Physical Sciences and Mathematics Commons^™

Machine Learning And Computer Vision In Solar Physics, Haodi Jiang

Dissertations

In the recent decades, the difficult task of understanding and predicting violent solar eruptions and their terrestrial impacts has become a strategic national priority, as it affects the life of human beings, including communication, transportation, the power grid, national defense, space travel, and more. This dissertation explores new machine learning and computer vision techniques to tackle this difficult task. Specifically, the dissertation addresses four interrelated problems in solar physics: magnetic flux tracking, fibril tracing, Stokes inversion and vector magnetogram generation.

First, the dissertation presents a new deep learning method, named SolarUnet, to identify and track solar magnetic flux elements in …

Ow-Detr: Open-World Detection Transformer, Akshita Gupta, Sanath Narayan, K.J. Joseph, Salman Khan, Fahad Shahbaz Khan, Mubarak Shah

Computer Vision Faculty Publications

Open-world object detection (OWOD) is a challenging computer vision problem, where the task is to detect a known set of object categories while simultaneously identifying unknown objects. Additionally, the model must incrementally learn new classes that become known in the next training episodes. Distinct from standard object detection, the OWOD setting poses significant challenges for generating quality candidate proposals on potentially unknown objects, separating the unknown objects from the background and detecting diverse unknown objects. Here, we introduce a novel end-to-end transformer-based framework, OW-DETR, for open-world object detection. The proposed OW-DETR comprises three dedicated components namely, attention-driven pseudo-labeling, novelty classification …

Auto-Curation Of Large Evolving Image Datasets, Sara Mousavicheshmehkaboodi

Doctoral Dissertations

Large image collections are becoming common in many fields and offer tantalizing opportunities to transform how research, work, and education are conducted if the information and associated insights could be extracted from them. However, major obstacles to this vision exist. First, image datasets with associated metadata contain errors and need to be cleaned and organized to be easily explored and utilized. Second, such collections typically lack the necessary context or may have missing attributes that need to be recovered. Third, such datasets are domain-specific and require human expert involvement to make the right interpretation of the image content. Fourth, the …

Situate: An Agent-Based System For Situation Recognition, Max Henry Quinn

Dissertations and Theses

Computer vision and machine learning systems have improved significantly in recent years, largely based on the development of deep learning systems, leading to impressive performance on object detection tasks. Understanding the content of images is considerably more difficult. Even simple situations, such as "a handshake", "walking the dog", "a game of ping-pong", or "people waiting for a bus", present significant challenges. Each consists of common objects, but are not reliably detectable as a single entity nor through the simple co-occurrence of their parts.

In this dissertation, toward the goal of developing machine learning systems that demonstrate properties associated with understanding, …

Fingerlings Mass Estimation: A Comparison Between Deep And Shallow Learning Algorithms, Adair Da Silva Oliveira Junior, Diego André Sant’Ana, Marcio Carneiro Brito Pache, Vanir Garcia, Vanessa Aparecida De Moares Weber, Gilberto Astolfi, Fabricio De Lima Weber, Geazy Vilharva Menezes, Gabriel Kirsten Menezes, Pedro Lucas França Albuquerque, Celso Soares Costa, Eduardo Quirino Arguelho De Queiroz, João Victor Araújo Rozales, Milena Wolff Ferreira, Marco Hiroshi Naka, Hemerson Pistori

School of Computing: Faculty Publications

The paper presents some results regarding the automatic mass estimation of Pintado Real fingerlings, using machine learning techniques to support the fish production process. For this purpose, an image dataset called FISHCV1206FSEG, was created which is composed of 1206 images of fingerlings with their respective annotated masses. Through the fish contours, the area and perimeter were extracted, and submitted to the J48, SVM, and KNN classification algorithms and a linear regression algorithm. The images were also submitted to ResNet50, In- ceptionV3, Exception, VGG16, and VGG19 convolutional neural networks. As a result, the classification algorithm J48 reached an accuracy of 58.2% …

Novel Statistical Modeling Methods For Traffic Video Analysis, Hang Shi

Dissertations

Video analysis is an active and rapidly expanding research area in computer vision and artificial intelligence due to its broad applications in modern society. Many methods have been proposed to analyze the videos, but many challenging factors remain untackled. In this dissertation, four statistical modeling methods are proposed to address some challenging traffic video analysis problems under adverse illumination and weather conditions.

