Can I recognise the old and new classes on one re-trained image classification model?

[Vuforiator]

I’m completely new to the topic “Hello AI World/ Deploying Deep Learning” and perhaps my question will seem stupid, but I’ve been trying to find the answer for days, but none of those around me can answer. The problem I have is : I want to recognise my own ("new") images using pre-trained image classification model.

• I created my own datasets for several new classes, I classify the Raspberry Pi series microcomputers (B3, B3+, B4 etc) - OK;
• I used the train.py utility to retrain the one of the pre-trained image classification models and convert the result into ONNX-format – OK;
• I used imagenet.py with onnx-file for recognition my microcomputers and I can recognize every of RaspberryPis under cаmera with some confidence – OK
  But I cannot recognize the “old” classes for the pre-trained image classification model/ I used both ResNet18 and GooleNet – the problem is the same: after retraining, none of the converted models could recognize any other images except images of my objects, the network only worked with my new classes and did not work with its “old” ones (apple, watermelon, computer, etc.)
  My question: does it possible to use imagenet.py for recognize both “old” and “new” (my own) classes on one re-trained image classification model?