Generation of microbial colonies dataset with deep learning style transfer

Abstract

We introduce an effective strategy to generate a synthetic dataset of microbiological images of Petri dishes that can be used to train deep learning models. The developed generator employs traditional computer vision algorithms together with a neural style transfer method for data augmentation. We show that the method is able to synthesize a dataset of realistic looking images that can be used to train a neural network model capable of localising, segmenting, and classifying five different microbial species. Our method requires significantly fewer resources to obtain a useful dataset than collecting and labeling a whole large set of real images with annotations. We show that starting with only 100 real images, we can generate data to train a detector that achieves comparable results to the same detector but trained on a real, several dozen times bigger dataset. We prove the usefulness of the method in microbe detection and segmentation, but we expect that it is general and flexible and can also be applicable in other domains of science and industry to detect various objects.

Self-Normalized Density Map (SNDM) for Counting Microbiological Objects

“…The biggest advantage of this method is that it is enough to label each image with the number of objects in it. If a dataset includes more detailed annotation with a bounding box for every object in an image, it is feasible to leverage detectors for object counting [23][24][25] . Eventually, an autoencoder can be used to estimate the density map (DM) based on a given image, which can be later integrated to obtain the number of objects [26][27][28][29][30] .…”

“…Traditionally, this task is done manually or semi-automatically using traditional computer vision methods 31,32 . However, recent studies prove that a DL-based methodology accelerates the process 23,25,29,[33][34][35] . We shall utilize the DM method to predict the number of microbial colonies on a Petri dish.…”