Convolutional Neuronal Networks Research Articles

Classification of Facial Expression of Post-Surgical Pain in Children

There are certain difficulties in differentiating between children's facial expression related to pain and other stimuli. In addition, the limited communication ability of children in the preverbal stage leads to misdiagnosis when the child feels pain, for example, post-surgical conditions. In this article, a classification approach of facial expression of child pain is presented based on models of pre-trained convolutional neuronal networks from the study carried out in a Colombian hospital of level 4 (Hospital Universitario San Vicente Fundación), in the recovery areas of child surgery services. AlexNet and VGG (16, 19 and Face) networks are evaluated in the own dataset using the FLACC scale and their performances are compared in three experiments. The results show that the VGG-19 model achieves the best performance (92.9%) compared to the other networks. The effectiveness of the model and transfer learning for the classification of facial expression of child pain shows a promising solution for the assessment of post-surgical pain.

Fetal birthweight prediction with measured data by a temporal machine learning method

BackgroundBirthweight is an important indicator during the fetal development process to protect the maternal and infant safety. However, birthweight is difficult to be directly measured, and is usually roughly estimated by the empirical formulas according to the experience of the doctors in clinical practice.MethodsThis study attempts to combine multiple electronic medical records with the B-ultrasonic examination of pregnant women to construct a hybrid birth weight predicting classifier based on long short-term memory (LSTM) networks. The clinical data were collected from 5,759 Chinese pregnant women who have given birth, with more than 57,000 obstetric electronic medical records. We evaluated the prediction by the mean relative error (MRE) and the accuracy rate of different machine learning classifiers at different predicting periods for first delivery and multiple deliveries. Additionally, we evaluated the classification accuracies of different classifiers respectively for the Small-for-Gestational-age (SGA), Large-for-Gestational-Age (LGA) and Appropriate-for-Gestational-Age (AGA) groups.ResultsThe results show that the accuracy rate of the prediction model using Convolutional Neuron Networks (CNN), Random Forest (RF), Linear-Regression, Support Vector Regression (SVR), Back Propagation Neural Network(BPNN), and the proposed hybrid-LSTM at the 40th pregnancy week for first delivery were 0.498, 0.662, 0.670, 0.680, 0.705 and 0.793, respectively. Among the groups of less than 39th pregnancy week, the 39th pregnancy week and more than 40th week, the hybrid-LSTM model obtained the best accuracy and almost the least MRE compared with those of machine learning models. Not surprisingly, all the machine learning models performed better than the empirical formula. In the SGA, LGA and AGA group experiments, the average accuracy by the empirical formula, logistic regression (LR), BPNN, CNN, RF and Hybrid-LSTM were 0.780, 0.855, 0.890, 0.906, 0.916 and 0.933, respectively.ConclusionsThe results of this study are helpful for the birthweight prediction and development of guidelines for clinical delivery treatments. It is also useful for the implementation of a decision support system using the temporal machine learning prediction model, as it can assist the clinicians to make correct decisions during the obstetric examinations and remind pregnant women to manage their weight.

Detection and Recognition of Moving Video Objects: Kalman Filtering with Deep Learning

Research in object recognition has lately found that Deep Convolutional Neuronal Networks (CNN) provide a breakthrough in detection scores, especially in video applications. This paper presents an approach for object recognition in videos by combining Kalman filter with CNN. Kalman filter is first applied for detection, removing the background and then cropping object. Kalman filtering achieves three important functions: predicting the future location of the object, reducing noise and interference from incorrect detections, and associating multi-objects to tracks. After detection and cropping the moving object, a CNN model will predict the category of object. The CNN model is built based on more than 1000 image of humans, animals and others, with architecture that consists of ten layers. The first layer, which is the input image, is of 100 * 100 size. The convolutional layer contains 20 masks with a size of 5 * 5, with a ruling layer to normalize data, then max-pooling. The proposed hybrid algorithm has been applied to 8 different videos with total duration of is 15.4 minutes, containing 23100 frames. In this experiment, recognition accuracy reached 100%, where the proposed system outperforms six existing algorithms.

