Engineering Commons^™

Human Depth Maps Restoration Based On Guided Gan, Jingfang Yin, Dengming Zhu, Shi Min, Zhaoqi Wang

Journal of System Simulation

Abstract: The depth maps captured by a small depth camera on mobile devices suffer from the problem of severe holes. The Guided Generative Adversarial Network (Guided GAN) based on deep learning is proposed to restore human depth maps with above problems. The high-precision human segmentation features and depth class features are extracted from the monocular RGB image by the guider based on the stacked hourglass network. The holes in the human depth maps are filled by the special generator under the guidance of the extracted human features. In order to get the more realistic results, the discriminator is introduced …

Noisy Importance Sampling Actor-Critic: An Off-Policy Actor-Critic With Experience Replay, Miriam A M Capretz, Norman Tasfi

Electrical and Computer Engineering Publications

This paper presents Noisy Importance Sampling Actor-Critic (NISAC), a set of empirically validated modifications to the advantage actor-critic algorithm (A2C), allowing off-policy reinforcement learning and increased performance. NISAC uses additive action space noise, aggressive truncation of importance sample weights, and large batch sizes. We see that additive noise drastically changes how off-sample experience is weighted for policy updates. The modified algorithm achieves an increase in convergence speed and sample efficiency compared to both the on-policy actor-critic A2C and the importance weighted off-policy actor-critic algorithm. In comparison to state-of-the-art (SOTA) methods, such as actor-critic with experience replay (ACER), NISAC nears the …

A Self Controlled Rdp Approach For Feature Extraction In Online Handwriting Recognition Using Deep Learning, Sukhdeep Singh, Vinod Kumar Chauhan, Elisa H. Barney Smith

Electrical and Computer Engineering Faculty Publications and Presentations

The identification of accurate features is the initial task for benchmarked handwriting recognition. For handwriting recognition, the objective of feature computation is to find those characteristics of a handwritten stroke that depict the class of a stroke and make it separable from the rest of the stroke classes. The present study proposes a feature extraction technique for online handwritten strokes based on a self controlled Ramer-Douglas-Peucker (RDP) algorithm. This novel approach prepares a smaller length feature vector for different shaped online handwritten strokes without preprocessing and without any control parameter to RDP. Thus, it also overcomes the shortcomings of the …

Deep Learning Based Poisson Solver In Particle Simulation Of Pn Junction With Transient Esd Excitation, Ling Zhang, Wenchang Huang, Zeyi Sun, Nicholas Erickson, Ryan From, Jun Fan

Engineering Management and Systems Engineering Faculty Research & Creative Works

In particle simulations of semiconductor devices for electro-static discharge (ESD) study at the microscopic level, solving Poisson's equation is an inevitable but time-consuming step. In this work, a deep learning technique is utilized to resolve Poisson's equation for a PN junction under an ESD event, namely using a trained deep neural network (DNN) to predict the potential distribution according to the charge distribution and the boundary condition under a transient ESD excitation. To improve the generalization performance of the DNN, multiple typical ESD curves with different parameters are used as the excitation boundary to generate large amounts of training data …

Holiday Highway Traffic Flow Prediction Method Based On Deep Learning, Xiaofeng Ji, Yicheng Ge

Journal of System Simulation

Abstract: Accurately predicting highway traffic holiday flow can provide important data for the emergency management of highway. The LSTM-SVR prediction model is established by using the theoretical framework of deep learning. The BP neural network is used to process the sample data, and the data features captured by LSTM are input into the SVR regression layer to realize the traffic flow prediction. Before and after the “Eleventh” Golden Week, the LSTM-SVR model was verified by using the traffic monitoring data of the intermodulation station in Lijiang City and the prediction results were compared with the others. It is found that …

Automatic Delamination Segmentation For Bridge Deck Based On Encoder-Decoder Deep Learning Through Uav-Based Thermography, Chongsheng Cheng, Zhexiong Shang, Zhigang Shen

Department of Construction Engineering and Management: Faculty Publications

Concrete deck delamination often demonstrates strong variations in size, shape, and temperature distribution under the influences of outdoor weather conditions. The strong variations create challenges for pure analytical solutions in infrared image segmentation of delaminated areas. The recently developed supervised deep learning approach demonstrated the potentials in achieving automatic segmentation of RGB images. However, its effectiveness in segmenting thermal images remains under-explored. The main challenge lies in the development of specific models and the generation of a large range of labeled infrared images for training. To address this challenge, a customized deep learning model based on encoder-decoder architecture is proposed …

