Search results for: visual saliency detection

Authors: Olivia A. Kjorlien, Maryam Asghari, Farshid Alizadeh-Shabdiz

Abstract:

The goal of this paper is to investigate the impact of mosaic augmentation on the performance of object detection solutions. To carry out the study, YOLOv4 and YOLOv4-Tiny models have been selected, which are popular, advanced object detection models. These models are also representatives of two classes of complex and simple models. The study also has been carried out on two categories of objects, simple and complex. For this study, YOLOv4 and YOLOv4 Tiny are trained with and without mosaic augmentation for two sets of objects. While mosaic augmentation improves the performance of simple object detection, it deteriorates the performance of complex object detection, specifically having the largest negative impact on the false positive rate in a complex object detection case.

Keywords: accuracy, false positives, mosaic augmentation, object detection, YOLOV4, YOLOV4-Tiny

Procedia PDF Downloads 83

Authors: S. Probine

Abstract:

The visual arts have been proven to be a central means through which young children can communicate their ideas, reflect on experience, and construct new knowledge. Despite this, perceptions of, and the degree to which the visual arts are valued within education, vary widely within political, educational, community and family contexts. These differing perceptions informed my doctoral research project, which explored the contextual factors that affect how young children come to value and use the visual arts in their lives and learning. The qualitative methodology of narrative inquiry with inclusion of arts-based methods was most appropriate for this inquiry. Using a sociocultural framework, the stories collected were analysed through the sociocultural theories of Lev Vygotsky as well as the work of Urie Bronfenbrenner, together with postmodern theories about identity formation. The use of arts-based methods such as teacher’s reflective art journals and the collection of images by child participants and their parent/caregivers allowed the research participants to have a significant role in the research. Three early childhood settings at which the visual arts were deeply valued as a meaning-making device in children’s learning, were purposively selected to be involved in the research. At each setting, the study found a unique and complex web of influences and interconnections, which shaped how children utilised the visual arts to mediate their thinking. Although the teachers' practices at all three centres were influenced by sociocultural theories, each settings' interpretations of these theories were unique and resulted in innovative interpretations of the role of the teacher in supporting visual arts learning. These practices had a significant impact on children’s experiences of the visual arts. For many of the children involved in this study, visual art was the primary means through which they learned. The children in this study used visual art to represent their experiences, relationships, to explore working theories, their interests (including those related to popular culture), to make sense of their own and other cultures, and to enrich their imaginative play. This research demonstrates that teachers have fundamental roles in fostering and disseminating the importance of the visual arts within their educational communities.

Keywords: arts-based methods, early childhood education, teacher's visual arts pedagogies, visual arts

Procedia PDF Downloads 109

Authors: P. Siriarchawatana, K. Leungchavaphongse, N. Covavisaruch, K. Rojananuangnit, P. Boondaeng, N. Panyayingyong

Abstract:

Glaucoma is a disease that causes visual loss in adults. Glaucoma causes damage to the optic nerve and its overall pathophysiology is still not fully understood. Vasculopathy may be one of the possible causes of nerve damage. Photographic imaging of retinal vessels by fundus camera during eye examination may complement clinical management. This paper presents an innovation for measuring central light reflex width-to-retinal vessel diameter ratio (CRR) from digital retinal photographs. Using our edge detection technique, CRRs from glaucoma and normal eyes were compared to examine differences and associations. CRRs were evaluated on fundus photographs of participants from Mettapracharak (Wat Raikhing) Hospital in Nakhon Pathom, Thailand. Fifty-five photographs from normal eyes and twenty-one photographs from glaucoma eyes were included. Participants with hypertension were excluded. In each photograph, CRRs from four retinal vessels, including arteries and veins in the inferotemporal and superotemporal regions, were quantified using edge detection technique. From our finding, mean CRRs of all four retinal arteries and veins were significantly higher in persons with glaucoma than in those without glaucoma (0.34 vs. 0.32, p < 0.05 for inferotemporal vein, 0.33 vs. 0.30, p < 0.01 for inferotemporal artery, 0.34 vs. 0.31, p < 0.01 for superotemporal vein, and 0.33 vs. 0.30, p < 0.05 for superotemporal artery). From these results, an increase in CRRs of retinal vessels, as quantitatively measured from fundus photographs, could be associated with glaucoma.

