You are here:
Harmful algal bloom smart device application: using image analysis and machine learning techniques for classification of harmful algal blooms
Citation:
Waters, M., M. Kannan, Jim Lazorchak, AND Joel Allen. Harmful algal bloom smart device application: using image analysis and machine learning techniques for classification of harmful algal blooms. SETAC NA Annual Meeting, Minneapolis, MN, November 12 - 16, 2017.
Impact/Purpose:
This presentation explains the app development for detection of Bluegreen Algae and the identification of phytoplankton
Description:
Northern Kentucky University and the U.S. EPA Office of Research Development in Cincinnati Agency are collaborating to develop a harmful algal bloom detection algorithm that estimates the presence of cyanobacteria in freshwater systems by image analysis. Green and blue-green algae exhibit different Hue-Saturation-Value color histograms in digital photographs. These differences are exploited by machine learning techniques to train a smart device (cellular phone, tablet, or similar) to detect the presence of cyanobacteria in a small surface portion of a freshwater system. The Harmful Algal Bloom Classification Application (HAB APP) has been field tested and verified to classify both green and blue-green algae. Specifically, the APP has been tested on several small streams and ponds, correctly classifying green algal blooms and has been tested on the Ohio River, correctly classifying blue-green algae in the 636-mile cyanobacteria bloom in summer 2015. The application is being tested via fixed camera monitoring stations and optimized at several locations along the Ohio River and in Lake Harsha, a 22,000-acre reservoir which supplies six million gallons per day of drinking water to the Ohio county in which it lies and is a source of many recreational activities, including swimming, boating, and fishing. The presence will be verified by other detection instruments and in vitro by agency scientists and hysteresis techniques will be used to monitor the presence of cyanobacteria on a periodic (e.g. daily, seasonally) basis at the monitoring stations. Further, the APP is being extended to classify harmful algae microscopically at the genus level using a convolutional neural network approach.