Action Recognition Using The Motion Taxonomy, Maxat Alibayev

USF Tampa Graduate Theses and Dissertations

In the last years, modern action recognition frameworks with deep architectures have achieved impressive results on the large-scale activity datasets. All state-of-the-art models share one common attribute: two-stream architectures. One deep model takes RGB frames, while the other model is fed with pre-computed optical flow vectors. The outputs of both models are combined to be used as a final probability distribution for the action classes. When comparing the results of individual models with the fused model, it is common to see that that latter method is more superior. Researchers explain that phenomena with the fact that optical flow vectors serve …

Intelligent Sdn Traffic Classification Using Deep Learning: Deep-Sdn, Ali Malik, Ruairí De Fréin, Mohammed Al-Zeyadi, Javier Andreu-Perez

Conference papers

Accurate traffic classification is fundamentally important for various network activities such as fine-grained network management and resource utilisation. Port-based approaches, deep packet inspection and machine learning are widely used techniques to classify and analyze network traffic flows. However, over the past several years, the growth of Internet traffic has been explosive due to the greatly increased number of Internet users. Therefore, both port-based and deep packet inspection approaches have become inefficient due to the exponential growth of the Internet applications that incurs high computational cost. The emerging paradigm of software-defined networking has reshaped the network architecture by detaching the control …

Travel Time Prediction Of Urban Road Based On Deep Learning, Weiwei Zhang, Ruimin Li, Zhongjiao Xie

Journal of System Simulation

Abstract: Travel time prediction of urban road is a significant support for urban intelligent transportation system. Four types of LSTM neural network architecture were selected to predict the urban road travel time. The number of nodes in the LSTM hidden layer was fixed to determine the optimal input length of the model. The input length of the model was fixed and the predictive performance of the four LSTM models under different hidden layer nodes and considering spatial correlation were tested respectively. The performance of spatial LSTM model was compared with four traditional models, for example, BP neural network. The results …

A Cnn Based Cognitive Method To Battlefields Encompassing Situation With Insufficient Samples, Zhu Feng, Xiaofeng Hu, Xiaoyuan He, Yisi Kong, Yang Lu

Journal of System Simulation

Abstract: To research the issue of how to grasp the commander's cognitive experience successfully and effectively facing to battlefields sight map, Convolution Neural Network (CNN) as a kind of the typical algorithm in deep learning can provide the key ways. However, CNN needs the enough samples for running. These samples are hardly to achieve for the time being. Aimed at these problems, some exploring researches were carried out. The issues of battlefields encompassing situation cognition met generally in the warfare and lacking enough samples were discussed. On the basis of analyzing the image characteristics of battlefields encompassing situation and the …

Research Of Air Mission Recognition Method Based On Deep Learning, Qingkai Yao, Shaojun Liu, Xiaoyuan He, Ou Wei

Journal of System Simulation

Abstract: In the large-scale simulation of war game, the air mission is the focus of the commander's attention. The rapid, accurate and automatic recognition of air missions is the prerequisite and basis for intelligent decision making. The rapid development of deep learning technology provided a practical and feasible solution for the extraction of complex battlefield posture features, and provided technical support for studying air mission recognition. The research progress of the traditional mission recognition research method and the mission recognition method based on the deep learning was summarized. The three methods of deep learning of Convolution Neural Network (CNN), Long-short …

Deep Learning For Quantitative Motion Tracking Based On Optical Coherence Tomography, Peter Abdelmalak

Theses

Optical coherence tomography (OCT) is a cross-sectional imaging modality based on low coherence light interferometry. OCT has been widely used in diagnostic ophthalmology and has found applications in other biomedical fields such as cancer detection and surgical guidance.