Keywords: glaucoma, retinal vessel, central light reflex, image processing, fundus photograph, edge detection

Procedia PDF Downloads 295

Authors: R. Rama Kishore, Sunesh Malik

Abstract:

Development in the usage of internet for different purposes in recent times creates great threat for the copy right protection of the digital images. Digital watermarking is the best way to rescue from the said problem. This paper presents detailed review of the different watermarking techniques, latest trends in the field and categorized like spatial and transform domain, blind and non-blind methods, visible and non visible techniques etc. It also discusses the different optimization techniques used in the field of watermarking in order to improve the robustness and imperceptibility of the method. Different measures are discussed to evaluate the performance of the watermarking algorithm. At the end, this paper proposes a watermarking algorithm using (k.n) shares of halftone visual cryptography (HVC) instead of (2, 2) share cryptography. (k,n) shares visual cryptography improves the security of the watermark. As halftone is a method of reprographic, it helps in improving the visual quality of watermark image. The proposed method uses fractional transformation to improve the robustness of the copyright protection of the method.

Keywords: digital watermarking, fractional transform, halftone, visual cryptography

Procedia PDF Downloads 320

Authors: Anton Stadler, Thorsten Ike

Abstract:

In this paper, we present a video based smoke detection algorithm based on TVL1 optical flow estimation. The main part of the algorithm is an accumulating system for motion angles and upward motion speed of the flow field. We optimized the usage of TVL1 flow estimation for the detection of smoke with very low smoke density. Therefore, we use adapted flow parameters and estimate the flow field on difference images. We show in theory and in evaluation that this improves the performance of smoke detection significantly. We evaluate the smoke algorithm using videos with different smoke densities and different backgrounds. We show that smoke detection is very reliable in varying scenarios. Further we verify that our algorithm is very robust towards crowded scenes disturbance videos.

Keywords: low density, optical flow, upward smoke motion, video based smoke detection

Procedia PDF Downloads 321

Authors: Salina Budin, Nor Liawati Abu Othman, Shaira Ismail

Abstract:

This study attempts to determine kinesthetic, verbal and visual intelligences among mechanical engineering undergraduate students and explores any probable relation with students’ learning styles and academic performance. The questionnaire used in this study is based on Howard Gardner’s multiple intelligences theory comprising of five elements of learning style; environmental, sociological, emotional, physiological and psychological. Questionnaires are distributed amongst undergraduates in the Faculty of Mechanical Engineering. Additional questions on students’ perception of learning styles and their academic performance are included in the questionnaire. The results show that one third of the students are strongly dominant in the kinesthetic intelligent (33%), followed by a combination of kinesthetic and visual intelligences (29%) and 21% are strongly dominant in all three types of intelligences. There is a statistically significant correlation between kinesthetic, verbal and visual intelligences and students learning styles and academic performances. The ANOVA analysis supports that there is a significant relationship between academic performances and level of kinesthetic, verbal and visual intelligences. In addition, it has also proven a remarkable relationship between academic performances and kinesthetic, verbal and visual learning styles amongst the male and female students. Thus, it can be concluded that, academic achievements can be enhanced by understanding as well as capitalizing the students’ types of intelligences and learning styles.

Keywords: kinesthetic intelligent, verbal intelligent, visual intelligent, learning style, academic performances

Procedia PDF Downloads 272

Authors: Kuang-Hsiung Tan, Chih-Chan Hu, Chien-Wu Lan, Shih-Sung Lin, Te-Jen Chang

Abstract:

An active islanding detection method using disturbance signal injection with intelligent controller is proposed in this study. First, a DC\AC power inverter is emulated in the distributed generator (DG) system to implement the tracking control of active power, reactive power outputs and the islanding detection. The proposed active islanding detection method is based on injecting a disturbance signal into the power inverter system through the d-axis current which leads to a frequency deviation at the terminal of the RLC load when the utility power is disconnected. Moreover, in order to improve the transient and steady-state responses of the active power and reactive power outputs of the power inverter, and to further improve the performance of the islanding detection method, two probabilistic fuzzy neural networks (PFNN) are adopted to replace the traditional proportional-integral (PI) controllers for the tracking control and the islanding detection. Furthermore, the network structure and the online learning algorithm of the PFNN are introduced in detail. Finally, the feasibility and effectiveness of the tracking control and the proposed active islanding detection method are verified with experimental results.