Sound-based remote real-time multi-device operational monitoring system using a Convolutional Neural Network (CNN)

Smart factory is the main keyword in the field of manufacturing processes about the fourth industrial revolution. To realize the smart factory, making all pieces of device into smart devices that are connected to the centralized system to enable a real-time exchange of information is essential. Sound can be efficient means to make devices as smart devices because sound can contain the status information of various devices simultaneously, and it can be recorded easily outside of a device using only a microphone. In this study, multi-device operation monitoring system by analyzing sound is developed. Mic arrays for acquiring the sound were installed at the outside the devices and recorded the sounds from several devices simultaneously. By analyzing the recorded sound with log-mel spectrogram and Convolutional Neuron Network (CNN), the system could detect the operational status of three devices with an accuracy of 71–92 %. To improve the performance, virtual data set was created by composition of individual device operating sounds of different intensities. With this virtual data set, accuracy can be enhanced to 87 % ∼ 99 % accuracy and, required sound data amount could be reduced. Developed system was applied successfully in monitoring experiments in two different environments: a workshop in which hand-operated device was used and a factory with a computer numerical control machine and verifying the performance.

A Pipeline Leak Detection and Localization Approach Based on Ensemble TL1DCNN

There is an increasing need for timely pipeline leak detection and localization methods, pipeline leak could lead to not only the loss of the goods but also considerable environmental and economic problems. With the rapid development of hardware and software, the pipeline leak detection and localization algorithms have been widely researched and applied in many Fields. However, traditional methods are usually limited by extracting features manually, which is inefficient and time-consuming. Convolutional neuron network is an effective method to extract features automatically. In this paper, a novel ensemble transfer learning one-dimension convolutional neural network (TL1DCNN) for the pipeline leak detection and localization is proposed. The TL1DCNN plays the role of base learner. The results of a set of obtained base learners are integrated to achieve the task of pipeline leak detection and localization. Firstly, one-dimension convolutional neural network (1DCNN) models with different parameters are pretrained with source domain data. A small learning rate is set to retrain the above 1DCNN models for target task with target domain data in order to obtain TL1DCNN base learners. Then, the four ensemble strategies with different number base learners whose ensemble weights are optimized by particle swarm optimization algorithm are obtained by minimizing the sum of similarity. The dataset simulated by pipeline network model is used to evaluate the effectiveness of the proposed approach. The indicators such as classification accuracy, precision, recall, F_score and confusion matrix are used to compare the proposed approach with traditional methods and other deep learning methods. The experimental results show that the performance of the proposed approach is superior to other compared methods.

An Advanced Ontology based Deep Learning for Computer-aided Interpretation of Mammography Images

Computer assisted detection (CAD) are able to detect and characterize suspicious mammographic images, micro calcifications, masses or more difficult, architectural distortion. With the exploitation of these different characteristics, the system can specify and predict the severity of the tumor to assess the risk in terms of Malignity/Benignness. Our work involves the development of a new method for screening breast cancer, this is achieved by developing a whole strategy of knowledge extraction through deep learning and medical ontology appropriate for the classification of regions selected from digital mammograms, for each radiological sign considered, namely, masses and micro calcifications. First, we extracted the parameters characterizing the images used as input to a deep convolutional neuron network CNN. The learning is supervised because the images used are images from the MARATHON database of the University of Florida; they are already diagnosed by experts. The second phase aims to add a semantic level to our classification through a specialized ontology developed for this purpose based on the BIRADS characterization system. Based on the evaluation performed, the proposed approach provides better classification results than the usual methods for assisting in the computer aided detection of breast cancer.

ПРОПОЗИЦІЇ ЩОДО ПІДВИЩЕННЯ ТОЧНОСТІ СЕГМЕНТАЦІЇ МІСЬКИХ БУДОВ НА ЦИФРОВИХ КОСМІЧНИХ І АЕРОФОТОЗНІМКАХ ПРИ АВТОМАТИЗОВАНОМУ МОНІТОРИНГУ МІСЬКОГО СЕРЕДОВ

In basis of topography that is used in a military sphere, - a capture lies the methods of study of locality. On it, receipt of quantitative and quality descriptions of locality on space and to the airphotos - is one of basic parts of military topography presently. Architecture of U-Net zdkztncz effective enough for the decision of different tasks, such as segmentation of neuron structures, sciagraphy et cetera. A network is characterized by a coder with the sequence of levels of convolutional and maximal pool. Thus, the layer of decoding contains the mirror sequence of convolutional that transponirovan’s. He behaves as traditional automatic encode. Extractor of functions of U-Net can be modernized for the improvement of maps of segmentation of municipal structures. Unclear to the neuron decorate a pattern Vanga-Мandelja it maybe to use the classifier of the modified decoder in quality U-Net. Formal presentation of neuron network ensemble of neuron networks is in-process offered on the basis of the modified rolled up neuron network for segmentation of municipal structures (automatic encode) of U-Net as super position of functions. Architecture of ensemble of neuron networks is worked out on the basis of the modified convolutional neuron network (СNN) for segmentation of municipal structures (automatic encode) of U-Net. Modification (automatic encode) of U-Net is conducted due to modification of subnet of withdrawal of signs with using as encoder preliminary trained deep СNN VGG, and similarly due to realization of classifier of automatic encode with the use of the modified unclear neuron network Vanga-Мandelja on the basis of INМТ2 for on a pixel classification of certain municipal structures and the generalized topology of by a neuron network model is worked out for segmentation of municipal structures. Modification (automatic encode) of U-Net will allow to promote exactness of segmentation of municipal structures on digital space and airphotos at the automated monitoring of municipal environment.

