Average Map Research Articles

A Novel Action Saliency and Context-Aware Network for Weakly-Supervised Temporal Action Localization

Temporal action localization is a challenging task in computer vision, and it tries to find the start time and the end time of the actions and predict their categories. However, compared to temporal action localization, weakly supervised temporal action localization (WTAL) is a more challenging task due to its poor annotations. With only video-level annotation, some background frames, similar to actions, would be classified as actions and produce inaccurate results. In addition, the two-stream fusion problem, ignored previously, also needs to be further considered. To resolve these issues, we propose a novel action saliency and context-aware network (ASCN) for weakly supervised temporal action localization tasks. Specifically, the temporal saliency and context module is designed to enhance the global saliency and context information of the RGB and the flow features to suppress the backgrounds and enhance the actions. In addition, a hybrid attention mechanism using frame differences and two-stream attention is designed to model the local action context information and further enlarge the scores of the potential action regions and suppress the background regions. Finally, to obtain two-stream consistency and solve the fusion problem, we use the similarity loss and a channel self-attention module to adaptively fuse the enhanced RGB and flow features. Extensive experiments demonstrate that ASCN can outperform all of the SOTA WTAL methods on the THUMOS14 dataset and the ActivityNet1.3 dataset with an average mAP that can reach 37.2% on the THUMOS14 dataset and attains an average mAP of 26.3% on the ActivityNet1.3 dataset. On the ActivityNet1.2 dataset, ASCN can also obtain comparable results. Compared with AdapNet (TNNLS20), MMSD (TIP22), and FTCL (CVPR22) on the THUMOS14 dataset, ASCN can outperform them by 13.5%, 2.9%, and 2.8%, respectively.

HE‐DETR‐DC5 method to rapidly detect objects on passive THz images in human safety inspection

As worldwide counterterrorism operations grow, screening people for concealed objects in public places utilizing Terahertz security technology is becoming increasingly critical. In this letter, a passive THz Human security image Intelligent detection method (HE-DETR-DC5) is proposed that combines histogram equalization (HE) processing and DEtection TRansformer (DETR). The proposed method first performs histogram equalization on passive THz images to improve image contrast. In addition, the cavity convolution is added to the ResNet-50 backbone network to enhance the model's attention to the concealed object in the passive THz image and improve the accuracy. The corresponding model is called DETR-DC5. The experimental results show that the average precision mAP of the proposed method can achieve 99.9%. At the same time, compared to other image enhancement methods, the histogram equalization method used in this letter has better performance and higher detection accuracy. The detection speed of the proposed method is about 0.5 FPS, which can meet the demand for real-time detection. Therefore, the proposed model can efficiently and automatically detect hidden objects in passive THz human security images.

Automated Road Damage Detection Using UAV Images and Deep Learning Techniques

This paper presents a novel automated road damage detection approach using Unmanned Aerial Vehicle (UAV) images and deep learning techniques. Maintaining road infrastructure is critical for ensuring a safe and sustainable transportation system. However, the manual collection of road damage data can be labor-intensive and unsafe for humans. Therefore, we propose using UAVs and Artificial Intelligence (AI) technologies to improve road damage detection’s efficiency and accuracy significantly. Our proposed approach utilizes three algorithms, YOLOv4, YOLOv5, and YOLOv7, for object detection and localization in UAV images. We trained and tested these algorithms using a combination of the RDD2022 dataset from China and a Spanish road dataset. The experimental results demonstrate that our approach is efficient and achieves 59.9% mean average precision mAP@.5 for the YOLOv5 version, 73.20% mAP@.5 for the YOLOv7 version, and 65.70% mAP@.5 for a YOLOv5 model with a Transformer Prediction Head. These results demonstrate the potential of using UAVs and deep learning for automated road damage detection and pave the way for future research in this field.

