Search results for: feature pyramid network

Authors: Pengfei Meng, Shuangcheng Jia, Qian Li

Abstract:

We proposed a real-time high-precision semantic segmentation network based on a multi-resolution feature fusion module, the auxiliary feature extracting module, upsampling module, and atrous spatial pyramid pooling (ASPP) module. We designed a feature fusion structure, which is integrated with sufficient features of different resolutions. We also studied the effect of side-branch structure on the network and made discoveries. Based on the discoveries about the side-branch of the network structure, we used a side-branch auxiliary feature extraction layer in the network to improve the effectiveness of the network. We also designed upsampling module, which has better results than the original upsampling module. In addition, we also re-considered the locations and number of atrous spatial pyramid pooling (ASPP) modules and modified the network structure according to the experimental results to further improve the effectiveness of the network. The network presented in this paper takes the backbone network of Bisenetv2 as a basic network, based on which we constructed a network structure on which we made improvements. We named this network deep multiple-resolution bilateral networks for real-time, referred to as DMBR-Net. After experimental testing, our proposed DMBR-Net network achieved 81.2% mIoU at 119FPS on the Cityscapes validation dataset, 80.7% mIoU at 109FPS on the CamVid test dataset, 29.9% mIoU at 78FPS on the COCOStuff test dataset. Compared with all lightweight real-time semantic segmentation networks, our network achieves the highest accuracy at an appropriate speed.

Keywords: multi-resolution feature fusion, atrous convolutional, bilateral networks, pyramid pooling

Procedia PDF Downloads 149

Authors: Jianghao Hu, Hongyu Wang

Abstract:

The effective and efficient human pose estimation algorithm is an important task for real-time human pose estimation on mobile devices. This paper proposes a light-weight human key points detection algorithm, Light-Weight Network for Real-Time Pose Estimation (LWPE). LWPE uses light-weight backbone network and depthwise separable convolutions to reduce parameters and lower latency. LWPE uses the feature pyramid network (FPN) to fuse the high-resolution, semantically weak features with the low-resolution, semantically strong features. In the meantime, with multi-scale prediction, the predicted result by the low-resolution feature map is stacked to the adjacent higher-resolution feature map to intermediately monitor the network and continuously refine the results. At the last step, the key point coordinates predicted in the highest-resolution are used as the final output of the network. For the key-points that are difficult to predict, LWPE adopts the online hard key points mining strategy to focus on the key points that hard predicting. The proposed algorithm achieves excellent performance in the single-person dataset selected in the AI (artificial intelligence) challenge dataset. The algorithm maintains high-precision performance even though the model only contains 3.9M parameters, and it can run at 225 frames per second (FPS) on the generic graphics processing unit (GPU).

Keywords: depthwise separable convolutions, feature pyramid network, human pose estimation, light-weight backbone

Procedia PDF Downloads 152

Authors: Saroj Bijarnia, Preety Singh

Abstract:

We propose a simple and effective biometrics system based on face verification across aging using a new variant of texture feature, Pyramid Binary Pattern. This employs Local Binary Pattern along with its hierarchical information. Dimension reduction of generated texture feature vector is done using Principal Component Analysis. Support Vector Machine is used for classification. Our proposed method achieves an accuracy of 92:24% and can be used in an automated age-invariant face verification system.

Keywords: biometrics, age invariant, verification, support vector machine

Procedia PDF Downloads 349

Authors: Avatharam Ganivada, Krishna Shah

Abstract:

The Mask R-CNN is recently introduced by the team of Facebook AI Research (FAIR), which is mainly concerned with instance segmentation in images. Here, the Mask R-CNN is based on ResNet and feature pyramid network (FPN), where a single dropout method is employed. This paper provides a modified Mask R-CNN by adding multiple dropout methods into the Mask R-CNN. The proposed model has also utilized the concepts of Resnet and FPN to extract stage-wise network feature maps, wherein a top-down network path having lateral connections is used to obtain semantically strong features. The proposed model produces three outputs for each object in the image: class label, bounding box coordinates, and object mask. The performance of the proposed network is evaluated in the segmentation of every instance in images using COCO and cityscape datasets. The proposed model achieves better performance than the state-of-the-networks for the datasets.

