How to perform image recognition using a pre-trained neural network?