An automatic high throughput colony counter based on Unet++ segmentation model

Abstract

Objective: To improve the detection efficiency of the aerobic plate counter, to meet the requirements of quality management system of the inspection and testing institutions and the application of laboratory LIMS system. Methods: It was used to continuously collected of colonies image through the GigE industrial camera, variable lens and multiple light sources combined lighting system, at the same time the Unet++ segmentation model was used for image recognition processing and colony counting. Results: The colony counter had the characteristics of high collection efficiency, it only took 38 s to complete the image acquisition of one plate. It only took 3~5 s to complete the whole process of image transmission and colony identification and counting of one plate, and it had fast data processing speed and good transmission performance. Compared with the counting method required by the current standard, the error rate of the counting result was less than 8%, and it had high accuracy of counting results and good repeatability. At the same time, it realized the automatic processing of the original data of the aerobic plate count detection. Conclusion: The equipment can not only carry out high-throughput image acquisition, automatic image processing and colony counting, but also can realize the fusion with the laboratory LIMS system, effectively improve the work efficiency. Meanwhile, it can ensure the data traceability, reduce the work intensity of the test personnel, meet the error requirements of the aerobic plate count method.

Keywords

• colony counting
• automatic
• high throughput
• unet++ image segmentation model