The idea is fairly straightforward. Once you have detected the object, you crop the image using the bounding box from the detection model. Then you can use an OCR technique such as tesseract to to read the text from the cropped image.