Backbone Extraction Research Articles

Improved YOLOv5 network method for remote sensing image-based ground objects recognition

High-resolution remote sensing images have the characteristics of complex background environment, clustering of objects, etc., the complex background makes the remote sensing image contain a large number of irrelevant ground objects with a high similarity or overlap, which makes the edge and texture of the objects not clear enough, and this leads to low recognition accuracy of ground objects such as airports, dams, and golf field, although the size of this object is large. Based on this problem, this paper proposes a remote sensing image object detection method based on the YOLOv5 network. By improving the backbone extraction network, the network structure can be deepened to get more information about large objects, and the detection effect can be improved by adding an attention mechanism and adding an output layer to enhance feature extraction and feature fusion. The pre-training weight is obtained by transfer learning and used as the training weight of the improved YOLOv5 to speed up the network convergence. The experiment is carried out on the DIOR dataset, the results show that the improved YOLOv5 network can significantly improve the accuracy of large object recognition compared with the YOLO series network and the EfficientDet model on DIOR dataset, and the mAP of the improved YOLOv5 network is 80.5%, which is 2% higher than the original YOLOv5 network.

A Complex Network Clustering and Phase Transition Models for Stock Price Dynamics before Crashes

Researchers from multiple disciplines have tried to understand the mechanism of stock market crashes. Precursory patterns before crashes agree with various empirical studies published by econophysicists, namely the prolific work of Didier Sornette. We intend to add more empirical evidence of synchronization of trading and demonstrate the prospect of predicting stock market crashes by analyzing clusters’ dynamics in the period of bubble build-up leading to a crash. We apply the Potential-based Hierarchical Agglomerative (PHA) Method, the Backbone Extraction Method, and the Dot Matrix Plot on S&P500 companies daily returns. Our innovative approach is proposed in this paper, empirical results and discussion presented in another publication.

Flavour network-based analysis of food pairing: Application to the recipes of the sub-cuisines from Northeast India

The flavour network-based analysis of food pairing was applied to the sub-cuisines from Northeast India to examine the food pairing behaviour in terms of the co-occurrence of ingredients with the shared flavouring compounds in food recipes. The method applied was based on an existing procedure in computational gastronomy, wherein the preference for positive pairing is attributed to dairy-based ingredients and negative pairing behaviour is attributed primarily to spice based ingredients. Recipe data was subjected to backbone extraction, projection of the recipe-ingredient-compound tri-partite network, and analysis for prevalence and authenticity of ingredients. Further, the average flavour sharing index of the cuisine was determined with the help of the flavour profiles of the ingredients. The extent of deviation for the original cuisine in comparison to a random cuisine was used to determine the degree of bias in the food pairing behaviour, with the sign as the indicator of the nature of pairing. The analysis identified the ingredients responsible to exhibit a positive or negative pairing pattern in the sub-cuisines. The ingredients from the spice category were the most prevalent and have resulted in the negative pairing behaviour in the cuisines. This role of spices in effecting a negative pairing behaviour is in line with the earlier reports for other Indian regional cuisines.

FRS-Net: An Efficient Ship Detection Network for Thin-Cloud and Fog-Covered High-Resolution Optical Satellite Imagery

According to statistics, more than 50% optical satellite images are covered by cloud or fog. Moreover, the cloud-cover rate is much higher in large bodies of water and nearby areas than inland areas due to the high amount of water evaporation and condensation. Therefore, ship detection from optical remote sensing images based on water surface analysis is more susceptible to cloud and fog interference, which affects the detection accuracy. In the time-sensitive application field of remote sensing images, due to the massive parameters in the large-scale network model, the detection speed is slow, and a lightweight detection model is commonly used. However, it is difficult for the lightweight detection model to achieve both high efficiency and accuracy for ship detection in a cloud-cover environment. To solve these problems, this manuscript proposes a lightweight algorithm called Fog Remote Sensing Ship Detection Network (FRS-Net) suitable for ship detection from remote sensing images in thin-cloud and fog covered environments. FRS-Net is developed based on the Deep Learning algorithm and can effectively improve ship detection accuracy under thin-cloud and fog cover. First, for the allocation strategy of anchor boxes, by using K-means clustering algorithm, FRS-Net simplifies the number of anchor boxes by utilizing the shape characteristics of the ship, which improves the time efficiency and the detection accuracy. Second, the FRS-Net network can meet the detection accuracy and has a fast inference speed. FRS-Net network is mainly composed of backbone extraction network, feature fusion network and prediction network. Experimental results on the Ship Detection in Optical Remote sensing images (SDIOR) dataset demonstrate the detection accuracy and computational efficiency of FRS-Net. The recognition mean Average Precision (mAP) achieved 43.20% for ship detection under thin-cloud and fog cover, with an efficiency of up to 424 FPS. FRS-Net has the potential to be applied in future scenarios such as embedded processing and on-board processing, where computing capabilities are strictly limited and the timeliness requirement is high.

