Development of new artificial learning models for colony morphology prediction

The Medium is environment that have been formulated for the growth of microorganisms. These can be for the different purposes such as growth, isolation, identification, counting, sensitivity tests of microorganisms, sterility testing, analysis of clinical samples, food, water, environmental controls, the acquisition of biological products, antibiotics and vitamins analysis, industrial analysis and so on. Structures which are formed by reproduced microorganisms and can be seen by eye are called colony. Colonies formed on the agar, creating images of different morphological characteristics depending on the microorganism and growth media. In the large number of colonies counted by hand, colony counting which is needed for many applications in areas such as biotechnology and pathology is boring, time consuming and it a process prone to human error. It depending on the decisions and abilities which can vary in relation to the laboratory worker fatigue, laborant's work rate limit the number of petri dishes can be evaluated at specific time interval. In this thesis, computer aided classification with image processing and machine learning methods on mold, yeast and bacteria colonie images which is used to determine the results of microbiological analysis of products in the dairy industry. Image processing is a method to convert the image to digital form and perform some operations on in order to extract some useful information from the image. This is a kind of signal processing whose input is an image such as video frame or photo and output is an image or features of the image. Artificial learning systems have emerged from needs such as automatic recognition of objects, signals, images by computers and automated decision-making process based on certain parameters by computers. There are two basic information on the classification of molds, yeasts and bacteria by the people: color information and shape information. When human eye making an assessment, the number of different colors in the colony image, color changes and transitions, different geometric shapes and corner shape changes in the colony image affects on the decision to be made. This hypothesis lead us to develeop two shape based and two color based feature extraction technique named ŞTY1,ŞTY2,RTY1 and RTY2 based on the characteristics that we think they effective to detection by human eye. For the determination of the appropriate classification system for solving the problem we applied PCA, LDA feature transformations, SBS feature selection algorithms to dataset and they have been classified by Naive Bayes, k-NN, C4.5 and ANN algorithms. According to empirical results obtained, the system that composed RTY2 feature extraction, PCA feature transformation and ANN classifier, have achieved better performance compared to other systems.