High-resolution optical imaging systems are quickly becoming universal tools to characterize the plankton diversity in marine ecosystems. Imaging FlowCytobot (IFCB) are most developed and well applied in Coast of US, which has reported to successfully forecast the blooms of Dinophysis ovum and Karenia brevis. The mixotrophic Mesodinium rubrum, a planktonic ciliate, generated impressive non-toxic blooms in coastal ocean but the food for Diarrhetic Shellfish Poisoning (DSP) producers Dinophysis, is proved to be the great precursor to serious DSP outbreaks. Usually, the common sampling and taxonomy methods are unable to detect the low biomass of Mesodinium. Therefore, large biomass of M. rubrum appeared suddenly and lack of warning period. The IFCB coupled automatic identification systems are used with highly resolution rate in the field study. Here, we unprecedentedly collected the original image data of phytoplankton (10~150μm) by IFCB at Zhoushan seashore, the western part of the East China Sea over March to June 2022, during which an unexpected M. rubrum bloom were captured and would be associated with the temperature. Automated detection systems, convolutional neural networks (CNN) were developed in present study to identify the plankton species. The process included 1) manually classify 12965 images collected by IFCB in the field into 35 categories, 2) select 17 categories containing 56 to 200 images with high imaging quality in an effort to reduce highly imbalanced class distributions and high intraclass variance, 3) obtain 2741 images as the training set and 4) develop CNN with the pretrained network EfficientNet-b0. The established CNN classifier had a result of overall accuracy, recall, and F1-score of 96.62%, 96.61% and 96.55% respectively. Our CNN was generally able to fast capture the plankton dynamic. On March 9^th, the population of Mesodinium rubrum blooming from 296 cells·L^-1 to 5,200 cells·L^-1 within 3 days with rising temperature from 14.3 °C to 17.2 °C, while its average population was 850 cells·L^-1 during study period, which is thought very likely to develop into a larger harmful bloom attracting Dinophysis. Besides, the population of Prorocentrum donghaiense reduced from 2,716 cells·L^-1 to 931 cells·L^-1, and Scrippsiella trochoidea appeared but Pinnularia was extinct according to the automatic classification results of CNN. Furthermore, we aim to establish the classification model with more classes and increase the identification accuracy to better characterize coastal phytoplankton communities threatened by rapidly changing environmental conditions.