Intelligent Counting And Classification Methods Of Living Colonies

Colony counting has important applications in food testing,environmental protection and health care.Most colony counting algorithms need to process composite colony images.The bacterial species of the composite colony image are complex,and the morphological characteristics of the colonies are different.The traditional colony counting method bases on gray scale,and various colony analyzers on the market faces certain difficulties when dealing with composite colony images.Moreover,the current colony counting algorithm only processes a single image or a single frame image in a video,and obtains the number of connected domains in the culture dish,so that interference of impurities or dead bacteria on the living colony count cannot be excluded.The living colony count is the basis of colony analysis,and it is usually necessary to classify and analyze colonies according to their cultural characteristics.To count living colonies,firstly to established a sequence of colony images,that was,to take an image of a medium at multiple time points.To sequentially sequenced the colony images to adjust the image position.Used the three-phase level set algorithm to segment the colony image.Obtained all connected domains in the culture dish,and compared multiple adhesion connected domain segmentation methods.Selected the optimal method to segment the connected domains.Tracked and recorded the growth process of each connected domain.Then established the area feature set of each connected domain in chronological order.Excluded the connected domains where the area had not changed significantly,the connected domains of the living colonies could be obtained.Finally,the remaining connected domains were counted,and the number of living colonies inside the medium could be accurately obtained.After counting live colonies,it was necessary to complete the classification task of the colonies.The living colonies were extracted from the connected domains and normalized into gray scale images of the same size.The normalized gray scale images were amplifying the data,and the training set and test set were established.The convolution neural network was constructed.The network was trained and tested,and the optimal model was selected.The average accuracy of the five classifications of colonies in the experiment reached 84.31%.After completing the living colony counting and classification task,the image rotation was optimized and accelerated by frequency domain processing.The GPU optimization library was accelerated by the colony counting module of Open CV,and the multi-core CPU was used to accelerate the adhesion colony segmentation by Open MP.Finally,the transplantation algorithm integrated some algorithm functions into the colony analysis software.The intelligent colony intelligent counting and classification algorithm designed in this paper could not only accurately obtain the number of living colonies in the culture dish,but also eliminate the interference of water stains.It can also morphologically classify the colonies according to the need of the experimenters.Compared with the existing colony analysis algorithm or the existing colony analysis software on the market,the proposed algorithm has advantages in counting and classification.