Keywords: instance segmentation, object detection, convolutional neural networks, deep learning, computer vision

Procedia PDF Downloads 72

Authors: Xing Yan Fan, Xing Lin Xu, Man Yuan Chen, Pei Tzu Lee, Yu Ting Wang, Yi Xiao Cao, Rui Yao

Abstract:

Pyramid sales have been a widespread issue in China. Victims who are defrauded not only lose money but damage interpersonal relationship. A deeper understanding of pyramid-sale models can be beneficial to prevent potential victims from fraud and improve the property security. The goals of this study were to detect psychological characteristics of pyramid-sale sellers, and analyse persuasive skills in pyramid organizations. A qualitative study was conducted in this study. Participants (n=6) recruited by 'snowball' sampling from present pyramid-sale sellers (n=3) and imprisoned pyramid-sale sellers (n=3). All participants accepted semi-structured interview for collecting data. Content analysis was adopted for data coding and analysis. The results indicate that pyramid organizations are used to utilize their appearance packaging and celebrity effect to strengthen the positions in participants’ mind. The status gap between pyramid-sale sellers in same organization, as well as rewards to increase reputation, are used to motivate participants in pyramid. The most significant common characteristics among all participants are that they tend to possess a high sense of belongingness within the firm. Moreover, the expression of pyramid-sale sellers on gambling mentality is expected to growth as constantly losing money. Findings suggest that the psychological characteristics of pyramid-sale sellers in accordance with Maslow’s hierarchy of needs, persuasive skills of pyramid organization confront to 'attitude-behaviour change model'. These findings have implication on 'immune education' that providing guidance for victims out of stuck and protecting ordinary people from the jeopardizing of pyramid sales.

Keywords: pyramid sales, characteristics, persuasive skills, qualitative study

Procedia PDF Downloads 254

Authors: Ma Lei-Lei, Zhou You

Abstract:

Aiming at the problems of low detection accuracy of deep learning algorithm in traffic sign detection, this paper proposes improved EfficientDet based traffic sign detection algorithm. Multi-head self-attention is introduced in the minimum resolution layer of the backbone of EfficientDet to achieve effective aggregation of local and global depth information, and this study proposes an improved feature fusion pyramid with increased vertical cross-layer connections, which improves the performance of the model while introducing a small amount of complexity, the Balanced L1 Loss is introduced to replace the original regression loss function Smooth L1 Loss, which solves the problem of balance in the loss function. Experimental results show, the algorithm proposed in this study is suitable for the task of traffic sign detection. Compared with other models, the improved EfficientDet has the best detection accuracy. Although the test speed is not completely dominant, it still meets the real-time requirement.

Keywords: convolutional neural network, transformer, feature pyramid networks, loss function

Procedia PDF Downloads 96

Authors: Jingting Luo, Yong Wang, Ying Wang

Abstract:

Pedestrian detection in image or video data is a very important and challenging task in security surveillance. The difficulty of this task is to locate and detect pedestrians of different scales in complex scenes accurately. To solve these problems, a deep neural network (RT-YOLOv3) is proposed to realize real-time pedestrian detection at different scales in security monitoring. RT-YOLOv3 improves the traditional YOLOv3 algorithm. Firstly, the deep residual network is added to extract vehicle features. Then six convolutional neural networks with different scales are designed and fused with the corresponding scale feature maps in the residual network to form the final feature pyramid to perform pedestrian detection tasks. This method can better characterize pedestrians. In order to further improve the accuracy and generalization ability of the model, a hybrid pedestrian data set training method is used to extract pedestrian data from the VOC data set and train with the INRIA pedestrian data set. Experiments show that the proposed RT-YOLOv3 method achieves 93.57% accuracy of mAP (mean average precision) and 46.52f/s (number of frames per second). In terms of accuracy, RT-YOLOv3 performs better than Fast R-CNN, Faster R-CNN, YOLO, SSD, YOLOv2, and YOLOv3. This method reduces the missed detection rate and false detection rate, improves the positioning accuracy, and meets the requirements of real-time detection of pedestrian objects.

