Intelligent Sensing System for Smart Buildings
REU student: Aden Scott
Project description: This research project aims to develop a machine learning approach for crack classification in structural health monitoring ( The study focuses on the application of machine learning algorithms to predict the opening and growth of cracks, their location, orientation and dimensions using simulation results obtained from CST Studio Suite The proposed approach aims to enhance the accuracy of crack detection sensors, specifically RFID based sensors in SHM systems The anticipated outcomes include the development of a robust machine learning algorithm capable of accurate crack classification with potential benefits for improved maintenance strategies, enhanced safety, and cost savings in SHM applications.
Comparing Models of Medical Image Classification with Segmentation
REU student: Anika Arifin
Project description: Image segmentation provides a vital method for image analysis in various fields and applications, and different models have been designed over time to optimize this process. General testing measures along with specific image segmentation statistic tests are applied to different architectures and classifiers to gauge similarities and differences. In this poster, we compare models of image segmentation (UNet and FCN) and classifiers (SVM and Naïve Bayes’) to measure the statistical significance of each model. The goal is to analyze the resulting statistical data to find which model provides the best application of image segmentation in the medical field. Finding this result will help improve the accuracy of medical diagnosis through image segmentation.
Parallel Glowworm Swarm Optimization Classification Algorithm Implemented with Apache Spark
REU student: Bren Hutchinson
Project description: With the ever-growing relevance of big data due to the growth of the internet and technology, there is large demand for large-scale data management and classification. In this paper, we proposed an implementation of the glowworm swarm optimization classification algorithm (SCGSO) that is parallelized using the Apache Spark framework. The main idea of SCGSO is to use the capabilities of the standardized GSO algorithm in finding multi-modal solutions and apply it to to severaltarget centroid labels, and assigning any unlabeled data points to the nearest centroid. For the experimentation, four datasets were used to evaluate the SCGSO algorithm with varying dimensionality and number of data points. The experimental results show that the algorithm performs better for lower dimensionality data sets, and scales nicely with size of dataset.
Exploring Generative Adversarial Networks for Diabetic Foot Ulcer Image Segmentation
REU student: David Roth
Project description: The early and accurate diagnosis of diabetic foot ulcers (DFUs) is crucial for effective patient care. However, the conventional approach of visual inspection and manual measurements by medical experts can be subject to human error, leading to limited precision in ulcer assessment. To overcome these limitations and enhance the diagnostic process, this research focuses on leveraging advanced image segmentation techniques. While certain convolutional neural network architectures, such as U-Net and SegNet, have been applied for image segmentation, this paper delves into exploring the untapped potential of Generative Adversarial Networks (GANs) in this domain. GANs have shown remarkable success in various computer vision tasks, including image generation and image-to-image translation. We aim to investigate the effectiveness of GAN-based image segmentation methods, particularly Pix2Pix and SegAN, in accurately identifying and segmenting DFU medical images. To accomplish this, we propose the use of performance measures such as the Dice Coefficient and the Jaccard Index, among others. By identifying the most effective GAN-based approach for DFU segmentation, this research seeks to contribute to the development of more reliable and automated diagnostic tools, leading to improved patient outcomes and reduced workload for healthcare professionals.
Mobile Application for Beef Cut Classification
REU student: Samantha Hong
Project description: The average consumer relies on accurate labeling for information about beef cut source, nutrition, and quality. Mass produced beef products are often mislabeled. This research aims to develop a machine-learning based mobile app that identifies beef quality to improve consumer knowledge when purchasing beef products.
Firefly Spark Classification Optimization
REU student: Jacob Olinger
Project description: Classification is a problem at the forefront of computer science. However, models for classification remain extremely computationally expensive. Therefore, stochastic algorithms provide a more efficient manner of model training. Furthermore, the Apache Spark context provides parallel processing capabilities in order to further improve classifier efficiency through increasing concurrency.
Using Chemiresistive Nanomaterials to Develop a Rapid Diabetes Sensor and Derive a Resistance-Acetone Relationship
REU student: Jonah Sachs
Project description: Specific nanomaterials have been shown to possess a functionalizable resistance when placed in contact with acetone, which is severely overrepresented in the breath product of patients with medical diabetes. The sensor created will take a form similar to that of an alcohol-based breathalyzer commonly used by Police for detecting DUIs. It will utilize the chemiresistive property of a unique nanomaterial mixture, tracking resistance changes based upon the presence of acetone. Different Regression Models will be utilized to attempt to quantify Acetone levels (ppm) when given a resistance-based time series, and also in exposure to other extraneous variables. In addition, Fourier analysis and other physical techniques will be utilized. The end goal is the creation of either a threshold or model which can accurately quantify acetone based on resistance over time.
AI-Enabled Breath Sensor
REU student: Kenny Tran
Project description: Current diagnosis methods for diabetes involve long wait times between data collection and diagnosis, as well as high costs associated with equipment and specialized operator requirements. The emergence and development of MAX phase 2-dimensional nanomaterials show significant promise for a low-cost rapid assessment device for use in the diagnosis of diabetes in its early stages.
Cell Culture Image Classification Using Deep Learning
REU student: Talia Gafrick
Project description: The emergence of deep learning has highlighted the cruciality of transfer learning in numerous fields, particularly in medical imaging. However, the availability of this data is scarce due to quality standards, patient privacy, monetary incentive,etc. Transfer learning has provided a solution to this challenge. This process modifies a pre-trained image neural network architecture by fine-tuning output layers with the smaller target dataset. This poster focuses on the examination of transfer learning with breast and prostate cancer cell culture imaging. The performance of three established deep neural networks- ResNet50, VGG19, and InceptionV3- on a target dataset with various fine-tuning configurations is evaluated.
Optimum Humanoid Footstep Planning
REU student: Scott Elgaard
Project description: Humanoid footstep planning is a challenging problem in robotics, requiring efficient algorithms to generate optimal paths while considering factors such as stability. feasibility, and computational complexity. Rapidly Exploring Random Tree (RRT) and RRT* have emerged as popular search algorithms for path planning due to their completeness. However, these algorithms often suffer from high computational costs and lack optimization, especially when applied to humanoid footstep planning.