Multiplane light conversion-based multichannel photonic computing

We propose a multichannel optical computing platform utilizing multi-plane light converters to directly process complex RGB data, overcoming the limitations of traditional grayscale encoding. By encoding multichannel information in the first active region and weight patterns in subsequent regions, we achieve enhanced performance on both image classification and regression tasks. Our experiments demonstrate that this multichannel approach outperforms single-channel systems, highlighting its potential for multi-channel optical computing applications.