. Inspired by two popular pruning criteria, i.e. magnitude and similarity, this paper proposes a novel structural pruning method based on Graph Convolution Network (GCN) to further promote compression performance. The channel features are firstly extracted by Global Average Pooling (GAP) from a batch of samples, and a graph model for each layer is generated based on the similarity of features. A set of agents for individual CNN layers are implemented by GCN and utilized to synthesize comprehensive channel information and determine the pruning scheme for the overall CNN model. The training process of each agent is carried out using Reinforcement Learning (RL) to ensure their convergence and adaptability to various network architectures. The proposed solution is assessed based on a range of image classification datasets i.e., CIFAR and Tiny-ImageNet. The numerical results indicate that the proposed pruning method outperforms the pure magnitude-based or similarity-based pruning solutions and other SOTA methods (e.g., HRank and SCP). For example, the proposed method can prune VGG16 by removing 93% of the model parameters without any accuracy reduction in the CIFAR10 dataset.

With the increasing demand for intelligent security in power plants, the rapid and accurate processing of massive surveillance video data is urgently needed. Researches on the detection and analysis of abnormal human behaviors in power plants still focus on traditional image processing technology, and most of them lack robustness. In this article, abnormal behavior detection and analysis system based on personnel information are proposed to solve the above problems. The proposed method using an improved Y oLov3 algorithm first detects persons and extracts abnormal behavior information on this basis. In the implementation, some training tricks are introduced to improve performance. Experimental results show that the system can effectively detect abnormal human behaviors, and the improved Y oLov3 algorithm can also effectively improve model performance. The proposed abnormal behavior detection and analysis system based on personnel information prove its effectiveness through experiments, which can efficiently perform in power plants with lower computing costs.

At the era of Artificial Intelligence and Internet of Things (AIoT), battery-powered mobile devices are required to perform more sophisticated tasks featured with fast varying workloads and constrained power supply, demanding more efficient run-time power management. In this paper, we propose a deep reinforcement learning framework for dynamic power and thermal co-management. We build several machine learning models that incorporate the physical details for an ARM Cortex-A72, with on average 3% and 1 % error for power and temperature predictions, respectively. We then build an efficient deep reinforcement learning control incorporating the machine learning models and facilitating the run-time dynamic voltage and frequency scaling (DVFS) strategy selection based on the predicted power, workloads and temperature. We evaluate our proposed framework, and compare the performance with existing management methods. The results suggest that our proposed framework can achieve 6.8% performance improvement compared with other alternatives.

Traffic controls in modern society are part of urban management. With the assistance of unmanned aerial vehicles (UAVs) equipped with mounted cameras, researchers could capture aerial (bird-view) images from appropriate altitude. The perspective in aerial images makes appearances of objects squat, although aerial images can supply more contextual information about the environment by a broader view angle, the object instances may be detected by mistake. This fact diminishes the aerial images that can be fed to a network with higher dimensions that increases the computational cost to prevent the diminishing of pixels belonging to small objects. To compare model performance on small objects with aerial images, this study trains and tests two object detectors, i.e. YOLOv4 and YOLOv3, on the AU-AIR dataset, and exploited the parameterization of YOLO based models for small object detection. Finally, the key numerical results and observations are presented.