In-situ segmentation and post-processing of phytoplankton image data suitable for an autonomous surface vehicle

The utilization of autonomous surface vehicles (ASVs) for in-situ and real-time identification of phytoplankton species can provide early warning and thereby offer rapid solutions for mitigating the detrimental impacts of harmful algal blooms (HABs). Onboard imaging systems on mobile platforms, where in-situ images of main phytoplankton species are captured, and data are transmitted back to a taxonomist or fed into a machine learning model for inference, are promising technologies on the horizon. When the vehicle is in a remote location with limited connectivity, unnecessary data should be removed before transmitting, and prior to inference, the data must be segmented and processed. Here, methods for post-processing microscopic images of phytoplankton captured by a system built for autonomous operation are presented. Lightweight image manipulation methods with open-source software such as FIJI are used to remove clogged particles from the image sequence, as well as normalize the background and perform thresholding to generate binary masks well suited for inference in a machine learning model. A proof of concept is demonstrated and evaluated by feeding image data of phytoplankton into a neural network for classification.