Keywords: pedestrian detection, feature detection, convolutional neural network, real-time detection, YOLOv3

Procedia PDF Downloads 141

Authors: Muhammad Azam, Micheal Olaolu Arowolo, Fei He, Mihail Popescu, Dong Xu

Abstract:

The number of biological papers is growing quickly, which means that the number of biological pathway figures in those papers is also increasing quickly. Each pathway figure shows extensive biological information, like the names of genes and how the genes are related. However, manually annotating pathway figures takes a lot of time and work. Even though using advanced image understanding models could speed up the process of curation, these models still need to be made more accurate. To improve gene name recognition from pathway figures, we applied a Siamese network to map image segments to a library of pictures containing known genes in a similar way to person recognition from photos in many photo applications. We used a triple loss function and a triplet spatial pyramid pooling network by combining the triplet convolution neural network and the spatial pyramid pooling (TSPP-Net). We compared VGG19 and VGG16 as the Siamese network model. VGG16 achieved better performance with an accuracy of 93%, which is much higher than OCR results.

Keywords: biological pathway, image understanding, gene name recognition, object detection, Siamese network, VGG

Procedia PDF Downloads 288

Authors: Yagya Raj Pandeya, Joonwhoan Lee

Abstract:

The dynamic behavior of music and video makes it difficult to evaluate musical instrument playing in a video by computer system. Any television or film video clip with music information are rich sources for analyzing musical instruments using modern machine learning technologies. In this research, we integrate the audio and video information sources using convolutional neural network (CNN) and pass network learned features through recurrent neural network (RNN) to preserve the dynamic behaviors of audio and video. We use different pre-trained CNN for music and video feature extraction and then fine tune each model. The music network use 2D convolutional network and video network use 3D convolution (C3D). Finally, we concatenate each music and video feature by preserving the time varying features. The long short term memory (LSTM) network is used for long-term dynamic feature characterization and then use late fusion with generalized mean. The proposed network performs better performance to recognize the musical instrument using audio-video multimodal neural network.

Keywords: multimodal, 3D convolution, music-video feature extraction, generalized mean

Procedia PDF Downloads 214

Authors: A. A. Abaker Omer, H. B. Mohamed Balh, W. Liu, A. Abas, J. Yu, S. Li, W. Ma, W. El Kolaly, Y. Y. Ahmed Abuker

Abstract:

We have discovered an important technical solution that could make new approaches in the processing of wet silicon etching, especially in the production of photovoltaic cells. During its inferior light-trapping and structural properties, the inverted pyramid structure outperforms the conventional pyramid textures and black silicone. The traditional pyramid textures and black silicon can only be accomplished with more advanced lithography, laser processing, etc. Importantly, our data demonstrate the feasibility of an inverted pyramidal structure of silicon via one-step Cu-catalyzed chemical etching (CCCE) in Cu (NO₃)₂/HF/H₂O₂/H₂O solutions. The effects of etching time and reaction temperature on surface geometry and light trapping were systematically investigated. The conclusion shows that the inverted pyramid structure has ultra-low reflectivity of ~4.2% in the wavelength of 300~1000 nm; introduce of Cu particles can significantly accelerate the dissolution of the silicon wafer. The etching and the inverted pyramid structure formation mechanism are discussed. Inverted pyramid structure with outstanding anti-reflectivity includes useful applications throughout the manufacture of semi-conductive industry-compatible solar cells, and can have significant impacts on industry colleagues and populations.

Keywords: Cu-catalyzed chemical etching, inverted pyramid nanostructured, reflection, solar cells

Procedia PDF Downloads 153

Authors: J. K. Alhassan, C. S. Actsu

Abstract:

This study is on the prediction of feature performance of undergraduate students with Artificial Neural Networks (ANN). With the growing decline in the quality academic performance of undergraduate students, it has become essential to predict the students’ feature academic performance early in their courses of first and second years and to take the necessary precautions using such prediction-based information. The feed forward multilayer neural network model was used to train and develop a network and the test carried out with some of the input variables. A result of 80% accuracy was obtained from the test which was carried out, with an average error of 0.009781.