Keywords: distributed generators, probabilistic fuzzy neural network, islanding detection, non-detection zone

Procedia PDF Downloads 360

Authors: Carlos Alberto Riveros, Edwin Fabián García, Javier Enrique Rivero

Abstract:

The measured data obtained from sensors in continuous monitoring of civil structures are mainly used for modal identification and damage detection. Therefore when modal identification analysis is carried out the quality in the identification of the modes will highly influence the damage detection results. It is also widely recognized that the usefulness of the measured data used for modal identification and damage detection is significantly influenced by the number and locations of sensors. The objective of this study is the numerical implementation of two widely known optimum sensor placement methods in beam-like structures

Keywords: optimum sensor placement, structural damage detection, modal identification, beam-like structures.

Procedia PDF Downloads 407

Authors: Jie Yang, Jian-Min Meng

Abstract:

A GPU-based floating object detection scheme is presented in this paper which is designed for floating mine detection tasks. This system uses contrast and motion information to eliminate as many false positives as possible while avoiding false negatives. The GPU computation platform is deployed to allow detecting objects in real-time. From the experimental results, it is shown that with certain configuration, the GPU-based scheme can speed up the computation up to one thousand times compared to the CPU-based scheme.

Keywords: object detection, GPU, motion estimation, parallel processing

Procedia PDF Downloads 444

Authors: Diego Luján Villarreal

Abstract:

Visual prostheses are designed to repair some eyesight in patients blinded by photoreceptor diseases, such as retinitis pigmentosa (RP) and age-related macular degeneration (AMD). Electrode-to-cell proximity has drawn attention due to its implications on secure single-localized stimulation. Yet, few techniques are available for understanding the relationship between the number of cells activated and the current injection. We propose an answering technique by solving the governing equation for time-dependent electrical currents using finite element discretization to obtain the volume of stimulation.

Keywords: visual prosthetic devices, volume for stimulation, FEM discretization, 3D simulation

Procedia PDF Downloads 41

Authors: Ya-Chuan Tseng, Hsin-Yi Liu, Meei-Fang Lou, Guey-Shiun Huang

Abstract:

Purpose: Sensory impairments are common in older adults. Hearing and visual impairments affect their physical and mental health and quality of life (QOL) adversely. However, systematic reviews of the relationship between hearing impairment, visual impairment, dual sensory impairment and quality of life are scarce. The purpose of this systematic review was to determine the relationship between hearing impairment, visual impairment, dual sensory impairment and quality of life. Methods: Searches of EMBASE, PubMed, CINAHL, MEDLINE, Cochrane Library and Airiti Library were conducted between January 2006 and December 2017 using the keywords ‘quality of life,’ ‘life satisfaction,’ ‘well-being,’ ‘hearing impairment’ and ‘visual impairment’ Two authors independently assessed methodologic quality using a modified Downs and Black tool. Data were extracted by the first author and then cross-checked by the second author. Results: Twenty-three studies consisting mostly of community-dwelling older adults were included in our review. Sensory impairment was found to be in significant association with quality of life, with an increase in hearing impairment or visual impairment severity resulting in a lower quality of life. Quality of life for dual sensory impairment was worse than for hearing impairment or visual impairment individually. Conclusions: A significant association was confirmed between hearing impairment, visual impairment, dual sensory impairment and quality of life. Our review can be used to enhance health care personnel’s understanding of sensory impairment in older adults and enable healthcare personnel to actively assess older adults’ sensory functions so that they can help alleviate the negative impact of sensory impairments on QOL in older adults.

Keywords: nursing, older adults, quality of life, systematic review, hearing impairment, visual impairment

Procedia PDF Downloads 195

Authors: Vedant Subhash

Abstract:

This paper discusses the physics of the detection of thermal neutrons using thin-film coated semiconductor detectors. The thermal neutron detection efficiency as a function of film thickness is calculated for the front and back irradiation detector devices coated with ¹⁰B, ⁶LiF, and pure Li thin films. The detection efficiency for back irradiation devices is 4.15% that is slightly higher than that for front irradiation detectors, 4.0% for ¹⁰B films of thickness 2.4μm. The theoretically calculated thermal neutron detection efficiency using ¹⁰B film thickness of 1.1 μm for the back irradiation device is 3.0367%, which has an offset of 0.0367% from the experimental value of 3.0%. The detection efficiency values are compared and proved consistent with the given calculations.