Distributed Banach-Picard Iteration: Application to Distributed Parameter Estimation and PCA

We recently proposed an algorithmic framework, <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">distributed Banach-Picard iteration</i> (DBPI), allowing a set of agents linked by a communication network to find a fixed point of a map that: (a) is the average of individual maps held by said agents; (b) is <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">locally contractive</i> (LC). Given such a map, DBPI yields a distributed algorithm provably inheriting the local linear convergence (LLC) of the standard Banach-Picard iteration for the centralized (average) map. Here, we instantiate DBPI in two classical problems, which amounts to proving that the conditions guaranteeing the LLC of DBPI hold. First, taking Sanger's algorithm for <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">principal component analysis</i> (PCA), we show that it corresponds to iterating an LC map that can be written as the average of local maps held by agents with private data subsets. Applying DBPI then recovers a previous distributed PCA algorithm, which lacked a convergence proof, thus closing that gap. In the second instantiation, we show that a variant of the <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">expectation-maximization</i> (EM) algorithm for parameter estimation from noisy, faulty measurements in sensor networks can be written as iterating an LC map that is the average of local maps. Consequently, the DBPI framework yields a distributed algorithm automatically inheriting the LLC guarantee of its centralized counterpart. Verifying the LC condition for EM is nontrivial (as the underlying operator depends on random samples) and a contribution in itself, possibly of independent interest. Finally, we illustrate experimentally the linear convergence of the proposed distributed EM algorithm, contrasting with the sub-linear rate of the previous version.

Polar particle flux distribution and its spatial extent

Context: The main challenge in atmospheric ionisation modelling is that sparse measurements are used to derive a global precipitation pattern. Typically this requires intense interpolation or scaling of long-term average maps. In some regions however, the particle flux might be similar and a combination of these regions would not limit the results even though it would dramatically improve the spatial and temporal data coverage. Aims: The paper intends to statistically analyse the particle flux distribution close to the geomagnetic poles labelled as Polar Particle Flux Distribution (PPFD) and identify similar distributions in neighbouring bins. Those bins are grouped and the size of the PPFD area is estimated. The benefit is that single measurements within the PPFD area should be able to represent the particle flux for the whole area at a given time. Methods: We use spatially binned energetic particle flux distributions measured by POES and Metop spacecraft during 2001–2018 to identify a Kp-dependent area with a similar flux distribution as the one found close to the geomagnetic poles (|magn.lat| &gt; 86°). First, the particle flux is mapped on a magnetic local time (MLT) vs. magnetic latitude grid. In the second step, the gridded data is split up according to Kp levels (forming the final bins). Third, the particle flux in every bin has been recalculated in order to replace zero-count rates with rates based on longer measurement periods which results in a more realistic low flux end of the particle distribution. Then the binned flux distributions are compared to the PPFD. A “Δ-test” indicates the similarity. A threshold for the Δ-test is defined using the standard deviation of Δ-test values inside the (|magn.lat| &gt; 86°) area. Bins that meet the threshold are attributed as PPFD area. Results: PPFDs and the corresponding PPFD areas have been determined for all investigated particle channels, covering an energy range of 154 eV–300 keV for electrons and 154 eV–2.5 MeV for protons. Concerning low energy channels a gradual flux increase with rising Kp has been identified. High energy channels show a combination of background population and solar particle event (SPE) population that adds up with increasing Kp. The size of the PPFD area depends on particle species, energy and geomagnetic disturbance, as well as MLT. The main findings are: a) There are small but characteristic hemispheric differences. b) Only above a certain energy threshold do the PPFD areas increase with particle energy. c) A clear enlargement with rising Kp is identified – with exceptions for very low Kp. d) The centre of the PPFD area is shifted towards midnight and moves with Kp. Asymmetries of the boundaries could be explained by auroral intensity. e) For low-energy particles the main restriction of the PPFD area seems to be the auroral precipitation.

Semantic and Temporal Contextual Correlation Learning for Weakly-Supervised Temporal Action Localization.