Keywords: academic performance, artificial neural network, prediction, students

Procedia PDF Downloads 466

Authors: Khalaf Khatatneh, Nabeel Al-Milli, Amjad Hudaib, Monther Ali Tarawneh

Abstract:

Software fault prediction identify potential faults in software modules during the development process. In this paper, we present a novel approach for software fault prediction by combining a feedforward neural network with particle swarm optimization (PSO). The PSO algorithm is employed as a feature selection technique to identify the most relevant metrics as inputs to the neural network. Which enhances the quality of feature selection and subsequently improves the performance of the neural network model. Through comprehensive experiments on software fault prediction datasets, the proposed hybrid approach achieves better results, outperforming traditional classification methods. The integration of PSO-based feature selection with the neural network enables the identification of critical metrics that provide more accurate fault prediction. Results shows the effectiveness of the proposed approach and its potential for reducing development costs and effort by detecting faults early in the software development lifecycle. Further research and validation on diverse datasets will help solidify the practical applicability of the new approach in real-world software engineering scenarios.

Keywords: feature selection, neural network, particle swarm optimization, software fault prediction

Procedia PDF Downloads 93

Authors: Krishan Kumar, Mohit Mittal, Pramod Kumar

Abstract:

Wireless sensor network is one of the most promising communication networks for monitoring remote environmental areas. In this network, all the sensor nodes are communicated with each other via radio signals. The sensor nodes have capability of sensing, data storage and processing. The sensor nodes collect the information through neighboring nodes to particular node. The data collection and processing is done by data aggregation techniques. For the data aggregation in sensor network, clustering technique is implemented in the sensor network by implementing self-organizing feature map (SOFM) neural network. Some of the sensor nodes are selected as cluster head nodes. The information aggregated to cluster head nodes from non-cluster head nodes and then this information is transferred to base station (or sink nodes). The aim of this paper is to manage the huge amount of data with the help of SOM neural network. Clustered data is selected to transfer to base station instead of whole information aggregated at cluster head nodes. This reduces the battery consumption over the huge data management. The network lifetime is enhanced at a greater extent.

Keywords: artificial neural network, data clustering, self organization feature map, wireless sensor network

Procedia PDF Downloads 516

Authors: Xiaoxiao Li, Shuangcheng Jia, Qian Li

Abstract:

Keypoint detection has always been a challenge in the field of image recognition. This paper proposes a novelty keypoint detection method which is called Multi-Scale Feature Fusion Convolutional Network with Attention (MFFCNA). We verified that the multi-scale features with the attention mechanism module have better feature expression capability. The feature fusion between different scales makes the information that the network model can express more abundant, and the network is easier to converge. On our self-made street sign corner dataset, we validate the MFFCNA model with an accuracy of 97.8% and a recall of 81%, which are 5 and 8 percentage points higher than the HRNet network, respectively. On the COCO dataset, the AP is 71.9%, and the AR is 75.3%, which are 3 points and 2 points higher than HRNet, respectively. Extensive experiments show that our method has a remarkable improvement in the keypoint recognition tasks, and the recognition effect is better than the existing methods. Moreover, our method can be applied not only to keypoint detection but also to image classification and semantic segmentation with good generality.

Keywords: keypoint detection, feature fusion, attention, semantic segmentation

Procedia PDF Downloads 117

Authors: Ayda Kurtulus

Abstract:

Heydar Aliyev Centre and Quéops Pyramid are two contrasting examples of sacred geometry perceived as metaphorical alchemy by linking cosmos and earth. Zaha Hadid’s modern building has a wave-like shape and semi-circular alternations that show fluidity and movement, while The Great Pyramid of Giza is triangular. The centre is reminding of the shape of planets, an attempt to regain the balance lost in the modern-day capitalist world, while the Great Pyramid of Giza represents a vortex of energy that connects heaven and earth, harmony and balance. The sacred geometric shapes link the past and the future through God and Ra, humanism and spiritualism in an architectural evolution continuum, mind and spirit into one. An analysis of two geometrical forms, a semi-circle, and a triangle, were carried out through a comprehensive literature review, indicating that behind the materialistic perceptual beauty of buildings, ancient and contemporary, there are mathematical and sacred geometrical constructions that add value to one superficiality can interpret.

