U2-Net and ResNet50-Based Automatic Pipeline for Bacterial Colony Counting

• DOI: 10.3390/microorganisms12010201
• Publication Date: 2024-01-18T16:28:46Z
• Authors: Libo Cao; Liping Zeng; Yaoxuan Wang; Jiayi Cao; Ziyu Han; Yang Chen; Yuxi Wang; Guowei Zhong; Shanlei Qiao
• Keywords: Bacterial colony

In this paper, an automatic colony counting system based on improved image preprocessing algorithm and convolutional neural network (CNN)-assisted method was developed. Firstly, we assembled LED backlighting illumination platform as capturing to obtain photographs of laboratory cultures. Consequently, a dataset introduced consisting 390 photos agar plate cultures, which included 8 microorganisms. Secondly, implemented new for light intensity correction, facilitated clearer differentiation between media areas. Thirdly, U2-Net used predict the probability distribution edge Petri dish in images locate region interest (ROI), then threshold segmentation applied separate it. This achieved F1 score 99.5% mean absolute error (MAE) 0.0033 validation set. Then, another within ROI. 96.5% MAE 0.005 After that, area segmented into multiple components containing single or adhesive colonies. Finally, (CC) were innovatively rotated crops resized input (with 14,921 training set 4281 set) ResNet50 automatically count number Our overall recovery 97.82% exhibited excellent performance adhesion classification. To best our knowledge, proposed “light correction-based preprocessing→U2-Net edge→U2-Net region→ResNet50-based counting” scheme represents attempt demonstrates high degree automation accuracy recognizing single-colony multi-colony targets.