In the Laboratory of Biophotonics Imaging and Sensing at New Jersey Institute of Technology, we developed a unique needle OCT imager based on a single fiber probe for breast cancer imaging. The needle OCT imager with sub-millimeter diameter can be inserted into tissue for minimally invasive in situ breast imaging. OCT imaging provides spatial resolution similar to histology and has the potential …

Modeling And Simulation Of A Robotic Bridge Inspection System, Md Monirul Karim, Cihan H. Dagli, Ruwen Qin

Engineering Management and Systems Engineering Faculty Research & Creative Works

Inspection and preservation of the aging bridges to extend their service life has been recognized as one of the important tasks of the State Departments of Transportation. Yet manual inspection procedure is not efficient to determine the safety status of the bridges in order to facilitate the implementation of appropriate maintenance. In this paper, a complex model involving a remotely controlled robotic platform is proposed to inspect the safety status of the bridges which will eliminate labor-intensive inspection. Mobile cameras from unmanned airborne vehicles (UAV) are used to collect bridge inspection data in order to record the periodic changes of …

Using Case-Level Context To Classify Cancer Pathology Reports, Shang Gao, Mohammed Alawad, Noah Schaefferkoetter, Lynne Penberthy, Xiao-Cheng Wu, Eric B. Durbin, Linda Coyle, Arvind Ramanathan, Georgia Tourassi

Kentucky Cancer Registry Faculty Publications

Individual electronic health records (EHRs) and clinical reports are often part of a larger sequence-for example, a single patient may generate multiple reports over the trajectory of a disease. In applications such as cancer pathology reports, it is necessary not only to extract information from individual reports, but also to capture aggregate information regarding the entire cancer case based off case-level context from all reports in the sequence. In this paper, we introduce a simple modular add-on for capturing case-level context that is designed to be compatible with most existing deep learning architectures for text classification on individual reports. We …

Evaluation Of Bridge Decks With Overlays Using Impact Echo, A Deep Learning Approach, Sattar Dorafshan, Hoda Azari

Civil Engineering Faculty Publications

In this paper, the feasibility of using deep learning models (DLMs) for evaluation of bridges with overlay systems is investigated. Several laboratory-made concrete specimens with artificial subsurface defects and overlay systems (bonded and debonded) made of cement and asphalt overlay materials were tested using impact echo (IE). One-dimensional (1D) and two-dimensional (2D) convolutional neural networks (CNNs) were developed, trained, and tested on the IE data. The proposed 1D CNN was the most successful in detecting debonding and subsurface defects; it achieved an average accuracy of 0.68 on the cement overlay specimens and 0.58 for asphalt overlay specimens. Maps of the …

Multiscale Modeling And Simulation Of Clutter In Isar Imaging, Jon Mitchell

Electrical Engineering Dissertations - Archive

Clutter is common in applications of radar imaging and can adversely impact target imaging by contributing scattered energy that is not accounted for in target signal models. One potential source of clutter is moving foliage in the vicinity of the target, such as a target embedded in a forest. ISAR imaging of moving clutter results in an equivalent current image that changes over each imaging sample. The stochastic nature of this clutter equivalent current presents challenges in detecting and imaging a weak embedded target using traditional algorithms. This dissertation proposes a multiscale model and analysis method to characterize the multiscale …

Stay-At-Home Motor Rehabilitation: Optimizing Spatiotemporal Learning On Low-Cost Capacitive Sensor Arrays, Reid Sutherland

Graduate Theses and Dissertations

Repeated, consistent, and precise gesture performance is a key part of recovery for stroke and other motor-impaired patients. Close professional supervision to these exercises is also essential to ensure proper neuromotor repair, which consumes a large amount of medical resources. Gesture recognition systems are emerging as stay-at-home solutions to this problem, but the best solutions are expensive, and the inexpensive solutions are not universal enough to tackle patient-to-patient variability. While many methods have been studied and implemented, the gesture recognition system designer does not have a strategy to effectively predict the right method to fit the needs of a patient. …

Computational Techniques In Medical Image Analysis Application For White Blood Cells Classification., Omar Dekhil

Electronic Theses and Dissertations

White blood cells play important rule in the human body immunity and any change in their count may cause serious diseases. In this study, a system is introduced for white blood cells localization and classification. The dataset used in this study is formed by two components, the first is the annotation dataset that will be used in the localization (364 images), and the second is labeled classes that will be used in the classification (12,444 images). For the localization, two approaches will be discussed, a classical approach and a deep learning based approach. For the classification, 5 different deep learning …

