Browsing by Subject "Convolutional Neural Network"

Now showing 1 - 5 of 5
  • Thumbnail Image
    Item
    The Challenges of Detecting Eurasian Watermilfoil with a Pseudo Labeling Semi-Supervised Convolutional Neural Network
    (2022-05-02) Pargman, Connor
    Eurasian Watermilfoil is an invasive aquatic plant found in many bodies of water in Minnesota. It tends to out grow and kill many native plants. The current solution to removing Eurasian Watermilfoil is to kill it using a herbicide. However, this has drawbacks because the herbicide can affect native plants, it contaminates the water, and is not sprayed accurately. A solution to this problem is by using autonomous underwater vehicles equipped with a deep learning model that can detect Eurasian Watermilfoil to map it for accurate spraying. However we found this not to be the case. While trying to train a model to detect Eurasian Watermilfoil using a pseudo labeling semi-supervised and supervised convolutional neural network, it could not detect the plant due to the scarce amount of images. However it was found the pseudo labeling a diver dataset proved to be more accurate and efficent than the supervised version.
  • Thumbnail Image
    Item
    Muon Neutrino Disappearance in NOvA with a Deep Convolutional Neural Network Classifier
    (2016-03) Rocco, Dominick
    The NuMI Off-axis Neutrino Appearance Experiment (NOvA) is designed to study neutrino oscillation in the NuMI (Neutrinos at the Main Injector) beam. NOvA observes neutrino oscillation using two detectors separated by a baseline of 810 km; a 14 kt Far Detector in Ash River, MN and a functionally identical 0.3 kt Near Detector at Fermilab. The experiment aims to provide new measurements of $\Delta m^2_{32}$ and $\theta_{23}$ and has potential to determine the neutrino mass hierarchy as well as observe CP violation in the neutrino sector. Essential to these analyses is the classification of neutrino interaction events in NOvA detectors. Raw detector output from NOvA is interpretable as a pair of images which provide orthogonal views of particle interactions. A recent advance in the field of computer vision is the advent of convolutional neural networks, which have delivered top results in the latest image recognition contests. This work presents an approach novel to particle physics analysis in which a convolutional neural network is used for classification of particle interactions. The approach has been demonstrated to improve the signal efficiency and purity of the event selection, and thus physics sensitivity. Early NOvA data has been analyzed (2.74$\times10^{20}$ POT, 14 kt equivalent) to provide new best-fit measurements of $\sin^2(\theta_{23}) = 0.43$ (with a statistically-degenerate compliment near 0.60) and $|\Delta m^2_{32}| = 2.48\times10^{-3}~\text{eV}^2$.
  • Thumbnail Image
    Item
    Noninvasive Cardiac Electrical Imaging of Activation Sequence and Activation Recovery Interval, and Localization of Ventricular Arrhythmias
    (2018-01) Yang, Ting
    This dissertation research aims to develop and evaluate methods for noninvasive cardiac imaging of activation sequence and activation recovery interval (ARI), and localization of ventricular arrhythmias. It includes (1) developing a novel imaging method (SSF, Spatial gradient Sparse in Frequency domain) for the reconstruction of activation sequences in ventricular arrhythmias, (2) developing ARI imaging technique to reconstruct the ARI maps in premature ventricular contraction (PVC) patients from body surface potential maps, (3) proposing a CNN-based (convolutional-neural-network-based) method to localize origins of PVCs from 12-lead electrocardiography (ECG). SSF is implemented in the frequency domain, and the activation time was encoded in the phase information of the solution. The performance of SSF was evaluated in computer simulation and a swine model with myocardial infarction. SSF is the first noninvasive imaging method reported that could reconstruct the reentry circuit in 3-dimensional space. SSF achieved better performance with less computational time. ARI imaging reconstructed 3D ARI maps in ventricles, which were compared with the endocardial ARI maps from CARTO recordings. From the analysis of 100 PVC beats in ten patients, the results suggest that it could serve as an alternative of evaluating global dispersion of ventricular repolarization and could guide ablation procedure in PVC patients. The CNN-based method consists of two CNNs (Segment CNN and Epi-Endo CNN). The inputs are from 12-lead ECG. The origins of PVC are localized by calculating the weighted center of gravity of classification returned by the CNNs. It was evaluated in computer simulation and in 90 PVC beats from nine patients. The results demonstrate the capability and merits of the proposed method for localization of PVC. This work suggests a new approach for cardiac source localization of origins of arrhythmias using only the 12-lead ECG by means of CNN.
  • Thumbnail Image
    Item
    On the Effectiveness of Neural Networks Classifying the MNIST Dataset
    (2017-03) Blum, Carter W
    Convolutional Neural Networks (CNNs) are the primary driver of the explosion of computer vision. Initially popularized by AlexNet's performance in the ImageNet Competition in 2012, convolutional neural networks have since far-surpassed the traditional `hand-wired' models that were previously used in computer vision. They have been a focus of major investment and research from major institutes such as Google and OpenAI. This project is part 1 of a 2 part project researching potential optimizations of CNNs in the areas of convergence, processing speed, over fitting and accuracy. The first semester of the project implemented several optimizations from literature and combined them with CNNs to analyze their effectiveness. It also lays the groundwork for the second semester of research, which will be focused on combining recurrency from Recurrent Neural Networks (particularly Long Short-Term Memory (LSTM) networks.
  • Thumbnail Image
    Item
    PermNet: Permuted Convolutional Neural Network
    (2021-05) Mehta, Rishabh
    Convolution filters in CNNs extract patterns from input by aggregating information across height, width and channel dimensions. Information aggregation across height and width dimensions performed using depthwise convolution, helps identify neighborhood patterns and hence is very intuitive. However the method in which channel dimension information is aggregated by channel summation seems mathematically simplistic and out of convenience. In this project we attempt to improve the channel dimension aggregation operations. The first approach introduces weighted summation channel aggregation in convolutions. The second approach introduces permuted convolutions which attempt to perform psuedo-width scaling by generating new constrained filters from existing filters. Implementing permuted convolutions comes with many challenges such as permutation explosion, stochasticity, higher memory and computation requirements. To resolve these issues, we come up with multiple variants of permuted convolutions and present their advantages and disadvantages. Lastly, we provide empirical results showcasing the performance of weighted channel summation networks and permuted convolution networks, present our findings and recommendations for future work.