Image processing pipeline for automated larva counting

• Published in: 2018 IEEE International Instrumentation and Measurement Technology Conference (I2MTC) pp. 1 - 5
• Main Authors: Clarke, Hayden, Horine, Brent, Thomas-Hall, Peter L.
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.05.2018
• Subjects: Image color analysis; Kernel; Ocean temperature; Pipelines; Shape; Thresholding (Imaging)
• DOI: 10.1109/I2MTC.2018.8409867

The unique biological traits that make the giant triton (Charonia tritonis) one of few natural predators of the coral-eating crown-of-thorns starfish (Acanthaster planci - COTS) are being investigated for potential use in COTS management strategies designed to protect tropical reef systems. Improved understanding of the early life history stages of the giant triton - particularly in relation to the processes necessary for successful transition between the egg, larvae and juvenile stages - is an essential prerequisite for this research. Current processes for lab-based assessment of survival rates between stages often rely on manual counts of physical samples, and with several thousand larvae needing to be counted each reproduction cycle, time and accuracy are significantly compromised. A novel image processing pipeline has been developed with functional effectiveness of the pipeline seen in the decrease in variance of root-mean-square error across different tests compared to conventional methods (sampling through radius of a known number of larvae). This is achieved using image processing techniques such as: thresholding (Otsu's method); morphology (erosion, dilation and opening); Hough transforms (Hough circle); color space transformation; and template matching. The proposed technique can significantly improve the efficiency, consistency and accuracy of data acquisition leaving researchers to invest a greater proportion of their time on data analysis and interpretation.