Keywords: detection efficiency, neutron detection, semiconductor detectors, thermal neutrons

Procedia PDF Downloads 108

Authors: Pooja Sharma, Renu Jain, Alka David

Abstract:

The primary goal of this research is to comprehend the many features of visual merchandising (VM) and impulsive or instant purchasing behavior. It aims to explain the link between visual merchandising and customer purchasing behavior. The reviews were compiled from research articles, professional journal articles, and the opinions of many authors. It also discusses the impact of different internal and external VM elements on instant purchasing. The visual merchandising elements are divided into two sections: interior element (inside the display, spaces, and layout, fixtures, mannequins, attention-grabbing device) and outside element (outside display, space, and layout, fixture, mannequins, attention-grabbing device) (Window Display, Exterior signs, Marquees, Entrance, color, and texture). By focusing on selected clothing stores from the four markets of Bhopal city, we discovered that the exterior elements (window display, color, and texture) and interior elements (mannequins like dummies and fixtures such as lighting) have a significant positive impact on instant buying among the elements of Visual merchandising.

Keywords: instant purchase, visual merchandising, instant buying behavior, consumer behavior, window display, fixtures, mannequins, marquees

Procedia PDF Downloads 86

Authors: Diego Luján Villarreal

Abstract:

Retinal prosthetic devices aim to repair some vision in visual impairment patients by stimulating electrically neural cells in the visual system. In this study, the 3D linear electrode carrier is presented. A simulation framework was developed by placing the 3D carrier 1 mm away from the fovea center at the highest-density cell. Cell stimulation is verified in COMSOL Multiphysics by developing a 3D computational model which includes the relevant retinal interface elements and dynamics of the voltage-gated ionic channels. Current distribution resulting from low threshold amplitudes produces a small volume equivalent to the volume confined by individual cells at the highest-density cell using small-sized electrodes. Delicate retinal tissue is protected by excessive charge density

Keywords: retinal prosthetic devices, visual devices, retinal implants., visual prosthetic devices

Procedia PDF Downloads 77

Authors: Manuel Alves, Angelica Reis, Armindo Lobo, Valdemar Leiras

Abstract:

In industry 4.0, it is common to have a lot of sensor data. In this deluge of data, hints of possible problems are difficult to spot. The digital twin concept aims to help answer this problem, but it is mainly used as a monitoring tool to handle the visualisation of data. Failure detection is of paramount importance in any industry, and it consumes a lot of resources. Any improvement in this regard is of tangible value to the organisation. The aim of this paper is to add the ability to forecast test failures, curtailing detection times. To achieve this, several anomaly detection algorithms were compared with a symbolic regression approach. To this end, Isolation Forest, One-Class SVM and an auto-encoder have been explored. For the symbolic regression PySR library was used. The first results show that this approach is valid and can be added to the tools available in this context as a low resource anomaly detection method since, after training, the only requirement is the calculation of a polynomial, a useful feature in the digital twin context.

Keywords: anomaly detection, digital twin, industry 4.0, symbolic regression

Procedia PDF Downloads 89

Authors: Ye Wang, Ickjai Lee

Abstract:

Due to the widespread of mobile sensing, there is a strong need to handle trails of moving objects, trajectories. This paper proposes three visual analytic approaches for higher order information of trajectory data sets based on the higher order Voronoi diagram data structure. Proposed approaches reveal geometrical information, topological, and directional information. Experimental results demonstrate the applicability and usefulness of proposed three approaches.

Keywords: visual analytics, higher order information, trajectory datasets, spatio-temporal data

Procedia PDF Downloads 382

Authors: S. Ghasemi, K. Khorasani

Abstract:

In this paper, the problem of fault detection and isolation in the attitude control subsystem of spacecraft formation flying is considered. In order to design the fault detection method, an extended Kalman filter is utilized which is a nonlinear stochastic state estimation method. Three fault detection architectures, namely, centralized, decentralized, and semi-decentralized are designed based on the extended Kalman filters. Moreover, the residual generation and threshold selection techniques are proposed for these architectures.