Convolutional neuronal networks combined with X-ray phase-contrast imaging for a fast and observer-independent discrimination of cartilage and liver diseases stages

We applied transfer learning using Convolutional Neuronal Networks to high resolution X-ray phase contrast computed tomography datasets and tested the potential of the systems to accurately classify Computed Tomography images of different stages of two diseases, i.e. osteoarthritis and liver fibrosis. The purpose is to identify a time-effective and observer-independent methodology to identify pathological conditions. Propagation-based X-ray phase contrast imaging WAS used with polychromatic X-rays to obtain a 3D visualization of 4 human cartilage plugs and 6 rat liver samples with a voxel size of 0.7 × 0.7 × 0.7 µm3 and 2.2 × 2.2 × 2.2 µm3, respectively. Images with a size of 224 × 224 pixels are used to train three pre-trained convolutional neuronal networks for data classification, which are the VGG16, the Inception V3, and the Xception networks. We evaluated the performance of the three systems in terms of classification accuracy and studied the effect of the variation of the number of inputs, training images and of iterations. The VGG16 network provides the highest classification accuracy when the training and the validation-test of the network are performed using data from the same samples for both the cartilage (99.8%) and the liver (95.5%) datasets. The Inception V3 and Xception networks achieve an accuracy of 84.7% (43.1%) and of 72.6% (53.7%), respectively, for the cartilage (liver) images. By using data from different samples for the training and validation-test processes, the Xception network provided the highest test accuracy for the cartilage dataset (75.7%), while for the liver dataset the VGG16 network gave the best results (75.4%). By using convolutional neuronal networks we show that it is possible to classify large datasets of biomedical images in less than 25 min on a 8 CPU processor machine providing a precise, robust, fast and observer-independent method for the discrimination/classification of different stages of osteoarthritis and liver diseases.

Selecting proper combination of mpMRI sequences for prostate cancer classification using multi-input convolutional neuronal network.

This study aims to develop a deep-learning-basedmethod to classify clinically significant (CS) and clinically insignificant (CiS) prostate cancer (PCa) on multiparametric magnetic resonance imaging (mpMRI) automatically, and to select suitable mpMRI sequences for PCa classificationin different anatomic zones. A multi-input selection network (MISN) is proposed for both PCa classification and theselection of the optimal combination of sequences for PCa classification in a specific zone.MISN is a multi-input/-output classification network consisting of nine branches to process nine input images from the mpMRI data. To improve classification accuracy and reduce model parameters, a pruning strategy is proposed to select a subset of the nine branches of MIST to form two more effective networks for the peripheral zone (PZ) PCa and transition zone (TZ) PCa, which are named as PZN and TZN, respectively. Besides, a new penalized cross-entropy loss function is adopted to train the networks tobalance the classification sensitivity and specificity. The proposed methods were evaluated on the PROSTATEx challenge dataset and achieved an area under the receiver operator characteristics curve of 0.95, which was much higher than currently published results and ranked first out of more than 1500 entries submitted to the challenge at the time of submission of this paper. For PZ-PCa and TZ-PCa classification, PZN and TZN achieved better performance than MISN. Higher performance can be achieved by selecting a suitable subset of the mpMRI sequences in PCa classification.

Identification of Gastritis Subtypes by Convolutional Neuronal Networks on Histological Images of Antrum and Corpus Biopsies.