A Complex Network Clustering and Phase Transition Models for Stock Price Dynamics before Crashes

Researchers from multiple disciplines have tried to understand the mechanism of stock market crashes. Precursory patterns before crashes agree with various empirical studies published by econophysicists, namely the prolific work of Didier Sornette. We intend to add more empirical evidence of synchronization of trading and demonstrate the prospect of predicting stock market crashes by analyzing clusters’ dynamics in the period of bubble build-up leading to a crash. We apply the Potential-based Hierarchical Agglomerative (PHA) Method, the Backbone Extraction Method, and the Dot Matrix Plot on S&P500 companies daily returns. Our innovative approach is proposed in this paper, empirical results and discussion presented in another publication.

Parameterless Pruning Algorithms for Similarity-Weight Network and Its Application in Extracting the Backbone of Global Value Chain

Abstract Purpose With the availability and utilization of Inter-Country Input-Output (ICIO) tables, it is possible to construct quantitative indices to assess its impact on the Global Value Chain (GVC). For the sake of visualization, ICIO networks with tremendous low- weight edges are too dense to show the substantial structure. These redundant edges, inevitably make the network data full of noise and eventually exert negative effects on Social Network Analysis (SNA). In this case, we need a method to filter such edges and obtain a sparser network with only the meaningful connections. Design/methodology/approach In this paper, we propose two parameterless pruning algorithms from the global and local perspectives respectively, then the performance of them is examined using the ICIO table from different databases. Findings The Searching Paths (SP) method extracts the strongest association paths from the global perspective, while Filtering Edges (FE) method captures the key links according to the local weight ratio. The results show that the FE method can basically include the SP method and become the best solution for the ICIO networks. Research limitations There are still two limitations in this research. One is that the computational complexity may increase rapidly while processing the large-scale networks, so the proposed method should be further improved. The other is that much more empirical networks should be introduced to testify the scientificity and practicability of our methodology. Practical implications The network pruning methods we proposed will promote the analysis of the ICIO network, in terms of community detection, link prediction, and spatial econometrics, etc. Also, they can be applied to many other complex networks with similar characteristics. Originality/value This paper improves the existing research from two aspects, namely, considering the heterogeneity of weights and avoiding the interference of parameters. Therefore, it provides a new idea for the research of network backbone extraction.

Motif-h: a novel functional backbone extraction for directed networks

Dense networks are very pervasive in social analytics, biometrics, communication, architecture, etc. Analyzing and visualizing such large-scale networks are significant challenges, which are generally met by reducing the redundancy on the level of nodes or edges. Motifs, patterns of the higher order organization compared with nodes and edges, are recently found to be the novel fundamental unit structures of complex networks. In this work, we proposed a novel motif h-backbone (Motif-h) method to extract functional cores of directed networks based on both motif strength and h-bridge. Compared with the state-of-the-art method Motif-DF and Entropy, our method solves two main issues which are often found in existing methods: the Motif-h reconsiders weak ties into our candidate set, and those weak ties often have critical functions of bridges in networks; moreover, our method provides a trade-off between the motif size and the edge strength, which quantifies the core edges accordingly. In the simulations, we compare our method with Motif-DF in four real-world networks and found that Motif-h can streamline the extraction of crucial structures compared with the others with limited edges.

