Potato is among one of income-generating crop grown in Ethiopia, especially in North west Amara regions. Selecting and planting clean and uninfected potato seed is an important activity before distributing and planting. Meanwhile, it will increase the production and quality yield of potato crop. Bactria Wilt(BW) was one of most commonly disease that affect the Ethiopian seed potato specially in North west Amhara zones. There were limited studies on globally on potato tuber and those studies were focused on Europe not localized on East African potato tuber disease identification problems. Those studies were focused on Europe and almost all studies were focused on potato plant and leaf disease identification problems. Thus, in this thesis we are presented a Convolutional Neural Network system for the automatic detection of seed potato disease focusing East African seed potato types. Our proposed system has five components, preprocessing, segmentation, augmentation, feature extraction and classification. In preprocessing stage, we normalize the image into standardize size of 224x224 pixel and converting RGB channel image into Gray Scaled image. Segmentation is used for extracting Region of Interest(ROI) we used canny edge segmentation techniques to detect the infected edges of the infected seed potato disease. For feature extraction we used Conventional Neural Network(CNN) to detect the important features of the infected disease and for classification we used softmax classifier to classify Bacterial Wilt(BW) and Clean potato types of class. Our Proposed system is implemented on Google Colab Notebook environment using Keras and tensor flow as a back end libraries using python as a programming code editor. Our sample images were collected from F/selam and Awi zones from World Vision Ethiopia Area Programs office seed stocks. The model achieved a diagnosis accuracy of 99.54% accuracy for the training data set. And 98.27% test accuracy. Our model is better performance than the state of the art models and its performance is improved using data augmentation techniques and reduced overfitting problem of the model. Keywords: Bacteria Wilt, CNN, Segmentation, Google Colab, RGB