Keywords: architectural shapes, The Great Pyramid of Giza, Heydar Aliyev Centre, sacred geometry, philosophy

Procedia PDF Downloads 106

Authors: Laicheng Cao, Xiangqian Su, Youxiao Wu

Abstract:

Feature selection is one of the important problems in network security, pattern recognition, data mining and other fields. In order to remove redundant features, effectively improve the detection speed of intrusion detection system, proposes a new feature selection method, which is based on the invasive weed optimization (IWO) algorithm and tabu search algorithm(TS). Use IWO as a global search, tabu search algorithm for local search, to improve the results of IWO algorithm. The experimental results show that the feature selection method can effectively remove the redundant features of network data information in feature selection, reduction time, and to guarantee accurate detection rate, effectively improve the speed of detection system.

Keywords: intrusion detection, feature selection, iwo, tabu search

Procedia PDF Downloads 529

Authors: Santhosh K. S. Annayappa, Nupur Pruthi

Abstract:

The region of dentate nucleus is a common site for various pathologies like hematomas, tumours, etc. We aimed to study in detail the relationship of this region with the vermis, especially the pyramid using microscopic fibre dissection technique. To achieve this aim, 20 cerebellar hemispheres were studied from the 11 cerebellums. Dissection was performed using wooden spatulas and micro dissectors under a microscope following Klingler’s preservation technique. The relationship between the pyramid of vermis and the dentate nucleus was studied in detail. A similar relationship was studied on the MRI of randomly selected trigeminal neuralgia patients and correlated with anatomical findings. Results show the mean distance of the lateral margin of the dentate nucleus from the midline on anatomic specimens was 21.4 ± 1.8 mm (19-25 mm) and 23.4 ± 3.4 mm (15-29 mm) on right and left side, respectively. Similar measurements made on the MRI were 22.97 ± 2.0 mm (20.03-26.15 mm) on the right side and 23.98 ± 2.1 mm (21.47-27.67 mm) on the left side. The amount of white matter dissection required to reach the dentate nucleus at the pyramidal attachment area was 7.3 ± 1.0 mm (6-9 mm) on the right side and 6.8 ± 1.4 mm (5-10 mm) on the left side. It was concluded that the pyramid of vermis has a constant relationship with the dentate nucleus and can be used as an excellent landmark during surgery to localise the dentate nucleus on the suboccipital surface.

Keywords: fiber dissection, micro neurosurgery, the dentate nucleus of cerebellum, the pyramid of vermis

Procedia PDF Downloads 153

Authors: Jiahao Cui

Abstract:

In the past centuries, due to climate change and intense whaling, the global whale population has dramatically declined. Among the various whale species, the fin whale experienced the most drastic drop in number due to its popularity in whaling. Under this background, identifying fin whale calls could be immensely beneficial to the preservation of the species. This paper uses feature extraction to process the input audio signal, then a network based on AlexNet and three networks based on the ResNet model was constructed to classify fin whale calls. A mixture of the DOSITS database and the Watkins database was used during training. The results demonstrate that a modified ResNet network has the best performance considering precision and network complexity.

Keywords: convolutional neural network, ResNet, AlexNet, fin whale preservation, feature extraction

Procedia PDF Downloads 121

Authors: Chih-Fong Tsai, Ya-Han Hu

Abstract:

The aim of feature selection (or dimensionality reduction) is to filter out unrepresentative features (or variables) making the classifier perform better than the one without feature selection. Since there are many well-known feature selection algorithms, and different classifiers based on different selection results may perform differently, very few studies consider examining the effect of performing different feature selection algorithms on the classification performances by different classifiers over different types of datasets. In this paper, two widely used algorithms, which are the genetic algorithm (GA) and information gain (IG), are used to perform feature selection. On the other hand, three well-known classifiers are constructed, which are the CART decision tree (DT), multi-layer perceptron (MLP) neural network, and support vector machine (SVM). Based on 14 different types of datasets, the experimental results show that in most cases IG is a better feature selection algorithm than GA. In addition, the combinations of IG with DT and IG with SVM perform best and second best for small and large scale datasets.

