A deep learning binary classification model that effectively identifying the presence or absence of landfill in aerial images.
This repository contains the code developed for a research paper on building a deep learning binary classification model for identifying sites of landfill. The paper was drafted for the 2023-2024 National Big Data And AI Challenge For High School Students and received the RBC Arnold Chan Memorial Award for Student Innovation at the Canada West Finals.
Read the manuscript or view a video presentation to learn more.
The VGG16 model, a pre-trained CNN on ImageNet, was chosen as the base of this project. The model was reconfigured and custom layers were added to suit the propose of landfill detection. The landfill-binary-classification file contains step by step descriptions on the constructing and training of the model.
AerialWaste (by Rocio Nahime Torresby and Piero Fraternali) was an aerial image dataset curated for the purpose of detecting illegal landfills. The dataset comprises a total of 10,977 images sourced from:
- The AGEA: Orthophotos generated through an airborne campaign executed by AGEA.
- WorldView3: High-resolution RGB images acquired from a commercial satellite sensor.
- Google Earth: RGB images obtained using the Google API.
Of the full dataset, 90% were used for training the model and 10% for testing.
This project is licensed under the MIT License.