Spatial–Spectral Diffusion Contrastive Representation Network for Hyperspectral Image Classification

Although efficient extraction of discriminative spatial–spectral features is critical for hyperspectral image classification (HSIC), it is difficult to achieve these features due to factors such as the spatial–spectral heterogeneity and noise effect. This article presents a spatial–spectral diffusion contrastive representation network (DiffCRN) based on the denoising diffusion probabilistic model (DDPM) combined with contrastive learning (CL) for HSIC, with the following characteristics. First, to improve spatial–spectral feature representation, instead of adopting the UNet-like structure, which is widely used for DDPM, we design a novel staged architecture with a spatial self-attention denoising (SSAD) module and a spectral group self-attention denoising (SGSAD) module in DiffCRN with improved efficiency for spectral–spatial feature learning. Second, to improve unsupervised feature learning efficiency, we design a new DDPM model with logarithmic absolute error (LAE) loss and CL that improve the loss function effectiveness and increase the instance-level and interclass discriminability. Third, to improve feature selection, we design a learnable approach based on pixel-level spectral angle mapping (SAM) for the selection of timesteps in the proposed DDPM model in an adaptive and automatic manner. Last, to improve feature integration and classification, we design an adaptive weighted addition module (AWAM) and a cross-timestep spectral–spatial fusion module (CTSSFM) to fuse timestepwise features and perform classification. Experiments conducted on widely used four hyperspectral image (HSI) datasets demonstrate the improved performance of the proposed DiffCRN over the classical backbone models and the state-of-the-art generative adversarial network (GAN), transformer models, and other pretrained methods. The source code and pretrained model will be made available publicly.