Extracting the signed backbone of intrinsically dense weighted networks

AbstractNetworks provide useful tools for analysing diverse complex systems from natural, social and technological domains. Growing size and variety of data such as more nodes and links and associated weights, directions and signs can provide accessory information. Link and weight abundance, on the other hand, results in denser networks with noisy, insignificant or otherwise redundant data. Moreover, typical network analysis and visualization techniques presuppose sparsity and are not appropriate or scalable for dense and weighted networks. As a remedy, network backbone extraction methods aim to retain only the important links while preserving the useful and elucidative structure of the original networks for further analyses. Here, we provide the first methods for extracting signed network backbones from intrinsically dense unsigned unipartite weighted networks. Utilizing a null model based on statistical techniques, the proposed significance filter and vigor filter allow inferring edge signs. Empirical analysis on migration, voting, temporal interaction and species similarity networks reveals that the proposed filters extract meaningful and sparse signed backbones while preserving the multiscale nature of the network. The resulting backbones exhibit characteristics typically associated with signed networks such as reciprocity, structural balance and community structure. The developed tool is provided as a free, open-source software package.

Extracting modular-based backbones in weighted networks

Networks are an adequate representation for modeling and analyzing a great variety of complex systems. However, understanding networks with millions of nodes and billions of connections can be pretty challenging due to memory and time constraints. Therefore, selecting the relevant nodes and edges of these large-scale networks while preserving their core information is a major issue. In most cases, the so-called backbone extraction methods are based either on coarse-graining or filtering approaches. Coarse-graining techniques reduce the network size by gathering similar nodes into super-nodes, while filter-based methods eliminate nodes or edges according to a statistical property.In this work, a filter-based method is proposed and investigated. It uses the overlapping community structure to build the backbone in weighted networks. While most filtering techniques rely on link features to extract the backbone, the proposed method exploits both nodes and links. It takes advantage of the network communities through their main features (overlapping nodes, hubs, and bridging connections) to select influential edges and nodes while preserving the ability of the information dissemination of the original network. The so-called “Modular filtering backbone” combines two components. The first one is the network connecting the overlapping nodes and the top connected nodes (also called the hubs).One discards the edges with the lowest weights as long as connected components are maintained. The second component uses the network of the inter-community links with the nodes at their extremities. The disparity filter algorithm allows preserving only its most crucial connections. An extensive investigation is performed on a set of real-world weighted networks of various sizes and a wide range of origin. Results show the advantage of the proposed method over alternative filtering-based methods used for comparative purposes. Furthermore, this sheds new light on the most relevant parts of empirical networks hidden by their complexity.

Method for detecting surface defects of runner blades of large hydraulic turbines based on improved real-time lightweight network

Cavitation erosion, wears and cracks caused by residual thermal stress occur during the operation of hydropower units. It is a key scientific problem that needs to be solved urgently to detect defects in the foundation pit of hydropower units. In this paper, a mobile robot is used to obtain the surface image of the over-current parts of hydropower units, and deep learning algorithm is used to detect defects. In order to detect blade surface defects more safely, reliably, efficiently and quickly, this paper proposes an improved real-time lightweight convolutional neural network MobileNetv3-YOLOv4-Lite, which is suitable for portable embedded devices. In this paper, MobileNetv3 is selected to replace CSPDarkNet53 as the new backbone extraction network, and all standard convolutions are replaced by depth separable convolutions. Compared with CSPDarkNet53, MobileNetv3 only needs 37.35MB in weight, which is 206.94MB lower. It shows that the network proposed in this paper has the advantage of low memory and can run on CPU. The defect detection accuracy of MobileNetv3-YOLOv4-Lite can reach 97.48%, and the GPU can process 3024x4023 images, which can process 44 images in one second. The standard convolution of the enhanced feature extraction layer network is replaced by the depth separable convolution, and the parameter quantity is reduced by 29.33%. Therefore, it can be considered that the real-time lightweight convolutional neural network MobileNetv3-YOLOv4-lite proposed in this paper can detect the surface defects of turbine blades well.

Characterizing dissimilarity of weighted networks

Measuring the dissimilarities between networks is a basic problem and wildly used in many fields. Based on method of the D-measure which is suggested for unweighted networks, we propose a quantitative dissimilarity metric of weighted network (WD-metric). Crucially, we construct a distance probability matrix of weighted network, which can capture the comprehensive information of weighted network. Moreover, we define the complementary graph and alpha centrality of weighted network. Correspondingly, several synthetic and real-world networks are used to verify the effectiveness of the WD-metric. Experimental results show that WD-metric can effectively capture the influence of weight on the network structure and quantitatively measure the dissimilarity of weighted networks. It can also be used as a criterion for backbone extraction algorithms of complex network.