Background: Gastritis is a prevalent disease and commonly classified into autoimmune (A), bacterial (B), and chemical (C) type gastritis. While the former two subtypes are associated with an increased risk of developing gastric intestinal adenocarcinoma, the latter subtype is not. In this study, we evaluated the capability to classify common gastritis subtypes using convolutional neuronal networks on a small dataset of antrum and corpus biopsies. Methods: 1230 representative 500 × 500 µm images of 135 patients with type A, type B, and type C gastritis were extracted from scanned histological slides. Patients were allocated randomly into a training set (60%), a validation set (20%), and a test set (20%). One classifier for antrum and one classifier for corpus were trained and optimized. After optimization, the test set was analyzed using a joint result from both classifiers. Results: Overall accuracy in the test set was 84% and was particularly high for type B gastritis with a sensitivity of 100% and a specificity of 93%. Conclusions: Classification of gastritis subtypes is possible using convolutional neural networks on a small dataset of histopathological images of antrum and corpus biopsies. Deep learning strategies to support routine diagnostic pathology merit further evaluation.

Towards a computer aided diagnosis (CAD) for brain MRI glioblastomas tumor exploration based on a deep convolutional neuronal networks (D-CNN) architectures

Manuel brain glioblastomas tumor exploration through MRI modalities is time-consuming. It is considered as a harmful and critical task due to highly inhomogeneous tumor regions composition. For this reason, clinicians recommend the Computer-Aided Diagnosis (CAD) tools to ensure a more accurate diagnostic. Based on convolutional Deep-Learning algorithms, this paper investigates a fully automatic CAD for brain Glioblastomas tumors exploration including three steps: pre-processing, segmentation, and finally classification. A denoising and an automatic contrast enhancement method have been applied to preprocess the MRI scans. A Multi-Modal Cascaded U-net architecture, based on Fully Convolutional deep Network (FCN), has been adopted for the Region of Interest (ROI) extraction and finally, Deep Convolutional Neural Network (D-CNN) architecture has been used to classify brain glioblastomas tumor into High-Grade (HG) and Low-Grade (LG). Experiments were performed on the Multimodal Brain Tumor Segmentation Challenge BraTS-2018 datasets benchmark. Several validations metric have been adopted to assess the CAD’s performances. The Dice Metric (DM) parameter has been calculated between the obtained segmentation results and the available ground truth data. The accuracy parameter has been computed for classification performance evaluation. The higher DM and accuracy values could attest the performance and the efficiency of the proposed CAD tool.

A novel unified deep neural networks methodology for use by date recognition in retail food package image

There exist various types of information on retail food packages, including use by date, food product name and so on. The correct coding of use by dates on food packages is vitally important for avoiding potential health risks to customers caused by erroneous mislabelling of use by dates. It is extremely tedious and laborious to check the use by dates coding manually by a human operator, which is prone to generate errors thus an automatic system for validating the correctness of the coding of use by dates is needed. In order to construct such a system, firstly it needs to correctly automatic recognize use by dates on food packages. In this work, we propose a novel dual deep neural networks-based methodology for automatic recognition of use by dates in food package photographs recorded by a camera, which is a combination of two networks: a fully convolutional network for use by date ROI detection and a convolutional recurrent neuron network for date character recognition. The proposed methodology is the first attempt to apply deep learning for automatic use by date recognition. From comprehensive experimental evaluations, it is shown that the proposed method can achieve high accuracies in use by date recognition (more than 95% on our testing dataset), given food package images with varying lighting conditions, poor printing quality and varied textual/pictorial contents collected from multiple real retailer sites.

Evaluating body composition by combining quantitative spectral detector computed tomography and deep learning-based image segmentation

PurposeAim of this study was to develop and evaluate a software toolkit, which allows for a fully automated body composition analysis in contrast enhanced abdominal computed tomography leveraging the strengths of both, quantitative information from dual energy computed tomography and simple detection and segmentation tasks performed by deep convolutional neuronal networks (DCNN). Methods and materialsBoth, public and private datasets were used to train and validate DCNN. A combination of two DCNN and quantitative thresholding was used to classify axial CT slices to the abdominal region, classify voxels as fat and muscle and to differentiate between subcutaneous and visceral fat. For validation, patients undergoing repetitive examination (±21 days) and patients who underwent concurrent bioelectrical impedance analysis (BIA) were analyzed. Concordance correlation coefficient (CCC), linear regression and Bland-Altman-Analysis were used as statistical tests. ResultsResults provided from the algorithm toolkit were visually validated. The automated classifier was able to extract slices of interest from the full body scans with an accuracy of 98.7 %. DCNN-based segmentation for subcutaneous fat reached a mean dice similarity coefficient of 0.95. CCCs were 0.99 for both muscle and subcutaneous fat and 0.98 for visceral fat in patients undergoing repetitive examinations (n = 48). Further linear regression and Bland-Altman-Analyses suggested good agreement (r2:0.67−0.88) between the software toolkit and patients who underwent concurrent BIA (n = 39). ConclusionWe describe a software toolkit allowing for an accurate analysis of body composition utilizing a combination of DCNN- and threshold-based segmentations from spectral detector computed tomography.