Weakly-supervised temporal action localization (WSTAL) aims to automatically identify and localize action instances in untrimmed videos with only video-level labels as supervision. In this task, there exist two challenges: (1) how to accurately discover the action categories in an untrimmed video (what to discover); (2) how to elaborately focus on the integral temporal interval of each action instance (where to focus). Empirically, to discover the action categories, discriminative semantic information should be extracted, while robust temporal contextual information is beneficial for complete action localization. However, most existing WSTAL methods ignore to explicitly and jointly model the semantic and temporal contextual correlation information for the above two challenges. In this article, a Semantic and Temporal Contextual Correlation Learning Network (STCL-Net) with the semantic (SCL) and temporal contextual correlation learning (TCL) modules is proposed, which achieves both accurate action discovery and complete action localization by modeling the semantic and temporal contextual correlation information for each snippet in the inter- and intra-video manners respectively. It is noteworthy that the two proposed modules are both designed in a unified dynamic correlation-embedding paradigm. Extensive experiments are performed on different benchmarks. On all the benchmarks, our proposed method exhibits superior or comparable performance in comparison to the existing state-of-the-art models, especially achieving gains as high as 7.2% in terms of the average mAP on THUMOS-14. In addition, comprehensive ablation studies also verify the effectiveness and robustness of each component in our model.

Research on Traffic Scene Recognition Algorithm Combining Object Detection and Semantic Segmentation

At present, the detection of instances in traffic scenes based on deep learning is mainly divided into two mainstream directions: object detection and semantic segmentation. Among them, object detection realizes the specific location of a single object in the road scene, and semantic segmentation realizes the pixel level classification of objects and background categories in the road scene. However, when pedestrian and other objects have occlusion problems, semantic segmentation is difficult to directly separate a single instance, and the anchor frame generated by object detection contains redundant information. To solve this problem, this paper proposes a method combining target detection and semantic segmentation. This method first uses YOLOv5 model to complete the target detection, and detects people, vehicles and other objects in the captured traffic scene image. At the same time, the improved DeepLabv3+ network model is used to capture the semantic and regional information of roads in the captured traffic scene image. Finally, the prediction results of the output of the two task branches are drawn in the image to be detected, and finally the drawing results are combined and output uniformly. This method can effectively distinguish different people, vehicles, roads and other information in the traffic scene, "complement each other" to understand the driverless road scene, and improve the detection accuracy. The experimental results show that the average mAP of this method is 79.11%, and the segmentation accuracy is high, which is suitable for the unmanned driving scene on real urban roads.

On approximating fixed points of weak enriched contraction mappings via Kirk's iterative algorithm in Banach spaces

Recently Berinde and Păcurar [Approximating fixed points of enriched contractions in Banach spaces. {\em J. Fixed Point Theory Appl.} {\bf 22} (2020), no. 2., 1--10], first introduced the idea of enriched contraction mappings and proved the existence of a fixed point of an enriched contraction mapping using the well-known fact that any fixed point of {the averaged mapping $T_\lambda$, where $\lambda\in (0,1]$, is also a fixed point of the initial mapping $T$}. In this work, we introduce the idea of weak enriched contraction mappings, and a new generalization of an averaged mapping called double averaged mapping. The first attempt is to prove the existence and uniqueness of the fixed point of a double averaged mapping associated with a weak enriched contraction mapping. Based on this result on Banach spaces, we give some sufficient conditions for the equality of all fixed points of a double averaged mapping and the set {of all fixed points of a weak enriched contraction mapping.} Moreover, our results show that an appropriate Kirk's iterative algorithm can be used to approximate a fixed point of a weak enriched contraction mapping. An illustrative example for showing the efficiency of our results is given.

Ultrafast prediction of somatic structural variations by filtering out reads matched to pan-genome k-mer sets.

Variant callers typically produce massive numbers of false positives for structural variations, such as cancer-relevant copy-number alterations and fusion genes resulting from genome rearrangements. Here we describe an ultrafast and accurate detector of somatic structural variations that reduces read-mapping costs by filtering out reads matched to pan-genome k-mer sets. The detector, which we named ETCHING (for efficient detection of chromosomal rearrangements and fusion genes), reduces the number of false positives by leveraging machine-learning classifiers trained with six breakend-related features (clipped-read count, split-reads count, supporting paired-end read count, average mapping quality, depth difference and total length of clipped bases). When benchmarked against six callers on reference cell-free DNA, validated biomarkers of structural variants, matched tumour and normal whole genomes, and tumour-only targeted sequencing datasets, ETCHING was 11-fold faster than the second-fastest structural-variant caller at comparable performance and memory use. The speed and accuracy of ETCHING may aid large-scale genome projects and facilitate practical implementations in precision medicine.

