06-28, 14:00–14:30 (Europe/Tirane), UBT F / N212 - Floor 3
Landfill sites are for storing waste in a secure and secluded manner but they can cause a lot of damage to the environment by generating greenhouse gases and contaminating soils by releasing heavy metals and toxins. Monitoring the area of landfill sites from space is a challenging problem because of the huge amount of unstructured data and unavailability of standard datasets and procedures. By combining open-source tools with geospatial data, we present a global dataset that monitors the changes in the landfill area. We have achieved this by developing a deep learning based segmentation model that uses multispectral satellite data and segments the landfill areas from them. In order to develop the model, we have labelled landfill sites from optical imagery from all over the world. Our current segmentation model has 31 million parameters and has achieved an accuracy of 77.6% on the test set. Currently, the dataset contains temporal data from 2021 of the major landfill sites from more than 7 countries and it is growing daily as new data is coming in. In future, we aim to enhance this dataset by adding more variables other than the area, for instance height of the landfill and will also explore other higher resolution data for validating our results further.
I am a data scientist currently working at Blue Sky Analytics, I am passionate about Geospatial data sciences. My core expertise exists in design, development and deploying of machine learning and deep learning models that run on gigabytes of data. I also have experience in developing Intelligent Geospatial REST API based services on cloud infrastructure.