Development Of Fully Balanced Ssfp And Computer Vision Applications For Mri-Assisted Radiosurgery (Mars), Jeremiah Sanders

Dissertations and Theses (Open Access)

Prostate cancer is the second most common cancer in men and the second-leading cause of cancer death in men. Brachytherapy is a highly effective treatment option for prostate cancer, and is the most cost-effective initial treatment among all other therapeutic options for low to intermediate risk patients of prostate cancer. In low-dose-rate (LDR) brachytherapy, verifying the location of the radioactive seeds within the prostate and in relation to critical normal structures after seed implantation is essential to ensuring positive treatment outcomes.

One current gap in knowledge is how to simultaneously image the prostate, surrounding anatomy, and radioactive seeds within the …

A Framework For Vector-Weighted Deep Neural Networks, Carter Chiu

UNLV Theses, Dissertations, Professional Papers, and Capstones

The vast majority of advances in deep neural network research operate on the basis of a real-valued weight space. Recent work in alternative spaces have challenged and complemented this idea; for instance, the use of complex- or binary-valued weights have yielded promising and fascinating results. We propose a framework for a novel weight space consisting of vector values which we christen VectorNet. We first develop the theoretical foundations of our proposed approach, including formalizing the requisite theory for forward and backpropagating values in a vector-weighted layer. We also introduce the concept of expansion and aggregation functions for conversion between real …

Research On Image Description Method Based On Neural Network, Kong Rui, Xie Wei, Lei Tai

Journal of System Simulation

Abstract: The automatic recognition and automatically describing image content is an important research direction to the artificial intelligence to connect the computer vision and the natural language processing. A method of describing the image content is proposed to generate the natural language by using the deep neural network model. The model consists of a convolutional neural network (CNN) and a recurrent neural network (RNN). The CNN is used to extract features of the input image to generate a fixed-length feature vector, which initializes the RNN to generate the sentences. Experimental results on the MSCOCO image description dataset show the syntactic …

Vehicle Velocity Prediction Using Artificial Neural Networks And Effect Of Real-World Signals On Prediction Window, Tushar Dnyaneshwar Gaikwad

Masters Theses

Prediction of vehicle velocity is essential since it can realize improvements in the fuel economy/energy efficiency, drivability, and safety. Many publications address velocity prediction problems, yet there is a need for the understanding effect of different signals for the prediction. There are numerous new sensor and signal technologies like vehicle-to-vehicle and vehicle-to-infrastructure communication that can be used to obtain comprehensive datasets. Several references considered deterministic and stochastic approaches that use the datasets as input to determine future operation predictions. These approaches include different traffic models and artificial neural networks such as Markov chain, nonlinear autoregressive model, Gaussian function, and recurrent …

Multi-Level-Phase Deep Learning Using Divide-And-Conquer For Scaffolding Safety, Sayan Sakhakarmi, Jee Woong Park

Civil and Environmental Engineering and Construction Faculty Research

A traditional structural analysis of scaffolding structures requires loading conditions that are only possible during design, but not in operation. Thus, this study proposes a method that can be used during operation to make an automated safety prediction for scaffolds. It implements a divide-and-conquer technique with deep learning. As a test scaffolding, a four-bay, three-story scaffold model was used. Analysis of the model led to 1411 unique safety cases for the model. To apply deep learning, a test simulation generated 1,540,000 datasets for pre-training, and an additional 141,100 datasets for testing purposes. The cases were then sub-divided into 18 categories …

Object Detection With Deep Learning To Accelerate Pose Estimation For Automated Aerial Refueling, Andrew T. Lee

Theses and Dissertations

Remotely piloted aircraft (RPAs) cannot currently refuel during flight because the latency between the pilot and the aircraft is too great to safely perform aerial refueling maneuvers. However, an AAR system removes this limitation by allowing the tanker to directly control the RP A. The tanker quickly finding the relative position and orientation (pose) of the approaching aircraft is the first step to create an AAR system. Previous work at AFIT demonstrates that stereo camera systems provide robust pose estimation capability. This thesis first extends that work by examining the effects of the cameras' resolution on the quality of pose …

Learning In The Machine: To Share Or Not To Share?, Jordan Ott, Erik Linstead, Nicholas Lahaye, Pierre Baldi

