Rudy, Ian, Computer Science, Susquehanna University, 1858 Weber Way box 3460, Selinsgrove, PA, 17870, email@example.com; Wilson, Matthew, Ecology, Susquehanna University, CEER Freshwater Research Institute, 514 University Ave, Selinsgrove, PA, 17870, firstname.lastname@example.org.
The measurement of turbidity serves as a key indicator of water quality and purity, crucial for informing decisions related to industrial, ecological, and public health applications. As existing processes require both additional expenses and additional steps to be taken during data collection, we seek to generate accurate estimations of turbidity, measured in Formazin Nephelometric Units (FNU), from underwater images. Such a process could give new insight to existing datasets, such as the turbidity levels different plants or animals will tolerate, or provide a cheaper alternative to measuring turbidity when high accuracy is not necessary. This will be achieved through the use of machine vision to create an image classification or regression model trained on image data and their corresponding turbidity values recorded from a turbidimeter using nephelometry measures. Data will be collected in the lab in flowing and standing water mesocosms. Varying amounts of sediment ranging from 0-55 FNU will be used, along with substances that produce visual differences, such as ink, without affecting turbidity. Data will also be collected in the field with different conditions, such as lighting, water depth, substrate, and ambient turbidity. Two categories of image will be collected, one with a Secchi disk in the image, and one without. Early testing has shown promise that the images with Secchi disks can create an accurate model, and this is likely to improve with increased model training. As of now, no testing has been conducted with images not including Secchi disks, but we plan to test the versatility and accuracy of a model trained on native substrate images. Such a process could provide a cost-effective way to measure turbidity without the acquisition of new equipment, and could ideally be applied retroactively to existing data, saving the time required to collect new data.
Turbidity, Machine Vision