Counting Method of Microfluidic Phytoplankton Based on Object Detection and Deduplication

Zihao Zhang,Gaofang Yin,Nanjing Zhao,Renqing Jia
DOI: https://doi.org/10.1109/ainit61980.2024.10581494
2024-01-01
Abstract:Phytoplankton is a crucial primary producer in aquatic ecosystems and serves as an indicator of their ecological health. Accurate counting of phytoplankton is essential for diagnosing and monitoring water quality. Microfluidic microscopic imaging counting is an important advancement for accurately counting phytoplankton in water ecosystems. In microfluidic microscopy images of phytoplankton, the same algae can easily appear at different locations in adjacent frames, leading to counting errors. To address this issue, this paper proposes a counting method of microfluidic phytoplankton based on object detection and deduplication. This method uses slicing operations to divide microscopic images into multiple parts and inputs them into the YOLOv5 model for inference to detect algae. Then, the weighted evaluation of perceptual hash and color histogram is utilized to comprehensively assess the similarity of algae in adjacent frames. Algae with high similarity are considered to be identical, and duplicate algae are removed to ensure accurate counting. In this study, we conducted comparative experiments on seven different species of algae. The results demonstrate that the detection method achieved higher accuracy, recall rate, mAP@0.5, and mAP@0.5:0.95 scores of 96.8%, 98.1%, 98.9%, and 92.5%, respectively, surpassing the performance of the original detection method. Furthermore, the deduplication algorithm effectively reduced the counting error of phytoplankton to 1.15%. The counting method proposed addresses the issue of duplicate algae counting and enables rapid and accurate counting of phytoplankton.
What problem does this paper attempt to address?