First, a new foreground detection method is presented to detect the foreground objects in videos. A novel Global Foreground Modeling (GFM) method, which estimates a global probability density function for the foreground and applies the Bayes decision rule …

Advances In Deep Learning With Applications To Computer Vision And Astronomy, Zhihang Hu

Dissertations

Deep Learning has spanned a variety of applications in computer vision as well as computational astronomy. These two aspects obtained similar data structure, therefore, their solutions can be transferable between each other. This dissertation look into two video-related tasks in computer vision and propose a novel problem in computational astronomy.

Specifically, acquiring an in-depth understanding of videos has been a cornerstone problem in computer vision. This problem has been studied by various researchers from different perspectives, among which video prediction has attracted much attention. Video prediction aims to generate the pixels of future frames given a sequence of context frames. …

Computer Vision Applications For Autonomous Aerial Vehicles, Burak Kakillioglu

Dissertations - ALL

Undoubtedly, unmanned aerial vehicles (UAVs) have experienced a great leap forward over the last decade. It is not surprising anymore to see a UAV being used to accomplish a certain task, which was previously carried out by humans or a former technology. The proliferation of special vision sensors, such as depth cameras, lidar sensors and thermal cameras, and major breakthroughs in computer vision and machine learning fields accelerated the advance of UAV research and technology. However, due to certain unique challenges imposed by UAVs, such as limited payload capacity, unreliable communication link with the ground stations and data safety, UAVs …

Understanding Complex Human Activities In Videos : The Study Of Concurrent Activity Detection And Group Activity Recognition, Yi Wei

Legacy Theses & Dissertations (2009 - 2024)

Human activity understanding, as one of the most important task in video analysis, has been studied for decades. Great efforts have been made to push the activity recognition models towards effective and efficient representation learning. However, it is difficult to define an explicit semantic organization of activities, even for human. Current activity recognition benchmarks only organize the activity labels with shallow hierarchies, which hinders the development of activity recognition system.

Mining Urban Perceptions From Social Media Data, Yu Liu, Yihong Yuan, Fan Zhang

Journal of Spatial Information Science

This vision paper summaries the methods of using social media data (SMD) to measure urban perceptions. We highlight two major types of data sources (i.e., texts and imagery) and two corresponding techniques (i.e., natural language processing and computer vision). Recognizing the data quality issues of SMD, we propose three criteria for improving the reliability of SMD-based studies. In addition, integrating multi-source data is a promising approach to mitigating the data quality problems.

Material Detection With Thermal Imaging And Computer Vision: Potentials And Limitations, Jared Poe

Graduate Theses and Dissertations

The goal of my masters thesis research is to develop an affordable and mobile infraredbased environmental sensoring system for the control of a servo motor based on material identification. While this sensing could be oriented towards different applications, my thesis is particularly interested in material detection due to the wide range of possible applications in mechanical engineering. Material detection using a thermal mobile camera could be used in manufacturing, recycling or autonomous robotics. For my research, the application that will be focused on is using this material detection to control a servo motor by identifying and sending control inputs based …

Methods For Detecting Floodwater On Roadways From Ground Level Images, Cem Sazara

Computational Modeling & Simulation Engineering Theses & Dissertations

Recent research and statistics show that the frequency of flooding in the world has been increasing and impacting flood-prone communities severely. This natural disaster causes significant damages to human life and properties, inundates roads, overwhelms drainage systems, and disrupts essential services and economic activities. The focus of this dissertation is to use machine learning methods to automatically detect floodwater in images from ground level in support of the frequently impacted communities. The ground level images can be retrieved from multiple sources, including the ones that are taken by mobile phone cameras as communities record the state of their flooded streets. …

Pedestrian Attribute Recognition Using Trainable Gabor Wavelets, Imran N Junejo, Naveed Ahmed, Mohammad Lataifeh

