OrcaAL | Orca detection Active Learning tool

Underwater noise pollution impedes the orcas’ ability to communicate and echolocate scarce salmon. To be able to detect orca calls, a deep learning model can be trained, but due to the lack of a vast amount of labeled data, an active learning tool is necessary.
This project provides an active learning web application, where the user can visualize the predictions of a deep learning binary classification algorithm on unlabeled data. Users are then allowed to annotate more data by hearing a sample and correcting its label. The project has been built in a modular way to allow the integration with different machine learning models.