Identifying driving factors of the runoff coefficient based on the geographic detector model in the upper reaches of Huaihe River Basin

Abstract Various climate and watershed characteristics determine the runoff coefficient (RC), and their interactions are complicated. Understanding the driving factors of the RC is important for understanding the long-term water balance and how it might change. Using the upper reaches of the Huaihe River Basin as the study area, remote sensing data were used to produce a RC map. The geographical detector was selected to quantify the individual and interactive influences of 13 driving factors on the RC. The results revealed that moderate resolution imaging spectroradiometer evapotranspiration (ET) data can be used to produce a mean average RC map based on the water balance equation. The dominant factors influencing the RC were found to vary at different scales. Precipitation had the largest correlation coefficient with the RC at the watershed scale. For the pixel scale, results from the geographical detector indicated that actual evapotranspiration (AET) and precipitation had the highest explanatory rate for the RC in the small watershed region and the whole study area (0.785 and 0.248, respectively). Climate factors, elevation, and normalized difference vegetation index had a substantial influence on the RC. Any two factors exhibited bilinear or nonlinear enhanced relationships in their interactions. The largest interactions between the factors were AET and precipitation, which exceeded 0.900. This study serves to better understand and explain runoff’s complex interrelationships.

The confound of head position in within-session connectome fingerprinting in infants

Individuals differ in their functional connectome, which can be demonstrated using a “fingerprinting” analysis in which the connectome from an individual in one dataset is used to identify the same person from an independent dataset. Recently, the origin of these fingerprints has been studied by examining if they are present in infants. The results have varied considerably, with identification rates from 10 to 90%. When fingerprinting has been performed by splitting a single imaging session into two split-sessions (within session), identification rates were higher than when two full-sessions (between sessions) were compared. This study examined whether a methodological difference could account for this variation. It was hypothesized that the infant's exact head position in the head coil may affect the measured connectome, due to the gradual inhomogeneity of signal-to-noise in phased-array coils and the breadth of possible positions for a small infant head in a head coil. This study examined the impact of this using resting state functional MRI data from the Developing Human Connectome Project second release. Using functional timeseries, fingerprinting identification was high (84-91%) within a session while between sessions it was low (7%).Using N = 416 infants’ head positions, a map of the average signal-to-noise across the physical volume of the head coil was calculated and was used (independent group of 44 infants with two scan sessions) to demonstrate a significant relationship between head position in the head coil and functional connectivity. Using only the head positions (signal-to-noise values extrapolated from the group average map) of the independent group of 44 infants, high identification success was achieved across split-sessions (within session) but not full-sessions (between sessions). Using a model examining factors influencing the stability of the functional connectome, head position was seen as the strongest of the explanatory variables. We conclude within-session fingerprinting is affected by head position and future infant functional fingerprint analyses must use a different strategy or account for this impact.

MTSCANet: Multi temporal resolution temporal semantic context aggregation network

AbstractTemporal action localisation is a challenging task, and video context is crucial to localisation actions. Most existing cases that incorporate temporal and semantic contexts into video features suffer from single contextual representation and blurred temporal boundaries. In this study, a multi‐temporal resolution pyramid structure model is proposed. Firstly, a temporal‐semantic context aggregation module (TSCF) is designed to assign different attention weights to temporal contexts and combine them with multi‐level semantics into video features. Secondly, for the problem of large differences in the time span between different actions in the video, a local‐global attention module is designed to combine local and global temporal dependencies for each temporal point to obtain a more flexible and robust representation of contextual relations. The redundant representation of the convolution kernel is reduced by modifying the convolution and the arithmetic power is redeployed at a microscopic granularity. To verify the effectiveness of the model, extensive experiments on three challenging datasets are performed. On THUMOS14, the best performance is obtained in IoU@0.3–0.6 with an average mAP of 47.02%. On ActivityNet‐1.3, an average mAP of 34.94% was obtained. On HACS, an average mAP of 28.46% was achieved.