Automated spheroid generation, drug application and efficacy screening using a deep learning classification: a feasibility study

The last two decades saw the establishment of three-dimensional (3D) cell cultures as an acknowledged tool to investigate cell behaviour in a tissue-like environment. Cells growing in spheroids differentiate and develop different characteristics in comparison to their two-dimensionally grown counterparts and are hence seen to exhibit a more in vivo-like phenotype. However, generating, treating and analysing spheroids in high quantities remains labour intensive and therefore limits its applicability in drugs and compound research. Here we present a fully automated pipetting robot that is able to (a) seed hanging drops from single cell suspensions, (b) treat the spheroids formed in these hanging drops with drugs and (c) analyse the viability of the spheroids by an image-based deep learning based convolutional neuronal network (CNN). The model is trained to classify between ‘unaffected’, ‘mildly affected’ and ‘affected’ spheroids after drug exposure. All corresponding spheroids are initially analysed by viability flow cytometry analysis to build a labelled training set for the CNN to subsequently reduce the number of misclassifications. Hence, this approach allows to efficiently examine the efficacy of drug combinatorics or new compounds in 3D cell culture. Additionally, it may provide a valuable instrument to screen for new and individualized systemic therapeutic strategies in second and third line treatment of solid malignancies using patient derived primary cells.

Deep Learning for the Classification of Small-Cell and Non-Small-Cell Lung Cancer.

Reliable entity subtyping is paramount for therapy stratification in lung cancer. Morphological evaluation remains the basis for entity subtyping and directs the application of additional methods such as immunohistochemistry (IHC). The decision of whether to perform IHC for subtyping is subjective, and access to IHC is not available worldwide. Thus, the application of additional methods to support morphological entity subtyping is desirable. Therefore, the ability of convolutional neuronal networks (CNNs) to classify the most common lung cancer subtypes, pulmonary adenocarcinoma (ADC), pulmonary squamous cell carcinoma (SqCC), and small-cell lung cancer (SCLC), was evaluated. A cohort of 80 ADC, 80 SqCC, 80 SCLC, and 30 skeletal muscle specimens was assembled; slides were scanned; tumor areas were annotated; image patches were extracted; and cases were randomly assigned to a training, validation or test set. Multiple CNN architectures (VGG16, InceptionV3, and InceptionResNetV2) were trained and optimized to classify the four entities. A quality control (QC) metric was established. An optimized InceptionV3 CNN architecture yielded the highest classification accuracy and was used for the classification of the test set. Image patch and patient-based CNN classification results were 95% and 100% in the test set after the application of strict QC. Misclassified cases mainly included ADC and SqCC. The QC metric identified cases that needed further IHC for definite entity subtyping. The study highlights the potential and limitations of CNN image classification models for tumor differentiation.

0792 Mandibular Movement Monitoring with Artificial Intelligence Analysis for the Diagnosis of Sleep Bruxism

Abstract Introduction Sleep bruxism (BXM) is the result of rhythmic muscular masticatory activity (RMMA) and can be captured by masseters surface electromyography (sEMG). Despite the multiple adverse negative consequences of BXM, a simple reliable home diagnostic device is currently unavailable, with in laboratory audio-video polysomnography (type I PSG) remaining the gold standard diagnostic tool. Mandibular movements (MM) recordings during sleep can readily identify RMMA, are simple to set up and can be easily repeated from night to night. Here, we aimed to identify stereotypical MM in patients with BXM, and to develop RMMA automatic detection and BXM diagnosis using an artificial intelligence-based approach. Methods MM were recorded by a dedicated sensor (Sunrise, Namur, Belgium) in 12 patients with BXM during type I PSG. The Sunrise system consists of a coin-sized hardware that is comfortably placed on the subject’s chin. Its embedded inertial measurement unit communicates via Bluetooth with a smartphone and automatically transfers MM signals to a cloud-based infrastructure at the end of the night. Data processing and analysis are then performed in Python programming language. A time series cluster analysis was applied to sequences of masseters sEMG and MM signals during BXM episodes (n=300) and during spontaneous micro-arousals (n=300). Then, a convolutional neuronal network (CNN) was developed to identify BXM and distinguish it from spontaneous micro-arousals while exclusively relying on MM signal. Results Based on the cluster analysis, BXM periods were characterized by a specific pattern of MM signals (higher frequency and amplitude), which was closely associated with the sEMG signals but clearly differed from the MM signal patterns during micro-arousals. CNN-based classifier distinguished the BXM events from other RMMAs during micro-arousals and respiratory efforts with an overall accuracy of 91%. Conclusion Sleep bruxism can be automatically identified, quantified, and characterized with mandibular movements analysis supported by artificial intelligence technology. Support This work was supported by the French National Research Agency (ANR-12-TECS-0010), in the framework of the “Investissements d’avenir” program (ANR-15-IDEX-02). https://life.univ-grenoble-alpes.fr.

