在训练好的softmax分类模型上检测多个对象

I have used the standard SVHN cropped digits dataset to generate a model that classifies into the 10 possible digits, with accuracy of 89.89% on the testing set. Moving on, I would like to detect multiple digits on an image. (e.g. the numbers on a car registration plate) How would I go about doing this? Would I need to retrain my model to detect multiple images? I have used the standard SVHN cropped digits da