False Positives Using Social Cognitive Mapping to Identify Children’s Peer Groups

Children and adolescents interact in peer groups, which are known to influence a range of psychological and behavioral outcomes. In developmental psychology and related disciplines, social cognitive mapping (SCM), as implemented with the SCM 4.0 software, is the most commonly used method for identifying peer groups from peer report data. However, in a series of four studies, we demonstrate that SCM has an unacceptably high risk of false positives. Specifically, we show that SCM will identify peer groups even when applied to random data. We introduce backbone extraction and community detection as one promising alternative to SCM, and offer several recommendations for researchers seeking to identify peer groups from peer report data.

Extracting backbones in weighted modular complex networks

Network science provides effective tools to model and analyze complex systems. However, the increasing size of real-world networks becomes a major hurdle in order to understand their structure and topological features. Therefore, mapping the original network into a smaller one while preserving its information is an important issue. Extracting the so-called backbone of a network is a very challenging problem that is generally handled either by coarse-graining or filter-based methods. Coarse-graining methods reduce the network size by grouping similar nodes, while filter-based methods prune the network by discarding nodes or edges based on a statistical property. In this paper, we propose and investigate two filter-based methods exploiting the overlapping community structure in order to extract the backbone in weighted networks. Indeed, highly connected nodes (hubs) and overlapping nodes are at the heart of the network. In the first method, called “overlapping nodes ego backbone”, the backbone is formed simply from the set of overlapping nodes and their neighbors. In the second method, called “overlapping nodes and hubs backbone”, the backbone is formed from the set of overlapping nodes and the hubs. For both methods, the links with the lowest weights are removed from the network as long as a backbone with a single connected component is preserved. Experiments have been performed on real-world weighted networks originating from various domains (social, co-appearance, collaboration, biological, and technological) and different sizes. Results show that both backbone extraction methods are quite similar. Furthermore, comparison with the most influential alternative filtering method demonstrates the greater ability of the proposed backbones extraction methods to uncover the most relevant parts of the network.

On computing the analytic-signal backbone of the unforced harmonic oscillator

Backbone curves are usually obtained by analyzing nonlinear oscillators (e.g., the Duffing equation) whereas the backbones of linear oscillators are incidental by-products of such studies. Nevertheless, we focus on the harmonic oscillator, a linear, homogeneous constant-coefficient second-order ordinary differential equation. The backbone of the sinusoidally forced harmonic oscillator is well known, but the backbone of the unforced harmonic oscillator seems to have received essentially no attention. Since the 1980s, various computerized methods of backbone extraction have been under development, one of which is based on analytic-signal theory, a subfield of digital signal processing that involves Hilbert transform computations. By focusing on linear-theory backbones, a prerequisite to nonlinear analysis, we aim to pinpoint some of the theoretical and computational issues that arise when dealing with analytic-signal backbones. This paper examines the Mirror Periodic Method (MPM) as a practical way to reduce end effects in Hilbert transform computations and approximate analytic-signal backbones of unforced harmonic oscillators with light damping (less than 10 percent of critical). Using numerical examples, we demonstrate that MPM works well with both displacement and acceleration data, even for short signals, meaning just one or two damped cycles of oscillation. We then apply MPM techniques to accelerometer data that were recorded as part of earthquake-engineering teaching demonstrations involving a small-scale laboratory model of a linear single-degree-of-freedom system.

How Do Filtering Choices Impact the Structures of Audience Networks? A Simulation Study Using Data from 26 Countries

ABSTRACT Although communication scholars have increasingly studied audience fragmentation by analyzing audience networks, recent cross-national research has spurred a methodological debate about how to construct these audience networks from media use data. One point of contention lies in the choice of filtering procedures that only keep meaningful ties, where the traditional deviation-from-random-duplication filter has been replaced by alternatives such as backbone extraction. Using survey data from the 2016 Reuters Institute Digital News Report on news usage in 26 countries, we examined how filtering choices impact the structures of audience networks through a resampling simulation experiment. The simulation revealed a strong divergence between the filtering approaches, which invited disparate interpretations regarding the fragmentation of modern news audiences. The backbone extraction approach produced the most distinct results, while dealing most effectively with randomness in the media use data. As the filtering approaches mapped the structural features of audiences to different values of the same network metrics, the findings imply that studies diverging in these analytical decisions do not produce compatible results.

