Hyperspectral imaging offers a way for computer vision to surpass human visual perception by increasing the available spectral information beyond RGB images. The requirement of specific cameras and difficulty of capturing hyperspectral images has led to a scarcity of data, with most belonging to the aerial remote sensing paradigm. In this paper, we present a novel and extensive hyperspectral dataset of construction debris for recycling applications that includes objects from 14 different material classes, measured by three different hyperspectral sensors. We compare a variety of hyperspectral image classification approaches and demonstrate that relative performance of common hyperspectral models differs for our new dataset. Furthermore, we demonstrate a positive effect of pre-training on our dataset for other similar hyperspectral image classification tasks.