Online Visualization and Analysis of Image Data

Ummu Yuzugulluer

Co-Presenters: Amir Jaraysa

College: The Dorothy and George Hennings College of Science, Mathematics and Technology

Major: Computer Science

Faculty Research Mentor: Ching-yu Huang

Abstract:

This project focuses on developing a web-based image classification system that allows users to upload, classify, and continuously improve the model over time. The system is built using TensorFlow, PHP, JavaScript, HTML, and Python and is hosted on Obi at Kean University.Unlike traditional classification models that stay the same after training, this system keeps learning and evolving. Users can add new objects, and once 10-15 new images are uploaded, the model is retrained to improve classification. To keep this process running smoothly, nohup is used to handle training in the background, so it continues even if the session is closed. This ensures uninterrupted learning without needing manual supervision.When an image is uploaded, the system processes it by resizing and formatting it before sending it through a convolutional neural network (CNN) for classification. The backend takes care of storing and managing images, ensuring that everything runs as expected, while the frontend is designed to be simple and easy to use. The goal is to make sure users can upload images without needing technical knowledge.One challenge has been integrating TensorFlow, PHP, and backend processes smoothly while maintaining system efficiency. Optimizing image uploads, processing, and classification has required careful tuning to prevent delays and ensure accurate results. Proper handling of images, from preprocessing to classification, is essential for seamless operation. This is particularly important in applications like object recognition, security, and automation, where performance and reliability are critical for accurate decision-making and smooth system functionality.