Multigraph Cluster Analysis of AD Candidate Genes via Human Brain Mapping

AbstractBackgroundBrain imaging genetics examines associations between imaging quantitative traits (QTs) and genetic factors such as single nucleotide polymorphisms (SNPs) to provide important insights into the pathogenesis of Alzheimer’s diseases (AD). Given the high dimensionality, the individual level SNP‐QT signals typically have small effect sizes and are hard to be detected and replicated. To overcome this limitation, this work proposes a new approach to identify high‐level imaging genetic associations through applying multi‐graph clustering to SNP‐QT association maps.MethodsParticipants include 255 cognitive normal (CN) and 218 late mild cognitive impairment (LMCI) subjects from the ADNI data. Linear regression is performed to regress regional AV‐45 measures on each of the 54 AD SNPs to obtain a brain association map across 116 AAL‐ROIs. To quantify the similarity between each SNP pair, 5 scoring functions are used to calculate 5 different similarity measures between the respective brain maps of the two SNPs. Each scoring function yields a similarity graph of all 54 SNPs. Multigraph min‐max clustering is applied to 5 different graphs simultaneously to group similar SNPs together. Functional annotation is performed on each SNP cluster and the corresponding average brain association map to provide high level interpretations.ResultsFigure 1 shows the genetic effect on regional AV‐45 measures for each pair of SNP and ROI. The multi‐graph clustering result is shown in Figure 2, where 54 AD‐related SNPs are clustered into two groups. Figure 3a shows the results of Enrichr Elsevier pathway analysis. SNP group 1 is associated with pathways such as amyloid beta clearance, while SNP group 2 is associated with pathways including amyloid beta formation and APP processing in AD. Figure 3b shows the NeuroSynth annotation of the average brain effect maps associated with the two SNP clusters, respectively.ConclusionWe proposed an approach that transforms individual level imaging genetic associations (SNP vs ROI) to high‐level associations (i.e., SNP cluster vs brain association map), and applied it to an LMCI study. These high‐level findings have the potential to provide valuable insights into relevant genetic pathways and brain circuits, which can help form new hypotheses for more detailed imaging and genetics studies in independent cohorts.

Machine Learning for Music Genre Classification Using Visual Mel Spectrum

Music is the most convenient and easy-to-use stress release tool in modern times. Many studies have shown that listening to appropriate music can release stress. However, since it is getting easier to make music, people only need to make it on the computer and upload it to streaming media such as Youtube, Spotify, or Beatport at any time, which makes it very infeasible to search a huge music database for music of a specific genre. In order to effectively search for specific types of music, we propose a novel method based on the visual Mel spectrum for music genre classification, and apply YOLOv4 as our neural network architecture. mAP was used as the scoring criterion of music genre classification in this study. After ten experiments, we obtained a highest mAP of 99.26%, and the average mAP was 97.93%.

Sixteen years of MOPITT satellite data strongly constrain Amazon CO fire emissions

Abstract. Despite the consensus on the overall downward trend in Amazon forest loss in the previous decade, estimates of yearly carbon emissions from deforestation still vary widely. Estimated carbon emissions are currently often based on data from local logging activity reports, changes in remotely sensed biomass, and remote detection of fire hotspots and burned area. Here, we use 16 years of satellite-derived carbon monoxide (CO) columns to constrain fire CO emissions from the Amazon Basin between 2003 and 2018. Through data assimilation, we produce 3 d average maps of fire CO emissions over the Amazon, which we verified to be consistent with a long-term monitoring programme of aircraft CO profiles over five sites in the Amazon. Our new product independently confirms a long-term decrease of 54 % in deforestation-related CO emissions over the study period. Interannual variability is large, with known anomalously dry years showing a more than 4-fold increase in basin-wide fire emissions relative to wet years. At the level of individual Brazilian states, we find that both soil moisture anomalies and human ignitions determine fire activity, suggesting that future carbon release from fires depends on drought intensity as much as on continued forest protection. Our study shows that the atmospheric composition perspective on deforestation is a valuable additional monitoring instrument that complements existing bottom-up and remote sensing methods for land-use change. Extension of such a perspective to an operational framework is timely considering the observed increased fire intensity in the Amazon Basin between 2019 and 2021.

