Phytoplankton enumeration is an essential part of monitoring ocean health and aquaculture management. Automated in situ sampling provides the capacity for high temporal resolution monitoring and imaging flow cytometry used in conjunction with machine learning can accurately monitor phytoplankton community composition. However, the accurate measurement of chain-forming diatoms is problematic for image classifiers where multiple cells may be visible in a single image. To address this issue, we propose a novel Diatom Chain Counter Model (DCCM) based on the “You Only Look Once” (YOLO) architecture, which uses object detection to efficiently count the number of cells in a chain. Focussing on four harmful genera (Chaetoceros, Pseudo-nitzschia, Skeletonema and Thalassiosira) the DCCM model shows a high level of accuracy when compared with manual counts resulting in a > 95% accuracy across all genera, rising to > 99% for Pseudo-nitzschia and Skeletonema. Applied to historic data from two Imaging FlowCytobots the increase in cell density estimations rose by a significant amount, particularly for Thalassiosira (multiplication factor of 4.93) and Chaetoceros (multiplication factor 6.97 average). The enumeration of diatom chain length is valuable for warning of harmful algal blooms, and enhanced time series based understanding of the relationship of phytoplankton with their environment.