Engineering Faculty Articles and Research

Weight-sharing is one of the pillars behind Convolutional Neural Networks and their successes. However, in physical neural systems such as the brain, weight-sharing is implausible. This discrepancy raises the fundamental question of whether weight-sharing is necessary. If so, to which degree of precision? If not, what are the alternatives? The goal of this study is to investigate these questions, primarily through simulations where the weight-sharing assumption is relaxed. Taking inspiration from neural circuitry, we explore the use of Free Convolutional Networks and neurons with variable connection patterns. Using Free Convolutional Networks, we show that while weight-sharing is a pragmatic optimization …

Knowledge Infused Learning (K-Il): Towards Deep Incorporation Of Knowledge In Deep Learning, Ugur Kursuncu, Manas Gaur, Amit Sheth

Publications

Learning the underlying patterns in data goes beyondinstance-based generalization to external knowledge repre-sented in structured graphs or networks. Deep learning thatprimarily constitutes neural computing stream in AI hasshown significant advances in probabilistically learning la-tent patterns using a multi-layered network of computationalnodes (i.e., neurons/hidden units). Structured knowledge thatunderlies symbolic computing approaches and often supportsreasoning, has also seen significant growth in recent years,in the form of broad-based (e.g., DBPedia, Yago) and do-main, industry or application specific knowledge graphs. Acommon substrate with careful integration of the two willraise opportunities to develop neuro-symbolic learning ap-proaches for AI, where conceptual and probabilistic repre-sentations are combined. …

Analysis Of Feature Extraction In Knee Cartilage Semantic Segmentation Convolutional Neural Networks, Logan Thorneloe

Undergraduate Honors Theses

Recent advances in deep learning and convolutional neural networks (CNNs) have shown promise for automatic segmentation in magnetic resonance images. However, because of the stochastic nature of the training process, it is difficult to interpret what information networks learn to represent. This study explores multiple difference metrics between networks to determine semantic relationships between knee cartilage tissues. It explores how differences in learned weights and output activations between networks can be used to express these relationships. These findings are further supported by training multi-class networks to segment multiple tissues to compare network accuracy across different tissue combinations. This study shows …

Flood Prediction And Uncertainty Estimation Using Deep Learning, Vinayaka Gude, Steven Corns, Suzanna Long

Engineering Management and Systems Engineering Faculty Research & Creative Works

Floods are a complex phenomenon that are difficult to predict because of their non-linear and dynamic nature. Therefore, flood prediction has been a key research topic in the field of hydrology. Various researchers have approached this problem using different techniques ranging from physical models to image processing, but the accuracy and time steps are not sufficient for all applications. This study explores deep learning techniques for predicting gauge height and evaluating the associated uncertainty. Gauge height data for the Meramec River in Valley Park, Missouri was used to develop and validate the model. It was found that the deep learning …

Improving Aeromagnetic Calibration Using Artificial Neural Networks, Mitchell C. Hezel

Theses and Dissertations

The Global Positioning System (GPS) has proven itself to be the single most accurate positioning system available, and no navigation suite is found without a GPS receiver. Even basic GPS receivers found in most smartphones can easily provide high quality positioning information at any time. Even with its superb performance, GPS is prone to jamming and spoofing, and many platforms requiring accurate positioning information are in dire need of other navigation solutions to compensate in the event of an outage, be the cause hostile or natural. Indeed, there has been a large push to achieve an alternative navigation capability which …

Improved Ground-Based Monocular Visual Odometry Estimation Using Inertially-Aided Convolutional Neural Networks, Josiah D. Watson

Theses and Dissertations

While Convolutional Neural Networks (CNNs) can estimate frame-to-frame (F2F) motion even with monocular images, additional inputs can improve Visual Odometry (VO) predictions. In this thesis, a FlowNetS-based [1] CNN architecture estimates VO using sequential images from the KITTI Odometry dataset [2]. For each of three output types (full six degrees of freedom (6-DoF), Cartesian translation, and transitional scale), a baseline network with only image pair input is compared with a nearly identical architecture that is also given an additional rotation estimate such as from an Inertial Navigation System (INS). The inertially-aided networks show an order of magnitude improvement over the …