Chern-Simons Current of Left and Right Chiral Superspace in Graphene Wormhole

Starting from the basic definitions of Chern-Simons current, it is possible to calculate its values with a quantum machine learning approach, the so-called supersymmetric support Dirac machine. The related supercurrent is generated from the coupling between three states of the quantum flux of a modified Wilson loop of Cooper pairs. We adopt the Holo-Hilbert spectrum, in frequency modulation, to visualize the network as the coupling of convolutional neuron network in a superstatistic theory where the theory of superconductors is applied. According to this approach, it is possible to calculate the number of carbon atoms in the spinor network of a graphene wormhole. A supercurrent of Cooper pairs is produced as graviphoton states by using the Holo-Hilbert spectral analysis.

Classification of hyperspectral images of the interior of fruits and vegetables using a 2D convolutional neuronal network

This work evaluates the performance of a Deep Learning technique for classification of challenging hyperspectral images of the interior of fruits and vegetables when they are combined. Some of these samples have low contrast, similar colour features, and their skins or characteristic shapes are lost when cut to expose their interiors. We implemented a two-dimensional convolutional neural network for this classification task and compared their results against the technique of support vector machines. We randomly selected a group of 13 hyperspectral images from a public database containing information of the interior of 42 fruits and vegetables. Using parts of these 13 selected images, we constructed three artificial hyperspectral images merging these parts differently. We applied the two proposed techniques over the three of them. The comparison of the classification results shows that the two-dimensional convolutional neural network over-performs the support vector machine in all three composite images. The two-dimensional convolutional neural network exceeded 98% classification accuracy in all of them. These results show that the two-dimensional convolutional network benefits from the spatial and spectral data in the images obtaining proper levels of classification even in samples mixed in complex contexts, as it can occur in the food or pharmaceutical industries.

GPS-based citywide traffic congestion forecasting using CNN-RNN and C3D hybrid model

Traffic congestion prediction in citywide road networks is a challenging research field in metropolitan transportation operation and management. Recent advances in GPS technology offer great opportunities to improve upon the limitations on the availability and quality of traffic data. Motivated by the success of deep neural networks and considering the spatial dependencies and temporal evolutions of network traffic, we propose an innovative deep learning-based mapping to cube architecture for network-wide urban traffic forecasting. Experiments using real Taxi GPS vehicle trajectory data confirm the accuracy and effectiveness of the proposed approach combining 3-Dimensional Convolutional Networks (C3D) with Convolutional Neuron Networks (CNNs) and Recurrent Neuron Networks (RNNs), called CRC3D as a hybrid method integrating CNN-RNNs and C3Ds. We also compared a variety of recurrent neural network architectures. Results show that CRC3D succeeds in inheriting the advantages of C3D and CNN-RNN, and show its consistent and satisfactory results in urban complex system.

Seismic phase picking in China Seismic Array using a deep convolutional neuron network*

Seismic phase picking is the preliminary work of earthquake location and body-wave travel time tomography. Manual picking is considered as the most accurate way to access the arrival times but time consuming. Many automatic picking methods were proposed in the past decades, but their precisions are not as high as human experts especially for events with low ratio of signal to noise and later arrivals. As the increasing deployment of large seismic array, the existing methods can not meet the requirements of quick and accurate phase picking. In this study, we applied a phase picking algorithm developed on the base of deep convolutional neuron network (PickNet) to pick seismic phase arrivals in ChinArray-Phase III. The comparison of picking error of PickNet and the traditional method shows that PickNet is capable of picking more precise phases and can be applied in a large dense array. The raw picked travel-time data shows a large variation deviated from the traveltime curves. The absolute location residual is a key criteria for travel-time data selection. Besides, we proposed a flowchart to determine the accurate location of the single-station earthquake via dense seismic array and phase arrival picked by PickNet. This research expands the phase arrival dataset and improves the location accuracy of single-station earthquake.