B-YOLOX-S: A Lightweight Method for Underwater Object Detection Based on Data Augmentation and Multiscale Feature Fusion

With the increasing maturity of underwater agents-related technologies, underwater object recognition algorithms based on underwater robots have become a current hotspot for academic and applied research. However, the existing underwater imaging conditions are poor, the images are blurry, and the underwater robot visual jitter and other factors lead to lower recognition precision and inaccurate positioning in underwater target detection. A YOLOX-based underwater object detection model, B-YOLOX-S, is proposed to detect marine organisms such as echinus, holothurians, starfish, and scallops. First, Poisson fusion is used for data amplification at the input to balance the number of detected targets. Then, wavelet transform is used to perform Style Transfer on the enhanced images to achieve image restoration. The clarity of the images and detection targets is further increased and the generalization of the model is enhanced. Second, a combination of BIFPN-S and FPN is proposed to fuse the effective feature layer obtained by the Backbone layer to enhance the detection precision and accelerate model detection. Finally, the localization loss function of the prediction layer in the network is replaced by EIoU_Loss to heighten the localization precision in detection. Experimental results comparing the B-YOLOX-S algorithm model with mainstream algorithms such as FasterRCNN, YOLOV3, YOLOV4, YOLOV5, and YOLOX on the URPC2020 dataset show that the detection precision and detection speed of the algorithm model have obvious advantages over other algorithm networks. The average detection accuracy mAP value is 82.69%, which is 5.05% higher than the benchmark model (YOLOX-s), and the recall rate is 8.03% higher. Thus, the validity of the algorithmic model proposed in this paper is demonstrated.

Offshore wind power potential analysis for different wind turbines in the Mediterranean Region, 1959–2020

We evaluated spatiotemporal characteristics of the wind power density (WPD) in the Mediterranean and the Black Sea. An hourly wind speed data from ERA5 of ECMWF is used covering 62 years between 1959 and 2020. Spatial distributions of WPD at 100 m altitude at monthly, yearly, and seasonal timescales are revealed. The average WPD in the study area is 373 W/m2, where the spatial maximums in the 62-year average map reach up to 1116 W/m2. The spatial distribution of yearly mean WPD considering 13 sub-basins revealed that the two most prominent regions are the Gulf of Lion and the Aegean Sea. The diurnal variation of WPD showed over ± 10 % variability within the considered sub-basins. Five commercial turbines are selected to investigate their performance in the study area. The capacity factors, and operational and rated power time percentages are spatially mapped. The variability in the WPD is investigated using several indicators of yearly variability index, intra-annual variability index, coefficient of variation, robust coefficient of variation, inter-quartile range, and inter-quartile ratio. The long-term trend in the monthly-mean values is investigated and it is revealed that the statistically significant downward trends manifest themselves in the Levantine Basin, Ligurian-Provençal Basin, Balearic Sea, and the Alboran Sea. The capacity factors of five widely used wind turbines are computed in the study area using both yearly and seasonal timescales.

Abstract 11159: Assessment of Atrial Cardiomyopathy by High-Density Mapping Can Guide Additional Procedures in AF Ablation