All Works

Surveillance cameras are everywhere keeping an eye on pedestrians or people as they navigate through the scene. Within this context, our paper addresses the problem of pedestrian attribute recognition (PAR). This problem entails the extraction of different attributes such as age-group, clothing style, accessories, footwear style etc. This is a multi-label problem with a host of challenges even for human observers. As such, the topic has rightly attracted attention recently. In this work, we integrate trainable Gabor wavelet (TGW) layers inside a convolution neural network (CNN). Whereas other researchers have used fixed Gabor filters with the CNN, the proposed layers …

A Quantitative Validation Of Multi-Modal Image Fusion And Segmentation For Object Detection And Tracking, Nicholas Lahaye, Michael J. Garay, Brian D. Bue, Hesham El-Askary, Erik Linstead

Mathematics, Physics, and Computer Science Faculty Articles and Research

In previous works, we have shown the efficacy of using Deep Belief Networks, paired with clustering, to identify distinct classes of objects within remotely sensed data via cluster analysis and qualitative analysis of the output data in comparison with reference data. In this paper, we quantitatively validate the methodology against datasets currently being generated and used within the remote sensing community, as well as show the capabilities and benefits of the data fusion methodologies used. The experiments run take the output of our unsupervised fusion and segmentation methodology and map them to various labeled datasets at different levels of global …

Reciprocal Transformations For Unsupervised Video Object Segmentation, Sucheng Ren, Wenxi Liu, Yongtuo Liu, Haoxin Chen, Guoqiang Han, Shengfeng He

Research Collection School Of Computing and Information Systems

Unsupervised video object segmentation (UVOS) aims at segmenting the primary objects in videos without any human intervention. Due to the lack of prior knowledge about the primary objects, identifying them from videos is the major challenge of UVOS. Previous methods often regard the moving objects as primary ones and rely on optical flow to capture the motion cues in videos, but the flow information alone is insufficient to distinguish the primary objects from the background objects that move together. This is because, when the noisy motion features are combined with the appearance features, the localization of the primary objects is …

Counterfactual Zero-Shot And Open-Set Visual Recognition, Zhongqi Yue, Tan Wang, Qianru Sun, Xian-Sheng Hua, Hanwang Zhang

Research Collection School Of Computing and Information Systems

We present a novel counterfactual framework for both Zero-Shot Learning (ZSL) and Open-Set Recognition (OSR), whose common challenge is generalizing to the unseen-classes by only training on the seen-classes. Our idea stems from the observation that the generated samples for unseen-classes are often out of the true distribution, which causes severe recognition rate imbalance between the seen-class (high) and unseen-class (low). We show that the key reason is that the generation is not Counterfactual Faithful, and thus we propose a faithful one, whose generation is from the sample-specific counterfactual question: What would the sample look like, if we set its …

Projecting Your View Attentively: Monocular Road Scene Layout Estimation Via Cross-View Transformation, Weixiang Yang, Qi Li, Wenxi Liu, Yuanlong Yu, Yuexin Ma, Shengfeng He, Jia Pan

Research Collection School Of Computing and Information Systems

HD map reconstruction is crucial for autonomous driving. LiDAR-based methods are limited due to the deployed expensive sensors and time-consuming computation. Camera-based methods usually need to separately perform road segmentation and view transformation, which often causes distortion and the absence of content. To push the limits of the technology, we present a novel framework that enables reconstructing a local map formed by road layout and vehicle occupancy in the bird's-eye view given a front-view monocular image only. In particular, we propose a cross-view transformation module, which takes the constraint of cycle consistency between views into account and makes full use …

Adaptive Aggregation Networks For Class-Incremental Learning, Yaoyao Liu, Bernt Schiele, Qianru Sun

Research Collection School Of Computing and Information Systems

Class-Incremental Learning (CIL) aims to learn a classification model with the number of classes increasing phase-by-phase. An inherent problem in CIL is the stability-plasticity dilemma between the learning of old and new classes, i.e., high-plasticity models easily forget old classes, but high-stability models are weak to learn new classes. We alleviate this issue by proposing a novel network architecture called Adaptive Aggregation Networks (AANets) in which we explicitly build two types of residual blocks at each residual level (taking ResNet as the baseline architecture): a stable block and a plastic block. We aggregate the output feature maps from these two …

