Research On Intelligent Detection Of Stem Cell Culture Equipment Based On Computer Vision

Stem cells are a kind of pluripotent cells with the ability of self-replication.With the development of medical technology,Stem cell culture has gradually become a common concern in the biomedical field.However,the design of the traditional cell culture tank is only limited to the completion of the required cell culture.Not only the equipment is not unified,but also the volume is too large,which requires frequent manual intervention.As a result,the number of cultured cells is far from meeting the market demand.Therefore,the development of automatic stem cell culture device is urgent.Automatic stem cell culture device has unique requirements for the universality,usability,reliability,high efficiency and platform computing power of the stem cell visual recognition algorithm.In this paper,the segmentation,localization and classification of stem cell images are systematically studied and the algorithm is realized by combining the requirements of the automatic stem cell culture device and referring to the existing research results of biomedical images.The study included:1.The stem cell image segmentation scheme was designed.The bimodal histogram method,adaptive threshold method,Ostu method and two-dimensional maximum entropy threshold method are analyzed,and the quantum particle swarm optimization algorithm is introduced to improve the two-dimensional maximum entropy threshold algorithm.2.The sub-image location and range calibration scheme are designed for the segmented binary image.The depth-first search algorithm and breadth-first search algorithm are improved for sub-image positioning.Combined with the working principle of breadth-first search traversal and the technical requirements of the task in this paper,the search algorithm is proposed as the localization scheme of cell sub-image.It is proved that the efficiency of this scheme is better than breadth-first search traversal.3.The classification scheme of the positioned sub-image is designedsThe convolutional neural network algorithm which excellent performance in the field of image classification is introduced to replace the traditional machine learning.The computing power level of the platform is fully considered and the classical model is simplified to a certain extent to reduce the computing cost.Model training was conducted with half of the sample images and the original image was classified and tested.The test results show that the model can recognize the majority of homologous cells and can eliminate the background and impurities of false segmentation,which basically meets the design requirements.The obtained results can be used in the development of the equipment and have engineering application value.