Keywords: component, formation flight of satellites, extended Kalman filter, fault detection and isolation, actuator fault

Procedia PDF Downloads 410

Authors: Soni Ariawan

Abstract:

The study aims to explore the cultural diversity highlighted in EFL textbook for Senior High School grade 10 in Indonesia. The visual images are selected as the data and qualitatively analysed using content analysis. The reason to choose visual images because images are not always neutral and they might impact teaching and learning process. In the current study, cultural diversity aspects are focused on religion (Muslim, Protestant, Catholic, Hindu, Buddhist, Confucian), gender (male, female, unclear), ethnic (Melanesian, Austronesian, Foreigner) and socioeconomic (low, middle, high, undetermined) diversity as the theoretical framework. The four aspects of cultural diversity are sufficiently representative to draw a conclusion in investigating Indonesian culture representation in EFL textbook. The finding shows that cultural diversity is not proportionally reflected in the textbook, particularly in the visual images.

Keywords: EFL textbook, cultural diversity, visual images, Indonesia

Procedia PDF Downloads 289

Authors: Jacek Turski

Abstract:

The visual space geometries are constructed in the Riemannian geometry framework from simulated iso-disparity conics in the horizontalvisual plane of the binocular system with the asymmetric eyes (AEs). For the eyes fixating at the abathic distance, which depends on the AE’s parameters, the iso-disparity conics are frontal straight lines in physical space. For allother fixations, the iso-disparity conics consist of families of the ellipses or hyperbolas depending on both the AE’s parameters and the bifoveal fixation. However, the iso-disparity conic’s arcs are perceived in the gaze direction asthe frontal lines and are referred to as visual geodesics. Thus, geometriesof physical and visual spaces are different. A simple postulate that combines simulated iso-disparity conics with basic anatomy od the human visual system gives the relative depth for the fixation at the abathic distance that establishes the Riemann matric tensor. The resulting geodesics are incomplete in the gaze direction and, therefore, give thefinite distances to the horizon that depend on the AE’s parameters. Moreover, the curvature vanishes in this eyes posture such that visual space is flat. For all other fixations, only the sign of the curvature canbe inferred from the global behavior of the simulated iso-disparity conics: the curvature is positive for the elliptic iso-disparity curves and negative for the hyperbolic iso-disparity curves.

Keywords: asymmetric eye model, iso-disparity conics, metric tensor, geodesics, curvature

Procedia PDF Downloads 123

Authors: Raffaele A. Calogero

Abstract:

RNAseq represents an attractive methodology for the detection of functional genomic variants. RNAseq results obtained from polyA+ RNA selection protocol (POLYA) and from exonic regions capturing protocol (ACCESS) indicate that ACCESS detects 10% more coding SNV/INDELs with respect to POLYA. ACCESS requires less reads for coding SNV detection with respect to POLYA. However, if the analysis aims at identifying SNV/INDELs also in the 5’ and 3’ UTRs, POLYA is definitively the preferred method. No particular advantage comes from ACCESS or POLYA in the detection of fusion transcripts.

Keywords: fusion transcripts, INDEL, RNA-seq, WES, SNV

Procedia PDF Downloads 262

Authors: M. Y. Kang, Euntaek Yoon, H. D. Choi

Abstract:

To calculate the full energy (FE) absorption peak efficiency for arbitrary volume sample, we developed and verified the EXVol (Efficiency calculator for EXtended Voluminous source) code which is based on effective solid angle method. EXVol is possible to describe the source area as a non-uniform three-dimensional (x, y, z) source. And decompose and set it into several sets of volume units. Users can equally divide (x, y, z) coordinate system to calculate the detection efficiency at a specific position of a cylindrical volume source. By determining the detection efficiency for differential volume units, the total radiative absolute distribution and the correction factor of the detection efficiency can be obtained from the nondestructive measurement of the source. In order to check the performance of the EXVol code, Si ingot of 20 cm in diameter and 50 cm in height were used as a source. The detector was moved at the collimation geometry to calculate the detection efficiency at a specific position and compared with the experimental values. In this study, the performance of the EXVol code was extended to obtain the detection efficiency distribution at a specific position in a large volume source.