YOLO-RD: A lightweight object detection network for range doppler radar images

Under the condition of limited memory and computing power of radar aircraft equipment, large-scale object detection network based deep learning can not be deployed. Based on the darknet framework, Our paper proposes a lightweight object detection network for range doppler(RD) radar images: YOLO-RD, and builds a lightweight RD dataset: mini-RD, for efficient network training. Firstly, YOLO-RD extracts features from the input image through a series of small convolutional. Secondly, the dense block connection module is used to design the backbone extraction network. Finally, the prediction layer is combined with multi-scale features for prediction. Experiments show that YOLO-RD has achieved good results on the mini-RD dataset with a smaller memory budget, with a detection accuracy of 97.54%.

Motif-based functional backbone extraction of complex networks

As a natural abstraction of a large number of real-world systems, the structure and function of complex networks have been attracting increasing attentions in recent years. Existing studies have highlighted the statistical heterogeneity of connection patterns in large-scale networks, where valuable information is usually overwhelmed by redundant intricacy. In this case, the extraction of truly relevant nodes/connections of a large-scale network, namely, network backbones, can help form reduced but meaningful representations of a large-scale complex network and understand its fundamental structure and function. However, so far as we know, most existing backbone extraction methods focus mainly on the extraction of structural backbones, such as centrality-based backbones. Few studies have studied the problem of how to extract the functional backbones of a network, which is relevant to certain functional properties of the network. Accordingly, in this paper, we present two motif-based extraction methods to extract functional backbones of complex networks based on higher-order organization of salient motifs. One is built upon the global threshold method, and the other is based on the disparity filter method. We implement our proposed methods on a set of real-world networks to evaluate the performance. The results show that our extraction methods are more effective than other existing methods in terms of extracting functional backbones of a network, measured by motif centrality, motif degree, and motif abundance.

Networks identify productive forum discussions

Discussion forums provide a channel for students to engage with peers and course material outside of class, accessible even to commuter and non-traditional populations. As such, forums can build classroom community as well as aid learning, but students do not always take up these tools. We use network analysis to compare three semesters of forum logs from an introductory calculus-based physics course. The networks show dense structures of collaboration that differ significantly between semesters, even though aggregate participation statistics remain steady. After characterizing network structure for each semester, we correlate centrality with final course grade. Finally, we use a backbone extraction procedure to clean up "noise" in the network and clarify centrality/grade correlations. We find that network centrality is positively linked with course success in the two semesters with denser forum networks, and is a more reliable indicator than non-network measures such as post count. Backbone extraction destroys these correlations, suggesting that the "noise" is in fact signal and further analysis of the discussion transcripts is required.

Transport network backbone extraction: A comparison of techniques

Network backbone extraction techniques reduce the size of networks while trying to preserve their key topological and spatial features. Various backbone extraction algorithms have been proposed in different scientific fields. Although of clear interest to transport geographers, backbone extraction techniques have been adopted unevenly and in an ad hoc fashion in transport geography research. In this paper we therefore present a conceptual and experimental comparison of backbone extraction techniques in a transport-geographical context, and explore the new insights each technique can offer to enhance our understanding of the Southeast Asian intercity air transport network (SAAN). We review six frequently-used methods, i.e. global weight thresholding method (GWTM), k-core decomposition method (KCDM), minimum spanning tree method (MSTM), primary linkage analysis method (PLAM), multiple linkage analysis method (MLAM), and the disparity filter algorithm method (DFAM), and elaborate their analytical essence by applying them to extract the backbone of the SAAN. The abstracted networks are compared in terms of their geographical and topological structures using the initial network as a benchmark. This comparison is then used to point out the different techniques' potential in light of different transport geography research applications.

Uncovering the essential links in online commercial networks

Recommender systems are designed to effectively support individuals' decision-making process on various web sites. It can be naturally represented by a user-object bipartite network, where a link indicates that a user has collected an object. Recently, research on the information backbone has attracted researchers' interests, which is a sub-network with fewer nodes and links but carrying most of the relevant information. With the backbone, a system can generate satisfactory recommenda- tions while saving much computing resource. In this paper, we propose an enhanced topology-aware method to extract the information backbone in the bipartite network mainly based on the information of neighboring users and objects. Our backbone extraction method enables the recommender systems achieve more than 90% of the accuracy of the top-L recommendation, however, consuming only 20% links. The experimental results show that our method outperforms the alternative backbone extraction methods. Moreover, the structure of the information backbone is studied in detail. Finally, we highlight that the information backbone is one of the most important properties of the bipartite network, with which one can significantly improve the efficiency of the recommender system.