Figure 2.
Method overview. The input image is first augmented by RCTA, then bacterial colonies need to be manually annotated. The annotation files and the augmented images will be used as the input of improved YOLOv3 for training. The trained model needs to be converted to the intermediate ONNX format first and then converted to the TRT format that can be deployed on jetson nano with high processing speed. The colony detection and analysis will be performed by the converted model.