Context
During my research project on fire and smoke detection, I developed a deep learning model designed to identify early signs of wildfires using aerial imagery. Wildfires present a serious environmental threat and are often difficult to detect before they become dangerous. The goal of this project was to build a system capable of classifying images into three categories: fire, smoke, or neutral. Using PyTorch in PyCharm, I trained a ResNet-50 neural network to analyze aerial images and detect these features. After training and evaluating the model, I deployed it on a Jetson Nano, a mobile computing platform capable of running AI models in real time. This project demonstrated how machine learning can be applied to real-world problems and potentially assist first responders in detecting wildfires earlier.
Artifact
An artifact from this project is my project documentation and implementation demonstrating the development and deployment of the fire and smoke detection model on a Jetson Nano device. The artifact includes the model architecture, dataset preparation process, training experiments, and deployment steps used to run the model on an embedded computing platform. This work highlights my ability to build and deploy machine learning systems beyond a purely theoretical environment.
Reflection
In this project, I was responsible for building the deep learning model from the ground up. This involved preparing the dataset, converting images into tensors, training the ResNet-50 architecture, and evaluating the model’s performance through multiple experiments. One of the main challenges I encountered was ensuring that the model could correctly distinguish between smoke, fire, and visually similar environmental features. I addressed this challenge by refining the training process, experimenting with different parameters, and evaluating model performance across multiple iterations.
Through this experience, I gained a deeper understanding of how machine learning models are developed, trained, and deployed in real-world environments. I strengthened my technical skills in Python, PyTorch, neural network training, and dataset preparation. Deploying the model on the Jetson Nano also introduced me to the challenges of running AI systems on edge devices, where computational resources are limited but real-time performance is required.
This experience played an important role in shaping my interest in applied AI and cybersecurity. It taught me that intelligent systems deployed in the real world must be reliable, efficient, and secure. Understanding how these models work internally also provides insight into how they might fail or be manipulated. This project represents one of my earliest hands-on experiences developing and deploying AI systems, and it helped build the foundation for my later research into adversarial machine learning and AI security.