Search Results (1 - 3 of 3 Results)

Sort By  
Sort Dir
 
Results per page  

Howard, Shaun MichaelDeep Learning for Sensor Fusion
Master of Sciences (Engineering), Case Western Reserve University, 2017, EECS - Computer and Information Sciences
The use of multiple sensors in modern day vehicular applications is necessary to provide a complete outlook of surroundings for advanced driver assistance systems (ADAS) and automated driving. The fusion of these sensors provides increased certainty in the recognition, localization and prediction of surroundings. A deep learning-based sensor fusion system is proposed to fuse two independent, multi-modal sensor sources. This system is shown to successfully learn the complex capabilities of an existing state-of-the-art sensor fusion system and generalize well to new sensor fusion datasets. It has high precision and recall with minimal confusion after training on several million examples of labeled multi-modal sensor data. It is robust, has a sustainable training time, and has real-time response capabilities on a deep learning PC with a single NVIDIA GeForce GTX 980Ti graphical processing unit (GPU).

Committee:

Wyatt Newman, Dr (Committee Chair); M. Cenk Cavusoglu, Dr (Committee Member); Michael Lewicki, Dr (Committee Member)

Subjects:

Artificial Intelligence; Computer Science

Keywords:

deep learning; sensor fusion; deep neural networks; advanced driver assistance systems; automated driving; multi-stream neural networks; feedforward; multilayer perceptron; recurrent; gated recurrent unit; long-short term memory; camera; radar;

Aspiras, Theus H.Emotion Recognition using Spatiotemporal Analysis of Electroencephalographic Signals
Master of Science (M.S.), University of Dayton, 2012, Electrical Engineering
Emotion recognition using electroencephalographic (EEG) recordings is a new area of research which focuses on recognition of emotional states of mind rather than impulsive responses. EEG recordings are found useful for the detection of emotions through monitoring the emotion characteristics of spatiotemporal variations of activations inside the brain. To distinguish between different emotions using EEG data, we need to provide specific spectral descriptors as features to quantify these spatiotemporal variations. We propose several new features, namely Normalized Root Mean Square (NRMS), Absolute Logarithm Normalized Root Mean Square (ALRMS), Logarithmic Power (LP), Normalized Logarithmic Power (NLP), and Absolute Logarithm Normalized Logarithmic Power (ALNLP) for the classification of emotions. A protocol has been established to elicit five distinct emotions: joy, sadness, disgust, fear, surprise, and neutral. EEG signals are collected using a 256-channel system, preprocessed using band-pass filters and a Laplacian Montage, and decomposed into five frequency bands using Discrete Wavelet Transform. The decomposed signals are transformed into different spectral descriptors and are classified using a two-layer Multilayer Perceptron (MLP) neural network. The Logarithmic Power descriptor produces the highest recognition rates, 91.82% and 94.27% recognition for two different experiments, which is more than 2% higher than when using other features.

Committee:

Vijayan Asari, PhD (Committee Chair); Tarek Taha, PhD (Committee Member); Eric Balster, PhD (Committee Member)

Subjects:

Computer Engineering; Electrical Engineering; Engineering; Neurosciences; Psychology

Keywords:

Emotion Recognition; Electroencephalography; Wavelet Decomposition; Multilayer Perceptron; Laplacian Montage; International Affective Picture System

Gao, ZhenningParallel and Distributed Implementation of A Multilayer Perceptron Neural Network on A Wireless Sensor Network
Master of Science, University of Toledo, Engineering (Computer Science)
This thesis presents a study on implementing the multilayer perceptron neural network on the wireless sensor network in a parallel and distributed way. We take advantage of the topological resemblance between the multilayer perceptron and wireless sensor network. A single neuron in the multilayer perceptron neural network is implemented on a wireless sensor node, and the connections between neurons are achieved by the wireless links between nodes. While the computation of the multilayer perceptron benefits from the massive parallelism and the fully distribution when the wireless sensor network is serving as the hardware platform, it is still unknown whether the delay and drop phenomena for message packets carrying neuron outputs would prohibit the multilayer perceptron from getting a decent performance. A simulation-based empirical study is conducted to assess the performance profile of the multilayer perceptron on a number of different problems. Simulation study is performed using a simulator which is developed in-house for the unique requirements of the study proposed herein. The simulator only simulates the major effects of wireless sensor network operation which influence the running of multilayer perceptron. A model for delay and drop in wireless sensor network is proposed for creating the simulator. The setting of the simulation is well defined. Back-Propagation with Momentum learning is employed as the learning algorithms for the neural network. A formula for the number of neurons in the hidden layer neuron is chosen by empirical study. The simulation is done under different network topology and condition of delay and drop for the wireless sensor network. Seven data sets, namely Iris, Wine, Ionosphere, Dermatology, Handwritten Numerical, Isolet and Gisette, with the attributes counts up to 5000 and instances counts up to 7797 are employed to profile the performance. The simulation results are compared with those from the literature and through the non-distributed multilayer perceptron. Comparative performance evaluation suggests that the performance of multilayer perceptron using wireless sensor network as the hardware platform is comparable with other machine learning algorithms and as good as the non-distributed multilayer perceptron. The time and message complexity have been analyzed and it shows the scalability of the proposed method is promising.

Committee:

Gursel Serpen (Advisor); Mohsin Jamali (Committee Member); Ezzatollah Salari (Committee Member)

Subjects:

Artificial Intelligence; Computer Science

Keywords:

Artificial Intelligence; Artificial Neural Network; Machine Learning; Multilayer Perceptron; Wireless Sensor Network; Parallel Computing; Distributed Computing