Keywords: attenuation, EXVol, detection efficiency, volume source

Procedia PDF Downloads 156

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 418

Authors: Yang Zhou, Kangfeng Zheng, Wei Ni, Ren Ping Liu

Abstract:

Software-defined networking (SDN) provides a solution for scalable network framework with decoupled control and data plane. However, this architecture also induces a particular distributed denial-of-service (DDoS) attack that can affect or even overwhelm the SDN network. DDoS attack detection problem has to date been mostly researched as entropy comparison problem. However, this problem lacks the utilization of SDN, and the results are not accurate. In this paper, we propose a DDoS attack detection method, which interprets DDoS detection as a signature matching problem and is formulated as Earth Mover’s Distance (EMD) model. Considering the feasibility and accuracy, we further propose to define the cost function of EMD to be a generalized Kullback-Leibler divergence. Simulation results show that our proposed method can detect DDoS attacks by comparing EMD values with the ones computed in the case without attacks. Moreover, our method can significantly increase the true positive rate of detection.

Keywords: DDoS detection, EMD, relative entropy, SDN

Procedia PDF Downloads 304

Authors: Emhimed Saffor

Abstract:

In this paper, the problem of edge detection in digital images is considered. Three methods of edge detection based on mathematical morphology algorithm were applied on two sets (Brain and Chest) CT images. 3x3 filter for first method, 5x5 filter for second method and 7x7 filter for third method under MATLAB programming environment. The results of the above-mentioned methods are subjectively evaluated. The results show these methods are more efficient and satiable for medical images, and they can be used for different other applications.

Keywords: CT images, Matlab, medical images, edge detection

Procedia PDF Downloads 303

Authors: Victoria Siriaki Jorry, I. S. Mbalawata, Hayong Shin

Abstract:

The main objective in a change detection problem is to develop algorithms for efficient detection of gradual and/or abrupt changes in the parameter distribution of a process or time series data. In this paper, we present a modified cumulative (MCUSUM) algorithm to detect the start and end of a time-varying linear drift in mean value of a time series data based on likelihood ratio test procedure. The design, implementation and performance of the proposed algorithm for a linear drift detection is evaluated and compared to the existing CUSUM algorithm using different performance measures. An approach to accurately approximate the threshold of the MCUSUM is also provided. Performance of the MCUSUM for gradual change-point detection is compared to that of standard cumulative sum (CUSUM) control chart designed for abrupt shift detection using Monte Carlo Simulations. In terms of the expected time for detection, the MCUSUM procedure is found to have a better performance than a standard CUSUM chart for detection of the gradual change in mean. The algorithm is then applied and tested to a randomly generated time series data with a gradual linear trend in mean to demonstrate its usefulness.

Keywords: average run length, CUSUM control chart, gradual change detection, likelihood ratio test

Procedia PDF Downloads 262

Authors: Bahloul Amel

Abstract:

The present study debates students’ perceptions of the use of technology in learning English as a Foreign Language. Its aim is to explore and understand students’ preparation and presentation of Posters, PowerPoint and Animated Videos by drawing attention to visual and oral elements. The data is collected through observations and semi-structured interviews and analyzed through phenomenological data analysis steps. The themes emerged from the data, visual learning satisfaction in using information and communication technology, providing structure to oral presentation, learning from peers’ presentations, draw attention to using Posters, PowerPoint and Animated Videos as each supports visual learning and organization of thoughts in oral presentations.

Keywords: EFL, posters, PowerPoint presentations, Animated Videos, visual learning

Procedia PDF Downloads 418

Authors: Min Jindong, Wang Mingxia

Abstract:

