Differentiation of identical actinobacterial strains by wavelet transform and artificial neural network

Recognition of bacteria type on microbiological culture plates is not only error-prone and time-consuming but also is a costly task and also has important role in other field of microbiology. In this paper, an efficient and dependable solution is proposed for the mentioned problem by using machine learning approaches and an image processing algorithm to reduce cost and time. In this regard, a 2-level wavelet transform is applied and statistical features are extracted from wavelet subbands as texture features, and some statistical features from color information as color feature. Afterward, Principle Component Analysis (PCA) is employed. PCA is a dimension reduction algorithm, which can help to remove redundant features on feature vector, and finally a Multi-Layer Preceptron (MLP) neural network is used for training the system. The proposed method was evaluated on two databases and results are reported. Also in this paper, the dataset UTMC.V2.DB is introduced which is set of microorganisms images including the images of UTMC.V1.DB. The accuracy of this method is about %76 on UTMC.V1.DB and about %62 on UTMC.V2.DB.