Rm-Net: Rasterizing Markov Signals To Images For Deep Learning, Kajal Gupta

Theses

Statistical machine learning approaches are quite famous for processing Markov signal data. They can model unobserved states and learn certain characteristics particular to a signal with good accuracy. However, with the advent of Deep learning the novice ways of solving a problem has shifted towards this more sophisticated algorithm, which is much better, powerful and more accurate. Specifically, Convolutional Neural Nets (CNN) have shown many promising results on images and videos. Here we illustrate how CNN can be applied to a 1D numeric signal using signal rasterization technique. We start by rasterizing a 1D numeric Markov signal into an image …

Towards Open World Object Detection, K. J. Joseph, Salman Khan, Fahad Shahbaz Khan, Vineeth N. Balasubramanian

Computer Vision Faculty Publications

Humans have a natural instinct to identify unknown object instances in their environments. The intrinsic curiosity about these unknown instances aids in learning about them, when the corresponding knowledge is eventually available. This motivates us to propose a novel computer vision problem called: 'Open World Object Detection', where a model is tasked to: 1) identify objects that have not been introduced to it as 'unknown', without explicit supervision to do so, and 2) incrementally learn these identified unknown categories without forgetting previously learned classes, when the corresponding labels are progressively received. We formulate the problem, introduce a strong evaluation protocol …

Using Deep Learning To Analyze Materials In Medical Images, Carson Molder

Computer Science and Computer Engineering Undergraduate Honors Theses

Modern deep learning architectures have become increasingly popular in medicine, especially for analyzing medical images. In some medical applications, deep learning image analysis models have been more accurate at predicting medical conditions than experts. Deep learning has also been effective for material analysis on photographs. We aim to leverage deep learning to perform material analysis on medical images. Because material datasets for medicine are scarce, we first introduce a texture dataset generation algorithm that automatically samples desired textures from annotated or unannotated medical images. Second, we use a novel Siamese neural network called D-CNN to predict patch similarity and build …

Regularized Deep Network Learning For Multi-Label Visual Recognition, Hao Guo

Theses and Dissertations

This dissertation is focused on the task of multi-label visual recognition, a fundamental task of computer vision. It aims to tell the presence of multiple visual classes from the input image, where the visual classes, such as objects, scenes, attributes, etc., are usually defined as image labels. Due to the prosperous deep networks, this task has been widely studied and significantly improved in recent years. However, it remains a challenging task due to appearance complexity of multiple visual contents co-occurring in one image. This research explores to regularize the deep network learning for multi-label visual recognition.

First, an attention concentration …

Automatic Detection Of Vehicles In Satellite Images For Economic Monitoring, Cole Hill

USF Tampa Graduate Theses and Dissertations

With the growing supply of satellites capturing images of the planet, governments andinvestors are looking for ways in which these new images may be used to determine which businesses are struggling and thriving. Recent works have shown that parking lot fill rates can provide valuable information about businesses’ earnings, however, the task of manually annotating the number of vehicles in a parking lot is expensive and time-consuming. Systems which can automate this process are therefore valuable as they are faster and cheaper than human labor. In this thesis, the problem of detection of small objects in large low-resolution images is …

Accurate Covariance Estimation For Pose Data From Iterative Closest Point Algorithm, Rick H. Yuan

Theses and Dissertations

One of the fundamental problems of robotics and navigation is the estimation of relative pose of an external object with respect to the observer. A common method for computing the relative pose is the Iterative Closest Point (ICP) algorithm, where a reference point cloud of a known object is registered against a sensed point cloud to determine relative pose. To use this computed pose information in down-stream processing algorithms, it is necessary to estimate the uncertainty of the ICP output, typically represented as a covariance matrix. In this thesis a novel method for estimating uncertainty from sensed data is introduced. …