With the widespread use of artificial intelligence in life, computer vision, especially deep convolutional neural network models, has developed rapidly. With the increase of the complexity of the real visual target detection task and the improvement of the recognition accuracy, the target detection network model is also very large. The huge deep neural network model is not conducive to deployment on edge devices with limited resources, and the timeliness of network model inference is poor. In this paper, knowledge distillation is used to compress the huge and complex deep neural network model, and the knowledge contained in the complex network model is comprehensively transferred to another lightweight network model. Different from traditional knowledge distillation methods, we propose a novel knowledge distillation that incorporates multi-faceted features, called M-KD. In this paper, when training and optimizing the deep neural network model for target detection, the knowledge of the soft target output of the teacher network in knowledge distillation, the relationship between the layers of the teacher network and the feature attention map of the hidden layer of the teacher network are transferred to the student network as all knowledge. in the model. At the same time, we also introduce an intermediate transition layer, that is, an intermediate guidance layer, between the teacher network and the student network to make up for the huge difference between the teacher network and the student network. Finally, this paper adds an exploration module to the traditional knowledge distillation teacher-student network model. The student network model not only inherits the knowledge of the teacher network but also explores some new knowledge and characteristics. Comprehensive experiments in this paper using different distillation parameter configurations across multiple datasets and convolutional neural network models demonstrate that our proposed new network model achieves substantial improvements in speed and accuracy performance.

Keywords: object detection, knowledge distillation, convolutional network, model compression

Procedia PDF Downloads 248

Authors: Kaveh Shahi, Helmi Z. M. Shafri, Ebrahim Taherzadeh

Abstract:

In remote sensing, shadow causes problems in many applications such as change detection and classification. It is caused by objects which are elevated, thus can directly affect the accuracy of information. For these reasons, it is very important to detect shadows particularly in urban high spatial resolution imagery which created a significant problem. This paper focuses on automatic shadow detection based on a new spectral index for multispectral imagery known as Shadow Detection Index (SDI). The new spectral index was tested on different areas of World-View 2 images and the results demonstrated that the new spectral index has a massive potential to extract shadows effectively and automatically.

Keywords: spectral index, shadow detection, remote sensing images, World-View 2

Procedia PDF Downloads 498

Authors: Hye-Young Shin, Hae-Young Lopilly Park, Chan Kee Park

Abstract:

We compared the 24-2 and 10-2 visual fields (VFs) and investigate the degree of discrepancy between the two tests in glaucomatous eyes with central VF defects. In all, 99 eyes of 99 glaucoma patients who underwent both the 24-2 VF and 10-2 VF tests within 6 months were enrolled retrospectively. Glaucomatous eyes involving a central VF defect were divided into three groups based on the average total deviation (TD) of 12 central points in the 24-2 VF test (N = 33, in each group): group 1 (tercile with the highest TD), group 2 (intermediate TD), and group 3 (lowest TD). The TD difference was calculated by subtracting the average TD of the 10-2 VF test from the average TD of 12 central points in the 24-2 VF test. The absolute central TD difference in each quadrant was defined as the absolute value of the TD value obtained by subtracting the average TD of four central points in the 10-2 VF test from the innermost TD in the 24-2 VF test in each quadrant. The TD differences differed significantly between group 3 and groups 1 and 2 (P < 0.001). In the superonasal quadrant, the absolute central TD difference was significantly greater in group 2 than in group 1 (P < 0.05). In the superotemporal quadrant, the absolute central TD difference was significantly greater in group 3 than in groups 1 and 2 (P < 0.001). Our results indicate that the results of VF tests for different VFs can be inconsistent, depending on the degree of central defects and the VF quadrant.

Keywords: central visual field defect, glaucoma, 10-2 visual field, 24-2 visual field

Procedia PDF Downloads 147

Authors: Sergey Ermolin, Olga Ermolin

Abstract:

A Deep Convolution Neural Network with Triplet Loss is used to identify duplicate images in real-estate advertisements in the presence of image artifacts such as watermarking, cropping, hue/brightness adjustment, and others. The effects of batch normalization, spatial dropout, and various convergence methodologies on the resulting detection accuracy are discussed. For comparative Return-on-Investment study (per industry request), end-2-end performance is benchmarked on both Nvidia Titan GPUs and Intel’s Xeon CPUs. A new real-estate dataset from San Francisco Bay Area is used for this work. Sufficient duplicate detection accuracy is achieved to supplement other database-grounded methods of duplicate removal. The implemented method is used in a Proof-of-Concept project in the real-estate industry.

Keywords: visual recognition, convolutional neural networks, triplet loss, spatial batch normalization with dropout, duplicate removal, advertisement technologies, performance benchmarking

Procedia PDF Downloads 305