Keywords: data mining, feature selection, pattern classification, dimensionality reduction

Procedia PDF Downloads 668

Authors: Salam M. H. Kareem

Abstract:

Instructional or teaching procedures and their proper sequence are essential for high-quality learning outcomes. These actions are the path that the teacher takes during the learning process after setting the learning objectives. Teachers and specialists in the education field should include teaching procedures with putting in place an effective mechanism for the procedure’s implementation to achieve a logical sequence with the desired output of overall education process. Determining the sequence of these actions may be a strategic process outlined by a strategic educational plan or drawn by teachers with a high level of experience, enabling them to determine those logical procedures. While specific actions may be necessary for a specific form, many Physical Education (PE) teachers can work out on various sports disciplines. This study was conducted to investigate the impact of using the teaching sequence of the teaching pyramid in raising the level of enjoyment in swimming classes. Four months later of teaching swimming skills to the control and experimental groups of the study, we figured that using the tools shown in the teaching pyramid with the experimental group led to statistically significant differences in the positive tendencies of students to participate in the swimming classes by using the traditional procedures of teaching and using of successive procedures in the teaching pyramid, and in favor of the teaching pyramid, The students are influenced by enhancing their tendency to participate in swimming classes when the teaching procedures followed are sensitive to individual differences and are based on the element of pleasure in learning, and less positive levels of the tendency of students when using traditional teaching procedures, by getting the level of skills' requirements higher and more difficult to perform. The level of positive tendencies of students when using successive procedures in the teaching pyramid was increased, by getting the level of skills' requirements higher and more difficult to perform, because of the high level of motivation and the desire to challenge the self-provided by the teaching pyramid.

Keywords: physical education, swimming classes, teaching process, teaching pyramid

Procedia PDF Downloads 145

Authors: Jihan Abdulazeez Ahmed, Adnan Mohsin Abdulazeez Brifcani

Abstract:

This paper suggests a new internal architecture of holon based on feature selection model using the combination of Bees Algorithm (BA) and Artificial Neural Network (ANN). BA is used to generate features while ANN is used as a classifier to evaluate the produced features. Proposed system is applied on the Wine data set, the statistical result proves that the proposed system is effective and has the ability to choose informative features with high accuracy.

Keywords: artificial neural network, bees algorithm, feature selection, Holon

Procedia PDF Downloads 456

Authors: Pengchao Wei, Xiangzhong Fang

Abstract:

Noise is ubiquitous in this world. Denoising is an essential technology, especially in image semantic segmentation, where noises are generally categorized into two main types i.e. feature noise and label noise. The main focus of this paper is aiming at modeling label noise, investigating the behaviors of different types of label noise on image semantic segmentation tasks using K-Nearest-Neighbor and Convolutional Neural Network classifier. The performance without label noise and with is evaluated and illustrated in this paper. In addition to that, the influence of feature noise on the image semantic segmentation task is researched as well and a feature noise reduction method is applied to mitigate its influence in the learning procedure.

Keywords: convolutional neural network, denoising, feature noise, image semantic segmentation, k-nearest-neighbor, label noise

Procedia PDF Downloads 219

Authors: Omer Cahana, Ofer Levi, Maya Herman

Abstract:

Magnetic Resonance Imaging (MRI) is a lengthy medical scan that stems from a long acquisition time. Its length is mainly due to the traditional sampling theorem, which defines a lower boundary for sampling. However, it is still possible to accelerate the scan by using a different approach such as Compress Sensing (CS) or Parallel Imaging (PI). These two complementary methods can be combined to achieve a faster scan with high-fidelity imaging. To achieve that, two conditions must be satisfied: i) the signal must be sparse under a known transform domain, and ii) the sampling method must be incoherent. In addition, a nonlinear reconstruction algorithm must be applied to recover the signal. While the rapid advances in Deep Learning (DL) have had tremendous successes in various computer vision tasks, the field of MRI reconstruction is still in its early stages. In this paper, we present an end-to-end method for MRI reconstruction from k-space to image. Our method contains two parts. The first is sensitivity map estimation (SME), which is a small yet effective network that can easily be extended to a variable number of coils. The second is reconstruction, which is a top-down architecture with lateral connections developed for building high-level refinement at all scales. Our method holds the state-of-art fastMRI benchmark, which is the largest, most diverse benchmark for MRI reconstruction.

