Hyperspectral Unmixing using Convolutional Autoencoder for Metal Detection in Lithium-Ion Battery Recycling Applications

Recent advancements in hyperspectral imaging systems have opened up possibilities for identifying and distinguishing materials based on their spectral characteristics, as every material has its unique spectral signature. In our work, we present a novel approach for detecting and distinguishing copper and aluminum foils present in shredded lithium-ion batteries (LIBs) using convolutional autoencoder for hyperspectral unmixing. In hyperspectral applications, unmixing is a key procedure for estimating spectral signatures of pure materials (endmembers) as well as the corresponding fractional spatial extent (abundances) of endmembers in mixed pixels of hyperspectral images (HSIs). We perform hyperspectral unmixing on a real hyperspectral dataset using a convolutional autoencoder with sparse regularization. We evaluate the performance of the autoencoder framework using VNIR (visible and near-infrared) HSI data acquired with the Specim FX10 hyperspectral sensor. Our experimental unmixing results demonstrate that convolutional autoencoder showed a significant improvement in unmixing performance compared with competing unmixing methods. To the best of our knowledge, this work is the first to implement hyperspectral unmixing using autoencoder in LIB recycling, which is highly significant for automated sorting of valuable metals in LIB recycling industrial applications.