Digital Commons Network^™

Star-Based Reachability Analysis Of Binary Neural Networks On Continuous Input, Mykhailo Ivashchenko

Department of Computer Science and Engineering: Dissertations, Theses, and Student Research

Deep Neural Networks (DNNs) have become a popular instrument for solving various real-world problems. DNNs’ sophisticated structure allows them to learn complex representations and features. However, architecture specifics and floating-point number usage result in increased computational operations complexity. For this reason, a more lightweight type of neural networks is widely used when it comes to edge devices, such as microcomputers or microcontrollers – Binary Neural Networks (BNNs). Like other DNNs, BNNs are vulnerable to adversarial attacks; even a small perturbation to the input set may lead to an errant output. Unfortunately, only a few approaches have been proposed for verifying …

Rescape: Transforming Coral-Reefscape Images For Quantitative Analysis, Zachary Ferris, Eraldo Ribeiro, Tomofumi Nagata, Robert Van Woesik

Ocean Engineering and Marine Sciences Faculty Publications

Ever since the first image of a coral reef was captured in 1885, people worldwide have been accumulating images of coral reefscapes that document the historic conditions of reefs. However, these innumerable reefscape images suffer from perspective distortion, which reduces the apparent size of distant taxa, rendering the images unusable for quantitative analysis of reef conditions. Here we solve this century-long distortion problem by developing a novel computer-vision algorithm, ReScape, which removes the perspective distortion from reefscape images by transforming them into top-down views, making them usable for quantitative analysis of reef conditions. In doing so, we demonstrate the …

Directed Acyclic Graph-Based Neural Networks For Tunable Low-Power Computer Vision, Abhinav Goel, Caleb Tung, Nick Eliopoulos, Xiao Hu, George K. Thiruvathukal, James C. Davis, Yung-Hisang Lu

Computer Science: Faculty Publications and Other Works

Processing visual data on mobile devices has many applications, e.g., emergency response and tracking. State-of-the-art computer vision techniques rely on large Deep Neural Networks (DNNs) that are usually too power-hungry to be deployed on resource-constrained edge devices. Many techniques improve DNN efficiency of DNNs by compromising accuracy. However, the accuracy and efficiency of these techniques cannot be adapted for diverse edge applications with different hardware constraints and accuracy requirements. This paper demonstrates that a recent, efficient tree-based DNN architecture, called the hierarchical DNN, can be converted into a Directed Acyclic Graph-based (DAG) architecture to provide tunable accuracy-efficiency tradeoff options. We …

Gauging The State-Of-The-Art For Foresight Weight Pruning On Neural Networks, Noah James

Computer Science and Computer Engineering Undergraduate Honors Theses

The state-of-the-art for pruning neural networks is ambiguous due to poor experimental practices in the field. Newly developed approaches rarely compare to each other, and when they do, their comparisons are lackluster or contain errors. In the interest of stabilizing the field of pruning, this paper initiates a dive into reproducing prominent pruning algorithms across several architectures and datasets. As a first step towards this goal, this paper shows results for foresight weight pruning across 6 baseline pruning strategies, 5 modern pruning strategies, random pruning, and one legacy method (Optimal Brain Damage). All strategies are evaluated on 3 different architectures …

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 …

Learning To Map The Visual And Auditory World, Tawfiq Salem

Theses and Dissertations--Computer Science

The appearance of the world varies dramatically not only from place to place but also from hour to hour and month to month. Billions of images that capture this complex relationship are uploaded to social-media websites every day and often are associated with precise time and location metadata. This rich source of data can be beneficial to improve our understanding of the globe. In this work, we propose a general framework that uses these publicly available images for constructing dense maps of different ground-level attributes from overhead imagery. In particular, we use well-defined probabilistic models and a weakly-supervised, multi-task training …

Integrity Monitoring For Automated Aerial Refueling: A Stereo Vision Approach, Thomas R. Stuart

Theses and Dissertations

Unmanned aerial vehicles (UAVs) increasingly require the capability to y autonomously in close formation including to facilitate automated aerial refueling (AAR). The availability of relative navigation measurements and navigation integrity are essential to autonomous relative navigation. Due to the potential non-availability of the global positioning system (GPS) during military operations, it is highly desirable that relative navigation can be accomplished without the use of GPS. This paper develops two algorithms designed to provide relative navigation measurements solely from a stereo image pair. These algorithms were developed and analyzed in the context of AAR using a stereo camera system modeling that …

