Computer Vision Algorithms for Microscopy Classification of Nanocomposites Using a Federated Machine Learning Approach

The work solves the problem of classifying nanocomposites using computer vision technology in conditions where microscopy images obtained for training cannot be used centrally for training for various confidentiality reasons. To solve this problem, a computer vision system is implemented and an approach based on federated machine learning is presented, allowing the system to be trained locally considering data confidentiality criteria. The computer vision system is based on a modified convolutional neural network. A horizontal federated machine learning architecture is used to ensure data privacy. A comparison was made of this system with classical machine learning methods. The accuracy of the model was 93.4% according to the accuracy metric.