Background: The efficacy of catheter ablation (CA) for atrial fibrillation (AF) has been lower in persistent AF (PeAF). Additional ablation strategies beyond pulmonary vein isolation (PVI) are not routinely recommended in PeAF because their efficacy is not established yet. The clinical AF patterns could not expect the degree of atrial abnormal voltage area (AVA). The inability to discriminate significant atrial cardiomyopathy may lead to diverse ablation outcomes beyond PVI in PeAF. Objectives: The objective of this study was to find out how much proportion of AVA can predict higher AF recurrence after PVI. Methods: We enrolled the consecutive 49 patients who underwent CA of AF. LA was mapped with Pentaray® catheter (2-6-2 spacing) during mid-CS pacing (600msec) using the CONFIDENCE Module. The peak-to-peak bipolar voltage was obtained using only 2mm spacing electrodes. The AVA was defined when the bipolar voltage was &lt; 1.5mV, which is known to be normal in LA. After excluding PVs, appendage, and mitral valve region, the proportion of AVA was calculated by the Carto 3 mapping system. The ablation strategy was PVI only regardless of the AF pattern &amp; degree of AVA. Results: The mean age was 58.1 ± 7.7 years, and 73.5% were male. The average mapping points were 3557.5 ± 760.9. The average proportion of AVA was 31.4 ± 15.3%. The paroxysmal AF (PAF) patients were 71.4%, and the mean AVA between PAF and PeAF was significantly different (28.7 ± 15.5 vs. 38.2 ± 12.7, p=0.049). The mean follow-up duration was 265.8 ± 150.4 days. Atrial tachyarrhythmia recurred in 35.7%. In receiver-operating curve analysis, the proportion of AVA of 33.4% predicts AF recurrence regardless of AF type with a sensitivity of 75% and specificity of 78.8% (AUC 0.864, P&lt;0.001). Conclusion: More than one-third of AVA assessed by high-density mapping could be used as a predictor of the recurrence of AF after CA, which may guide the consideration of additional ablation. However, a randomized controlled trial is warranted.

Identification and analysis of long-term changes in river deltas and riparian zones using time-series multispectral data

Today’s remote sensing data and technologies offer the capability to effectively monitor diverse and challenging environments around the world, such as coastal river and riparian zones. Coastal riparian zones and river deltas usually demonstrate extreme coastline changes in terms of the extent of water coverage of inland territories due to flood events, low and high tides, the climate, specific environmental characteristics, etc. In this paper, we exploit freely available multispectral time series data for previous decades, utilizing Landsat missions in order to develop an open-source-based image processing pipeline for the extraction of the actual yearly average coastline status of riparian river delta areas. The latter present significant temporal coastline changes between years, semesters, and months. Average mean maps are generated and then compared to several temporal levels in order to distinguish long-term significant changes and ecosystem threats. Additionally, a custom long short-term memory (LSTM) neural network is deployed to forecast the evolution of the coastline by exploiting the average value for each pixel across all available images as a training sample and producing a forecast output for the next period. The network achieves accuracy scores of 89.77% over 'non-water' depicting pixels and 84.26% over 'water' depicting pixels, for regions that present frequent changes between land and water coverage over time. The predicted map presents high statistical agreement with the respective average map generated in two different validation approaches, with kappa coefficients of 85.9% and 91.4%, respectively.

Cone of Vision as a Behavioural Cue for VR Collaboration

Mutual awareness of visual attention is essential for collaborative work. In the field of collaborative virtual environments (CVE), it has been proposed to use Field-of-View (FoV) frustum visualisations as a cue to support mutual awareness during collaboration. Recent studies on FoV frustum visualisations focus on asymmetric collaboration with AR/VR hardware setups and 3D reconstructed environments. In contrast, we focus on the general-purpose CVEs (i.e., VR shared offices), whose popularity is increasing due to the availability of low-cost headsets, and the restrictions imposed by the pandemic. In these CVEs collaboration roles are symmetrical, and the same 2D content available on desktop computers is displayed on 2D surfaces in a 3D space (VR screens). We prototyped one such CVE to evaluate FoV frustrum visualisation within this collaboration scenario. We also implement a FoV visualisation generated from an average fixation map (AFM), therefore directly generated by users' gaze behaviour which we call Cone of Vision (CoV). Our approach to displaying the frustum visualisations is tailored for 2D surfaces in 3D space and allows for self-awareness of this visual cue. We evaluate CoV in the context of a general exploratory data analysis (EDA) with 10 pairs of participants. Our findings indicate that CoV is beneficial during shifts between independent and collaborative work and supports collaborative progression across the visualisation. Self-perception of the CoV improves visual attention coupling, reduces the number of times users watch the collaborator's avatars and offers a consistent representation of the shared reality.