Keywords: magnetic resonance imaging, image reconstruction, pyramid network, deep learning

Procedia PDF Downloads 91

Authors: Sandra Mitrovic, Gaurav Singh

Abstract:

In many industries, including telecommunications, churn prediction has been a topic of active research. A lot of attention has been drawn on devising the most informative features, and this area of research has gained even more focus with spread of (social) network analytics. The call detail records (CDRs) have been used to construct customer networks and extract potentially useful features. However, to the best of our knowledge, no studies including network features have yet proposed a generic way of representing network information. Instead, ad-hoc and dataset dependent solutions have been suggested. In this work, we build upon a recently presented method (node2vec) to obtain representations for nodes in observed network. The proposed approach is generic and applicable to any network and domain. Unlike node2vec, which assumes a static network, we consider a dynamic and time-evolving network. To account for this, we propose an approach that constructs the feature representation of each node by generating its node2vec representations at different timestamps, concatenating them and finally compressing using an auto-encoder-like method in order to retain reasonably long and informative feature vectors. We test the proposed method on churn prediction task in telco domain. To predict churners at timestamp ts+1, we construct training and testing datasets consisting of feature vectors from time intervals [t1, ts-1] and [t2, ts] respectively, and use traditional supervised classification models like SVM and Logistic Regression. Observed results show the effectiveness of proposed approach as compared to ad-hoc feature selection based approaches and static node2vec.

Keywords: churn prediction, dynamic networks, node2vec, auto-encoders

Procedia PDF Downloads 314

Authors: Bubai Maji, Monorama Swain

Abstract:

Speech is considered an essential and most natural medium for the interaction between machines and humans. However, extracting effective features for speech emotion recognition (SER) is remains challenging. The present studies show that the temporal information captured but high-level temporal-feature learning is yet to be investigated. In this paper, we present an efficient novel method using the Self-attention (SA) mechanism in a combination of Convolutional Neural Network (CNN) and Bi-directional Gated Recurrent Unit (Bi-GRU) network to learn high-level temporal-feature. In order to further enhance the representation of the high-level temporal-feature, we integrate a Bi-GRU output with learnable weights features by SA, and improve the performance. We evaluate our proposed method on our created SITB-OSED and IEMOCAP databases. We report that the experimental results of our proposed method achieve state-of-the-art performance on both databases.

Keywords: Bi-GRU, 1D-CNNs, self-attention, speech emotion recognition

Procedia PDF Downloads 112

Authors: Li Qingjian, Li Ke, He Chun, Huang Yong

Abstract:

In this paper, the method of combining the Pohl Seidman's deep belief network with the self-organizing neural network is proposed to classify the target. This method is mainly aimed at the high nonlinearity of the hyperspectral image, the high sample dimension and the difficulty in designing the classifier. The main feature of original data is extracted by deep belief network. In the process of extracting features, adding known labels samples to fine tune the network, enriching the main characteristics. Then, the extracted feature vectors are classified into the self-organizing neural network. This method can effectively reduce the dimensions of data in the spectrum dimension in the preservation of large amounts of raw data information, to solve the traditional clustering and the long training time when labeled samples less deep learning algorithm for training problems, improve the classification accuracy and robustness. Through the data simulation, the results show that the proposed network structure can get a higher classification precision in the case of a small number of known label samples.

