| The morphology of microorganisms changes with the growth state,product accumulation and cell stress,and the intracellular physiological and fermentation status can be understood through the information of microbial morphology.Schizochytrium sp.is a heterotrophic marine protist with extremely high lipid content,and there are few studies on the relationship between morphology and lipid production of Schizochytrium sp.Image recognition technology can extract effective information from images,and has the advantages of automatic,rapid,objective,and high analysis accuracy.It has been widely used in microbial segmentation,counting,classification,and target detection.This paper took Schizochytrium sp.S31 as the starting strain,analyzed the relationship between its morphology and fermentation state,established seven typical single-cell classification models in the life cycle of Schizochytrium sp.S31 based on deep learning and established population cell target detection in the lipid production stage of Schizochytrium sp.S31.The model was used to assist in the determination of the fermentation state of Schizochytrium sp.S31 in different fermentation batches.Meanwhile,a strain with high biomass and short fermentation period was obtained through directional domestication combined with target detection technology.（1）Based on the observation and analysis of the growth and lipid accumulation process of Schizochytrium sp.S31,a single-cell classification model in the life cycle of Schizochytrium sp.S31 was established.Observed the morphological changes of Schizochytrium sp.S31 during the fermentation process,and analyzed the relationship between the morphologies and the fermentation state.The single-cell images of Schizochytrium sp.S31 were divided into seven typical types,including single cells,dyads,tetrads,polydites,cells not filled with lipid,cells filled with large lipid droplets,and cell fragments.Taking 881 single-cell images as the training set,after data preprocessing,a densely connected convolutional network based on the triple attention mechanism was used for modeling,and an additional 219 cell images were selected as the test set.The results showed that the classification accuracy rate is 92.24%.（2）Established a population cell target detection model during the lipid production process of Schizochytrium sp.S31 to assist in detecting the fermentation status of different fermentation batches.Microscopic images of the lipid production process of Schizochytrium sp.S31 were collected,manually framed and marked,and combined with YOLOv3 to establish a target detection model of Schizochytrium sp.S31,with an average accuracy rate of74.35%.Different fermentation batches were simulated,combined with target detection to assist in judging the state of lipid accumulation.The visualization and statistical results of target detection combined with shake flask fermentation data showed that target detection technology could be used to assist in judging the fermentation state of Schizochytrium sp.S31.（3）Targeted high oxygen domestication of Schizochytrium sp.S31,combined with target detection to monitor the lipid production process of Schizochytrium sp.S31.Schizochytrium sp.S31 was domesticated and cultivated at 260 r·min-1 for 4 rounds,and 4 domesticated strains were obtained.The biomass of the 4 strains were 28.87 g·L-1,29.96 g·L-1,32.25 g·L-1,36.77 g·L-1,the lipid yields were 11.55 g·L-1,12.54 g·L-1,13.69 g·L-1,15.85 g·L-1.There was no significant difference in the strain morphology during the domestication process.Compared with the original strain,the biomass of the ALE40 strain increased by 29.34%,the lipid production increased by 35.70%,and the fermentation period was shortened by 12 h.Target detection was carried out on the 72 h and 108 h microscopic images,and the statistical results combined with the visualized images showed that target detection can assist in monitoring the lipid production process of Schizochytrium sp.S31. |
PDF Full Text Request