Figure 1

Overview of the bioprocess control and monitoring and implementation of the CNN-based classifier. (A) Set up of a labelled database obtained thanks to the annotation of 4000 representative images taken by ISM. (B) Training and test phases of the model. Evaluation of the performance of the object detection algorithm is done with optimizing its mean average precision (mAP). (C) Diagram of the ISM mounted on a bioreactor for the observation of cells suspended in a nutrient medium during a biopharmaceutical production process. LED-flash illumination generates still images of cells drifting by the quartz window in front of the ISM objective. The images showing cells with different morphological patterns are acquired on-line during the bioprocess. (D) The prediction of the different classes of cells localized in the images is done on-line by the trained and tested convolutional neural network. The evolution of the concentration of each class of cells is used for process monitoring and control.