Online Ocean Monitoring using Edge IoT

Online monitoring of oceans like pollution monitoring, algae blooms, tsunami warnings, etc., are really challenging because of the latency involved in sending, storing, and processing of data in the cloud. To mitigate this issue, machine learning models have to be run at the edge close to where the data is generated but requires a huge dataset at monitoring equipment and must withstand the intermittent network connectivity thus be able to take the own decisions. This paper describes the implementation of Greengrass IoT, an Edge computing software developed by Amazon Web Services at maritime environments by running the local machine learning inferences without the need for datasets at the monitoring equipment. In this paper, the amazon sagemaker neo deep learning runtime is deployed at the Greengrass core (Raspberry-pi) for marine plastic pollution monitoring which predicts the images taken from the raspberry-pi camera based on Resnet-50 neural network model before sending data to the cloud.