Stereo Camera Calibrations With Optical Flow, Joshua D. Larson

Theses and Dissertations

Remotely Piloted Aircraft (RPA) are currently unable to refuel mid-air due to the large communication delays between their operators and the aircraft. AAR seeks to address this problem by reducing the communication delay to a fast line-of-sight signal between the tanker and the RPA. Current proposals for AAR utilize stereo cameras to estimate where the receiving aircraft is relative to the tanker, but require accurate calibrations for accurate location estimates of the receiver. This paper improves the accuracy of this calibration by improving three components of it: increasing the quantity of intrinsic calibration data with CNN preprocessing, improving the quality …

Multi-Modal Classification Using Images And Text, Stuart J. Miller, Justin Howard, Paul Adams, Mel Schwan, Robert Slater

SMU Data Science Review

This paper proposes a method for the integration of natural language understanding in image classification to improve classification accuracy by making use of associated metadata. Traditionally, only image features have been used in the classification process; however, metadata accompanies images from many sources. This study implemented a multi-modal image classification model that combines convolutional methods with natural language understanding of descriptions, titles, and tags to improve image classification. The novelty of this approach was to learn from additional external features associated with the images using natural language understanding with transfer learning. It was found that the combination of ResNet-50 image …

Ship Deck Segmentation In Engineering Document Using Generative Adversarial Networks, Mohammad Shahab Uddin, Raphael Pamie-George, Daron Wilkins, Andres Sousa Poza, Mustafa Canan, Samuel Kovacic, Jiang Li

Engineering Management & Systems Engineering Faculty Publications

Generative adversarial networks (GANs) have become very popular in recent years. GANs have proved to be successful in different computer vision tasks including image-translation, image super-resolution etc. In this paper, we have used GAN models for ship deck segmentation. We have used 2D scanned raster images of ship decks provided by US Navy Military Sealift Command (MSC) to extract necessary information including ship walls, objects etc. Our segmentation results will be helpful to get vector and 3D image of a ship that can be later used for maintenance of the ship. We applied the trained models to engineering documents provided …

A Deep Transfer Learning Based Model For Automatic Detection Of Covid-19from Chest X-Rays, Prateek Chhikara, Prakhar Gupta, Prabhjot Singh, Tarunpreet Bhatia

Turkish Journal of Electrical Engineering and Computer Sciences

Deep learning in medical imaging has revolutionized the way we interpret medical data, as high computational devices' capabilities are far more than their creators. With the pandemic causing havoc for the second straight year, the findings in our paper will allow researchers worldwide to use and create state-of-the-art models to detect affected persons before it reaches the R number. The paper proposes an automated diagnostic tool using the deep learning models on chest x-rays as an input to reach a point where we surpass this pandemic (COVID-19 disease). A deep transfer learning-based model for automatic detection of COVID-19 from chest …

Inference Of Surface Velocities From Oblique Time Lapse Photos And Terrestrial Based Lidar At The Helheim Glacier, Franklyn T. Dunbar Ii

Graduate Student Theses, Dissertations, & Professional Papers

Using time dependent observations derived from terrestrial LiDAR and oblique
time-lapse imagery, we demonstrate that a Bayesian approach to glacial motion es-
timation provides a concise way to incorporate multiple data products into a single
motion estimation procedure effectively producing surface velocity estimates with
an associated uncertainty. This approach brings both improved computational effi-
ciency, and greater scalability across observational time-frames when compared to
existing methods. To gauge efficacy, we apply these methods to a set of observa-
tions from the Helheim Glacier, a critical actor in contemporary mass loss trends
observed in the Greenland Ice Sheet. We find that …

Deep Unsupervised Anomaly Detection, Tangqing Li, Zheng Wang, Siying Liu, Wen-Yan Lin

Research Collection School Of Computing and Information Systems

This paper proposes a novel method to detect anomalies in large datasets under a fully unsupervised setting. The key idea behind our algorithm is to learn the representation underlying normal data. To this end, we leverage the latest clustering technique suitable for handling high dimensional data. This hypothesis provides a reliable starting point for normal data selection. We train an autoencoder from the normal data subset, and iterate between hypothesizing normal candidate subset based on clustering and representation learning. The reconstruction error from the learned autoencoder serves as a scoring function to assess the normality of the data. Experimental results …