Our train model is only about 0~9 digits. we need to rebuild the model adding blank picture cases to improve image recognition.