Interpreting Individual Classifications Of Hierarchical Networks, Will Landecker, Michael David Thomure, Luis M.A. Bettencourt, Melanie Mitchell, Garrett T. Kenyon, Steven P. Brumby

Computer Science Faculty Publications and Presentations

Hierarchical networks are known to achieve high classification accuracy on difficult machine-learning tasks. For many applications, a clear explanation of why the data was classified a certain way is just as important as the classification itself. However, the complexity of hierarchical networks makes them ill-suited for existing explanation methods. We propose a new method, contribution propagation, that gives per-instance explanations of a trained network's classifications. We give theoretical foundations for the proposed method, and evaluate its correctness empirically. Finally, we use the resulting explanations to reveal unexpected behavior of networks that achieve high accuracy on visual object-recognition tasks using well-known …

Toward Automatic Subpixel Registration Of Unmanned Airborne Vehicle Images, Amr Hussein Yousef, Jiang Li, Mohammad Karim, Mark Allen Neifeld (Ed.), Amit Ashok (Ed.)

Electrical & Computer Engineering Faculty Publications

Many applications require to register images within subpixel accuracy like computer vision especially super-resolution (SR) where the estimated subpixel shifts are very crucial in the reconstruction and restoration of SR images. In our work we have an optical sensor that is mounted on an unmanned airborne vehicle (UAV) and captures a set of images that contain sufficient overlapped area required to reconstruct a SR image. Due to the wind, The UAV may encounter rotational effects such as yaw, pitch and roll which can distort the acquired as well as processed images with shear, tilt or perspective distortions. In this paper …

Enhancement Technique For Aerial Images, Sertan Erkanli, Ahmet Gungor Pakfiliz, Jiang Li

Electrical & Computer Engineering Faculty Publications

Recently, we proposed an enhancement technique for uniformly and non-uniformly illuminated dark images that provides high color accuracy and better balance between the luminance and the contrast in images to improve the visual representations of digital images. In this paper we define an improved version of the proposed algorithm to enhance aerial images in order to reduce the gap between direct observation of a scene and its recorded image.

Object Detection And Classification With Applications To Skin Cancer Screening, Jonathan Blackledge, Dmitryi Dubovitskiy

Articles

This paper discusses a new approach to the processes of object detection, recognition and classification in a digital image. The classification method is based on the application of a set of features which include fractal parameters such as the Lacunarity and Fractal Dimension. Thus, the approach used, incorporates the characterisation of an object in terms of its texture.

The principal issues associated with object recognition are presented which includes two novel fast segmentation algorithms for which C++ code is provided. The self-learning procedure for designing a decision making engine using fuzzy logic and membership function theory is also presented and …

Vegetation Identification Based On Satellite Imagery, Vamsi K.R. Mantena, Ramu Pedada, Srinivas Jakkula, Yuzhong Shen, Jiang Li, Hamid R. Arabnia (Ed.)

Electrical & Computer Engineering Faculty Publications

Automatic vegetation identification plays an important role in many applications including remote sensing and high performance flight simulations. This paper presents a method to automatically identify vegetation based upon satellite imagery. First, we utilize the ISODATA algorithm to cluster pixels in the images where the number of clusters is determined by the algorithm. We then apply morphological operations to the clustered images to smooth the boundaries between clusters and to fill holes inside clusters. After that, we compute six features for each cluster. These six features then go through a feature selection algorithm and three of them are determined to …

Determination Of Structure From Motion Using Aerial Imagery, Paul R. Graham

Theses and Dissertations

The structure from motion process creates three-dimensional models from a sequence of images. Until recently, most research in this field has been restricted to land-based imagery. This research examines the current methods of land-based structure from motion and evaluates their performance for aerial imagery. Current structure from motion algorithms search the initial image for features to track though the subsequent images. These features are used to create point correspondences between the two images. The correspondences are used to estimate the motion of the camera and then the three-dimensional structure of the scene. This research tests current algorithms using synthetic data …