How to fine-tune deep neural networks in few-shot learning?

Segmentation overlapping wear particles with few labelled data and imbalance sample.

Ferrograph image classification.