Research on autonomous waste detection is primarily focused on conveyor belt systems. Large objects are typically shredded to fit within a conveyor belt system. This work investigates material detection in bulky waste before it is processed by shredders, as sorting large objects before shredding has the potential to improve the recycling process. Multispec-tral cameras are employed to capture high dynamic range images across the ultraviolet, visible, near-infrared, and shortwave infrared spectra. Deep learning techniques are applied for pixel classification and patch segmentation. We evaluate our approach on a small laboratory dataset consisting of 17 images. The results demonstrate that the multispectral imaging approach outperforms RGB-only imaging, achieving a 10% higher accuracy. Furthermore, the study demonstrates that spectral and spatial convolutions enhance the performance of material detection.