Keywords: DBN, SOM, pattern classification, hyperspectral, data compression

Procedia PDF Downloads 340

Authors: Kyugneun Lee, Ikjin Lee

Abstract:

Parameter estimation with inverse problem often suffers from unfavorable conditions in the real world. Useless data and many input parameters make the problem complicated or insoluble. Data refinement and reformulation of the problem can solve that kind of difficulties. In this research, a method to solve the rank deficient inverse problem is suggested. A multi-physics system which has rank deficiency caused by response correlation is treated. Impeditive information is removed and the problem is reformulated to sequential estimations using Bayesian network (BN) and subset groups. At first, subset grouping of the responses is performed. Feature selection with singular value decomposition (SVD) is used for the grouping. Next, BN inference is used for sequential conditional estimation according to the group hierarchy. Directed acyclic graph (DAG) structure is organized to maximize the estimation ability. Variance ratio of response to noise is used to pairing the estimable parameters by each response.

Keywords: Bayesian network, feature selection, rank deficiency, statistical inverse analysis

Procedia PDF Downloads 313

Authors: Han Xue, Zhang Lanyue

Abstract:

In order to seek more effective feature extraction technology, feature extraction method based on MFCC combined with vector hydrophone is exposed in the paper. The sound pressure signal and particle velocity signal of two kinds of ships are extracted by using MFCC and its evolution form, and the extracted features are fused by using fisher-ratio and correlated distance criterion. The features are then identified by BP neural network. The results showed that MFCC, First-Order Differential MFCC and Second-Order Differential MFCC features can be used as effective features for recognition of underwater targets, and the fusion feature can improve the recognition rate. Moreover, the results also showed that the recognition rate of the particle velocity signal is higher than that of the sound pressure signal, and it reflects the superiority of vector signal processing.

Keywords: vector information, MFCC, differential MFCC, fusion feature, BP neural network

Procedia PDF Downloads 527

Authors: Mohd Zamri Osman, Mohd Aizaini Maarof, Mohd Foad Rohani

Abstract:

Human skin detection recognized as the primary step in most of the applications such as face detection, illicit image filtering, hand recognition and video surveillance. The performance of any skin detection applications greatly relies on the two components: feature extraction and classification method. Skin color is the most vital information used for skin detection purpose. However, color feature alone sometimes could not handle images with having same color distribution with skin color. A color feature of pixel-based does not eliminate the skin-like color due to the intensity of skin and skin-like color fall under the same distribution. Hence, the statistical color analysis will be exploited such mean and standard deviation as an additional feature to increase the reliability of skin detector. In this paper, we studied the effectiveness of statistical color feature for human skin detection. Furthermore, the paper analyzed the integrated color and texture using eight classifiers with three color spaces of RGB, YCbCr, and HSV. The experimental results show that the integrating statistical feature using Random Forest classifier achieved a significant performance with an F1-score 0.969.

Keywords: color space, neural network, random forest, skin detection, statistical feature

Procedia PDF Downloads 461

Authors: Shimaa Holail, Tamer Saleh, Xiongwu Xiao

Abstract:

Building change detection (BCD) from satellite imagery is an essential topic in urbanization monitoring, agricultural land management, and updating geospatial databases. Recently, methods for detecting changes based on deep learning have made significant progress and impressive results. However, it has the problem of being insensitive to changes in buildings with complex spectral differences, and the features being extracted are not discriminatory enough, resulting in incomplete buildings and irregular boundaries. To overcome these problems, we propose a dual Siamese network based on the Unet model with the addition of a deep supervision strategy (DS) in this paper. This network consists of a backbone (encoder) based on ImageNet pre-training, a fusion block, and feature pyramid networks (FPN) to enhance the step-by-step information of the changing regions and obtain a more accurate BCD map. To train the proposed method, we created a new dataset (EGY-BCD) of high-resolution and multi-temporal aerial images captured over New Cairo in Egypt to detect building changes for this purpose. The experimental results showed that the proposed method is effective and performs well with the EGY-BCD dataset regarding the overall accuracy, F1-score, and mIoU, which were 91.6 %, 80.1 %, and 73.5 %, respectively.

Keywords: building change detection, deep supervision, semantic segmentation, EGY-BCD dataset

Procedia PDF Downloads 118