GhoMR

GhoMR introduces an approach to hyperspectral image (HSI) classification, leveraging the power of convolutional neural networks (CNNs) while addressing the challenge of network heaviness due to the large number of receptive fields (RFs) and associated weights. HSIs, characterized by their vast number of spectral bands, offer rich information for computer vision applications, particularly in remote sensing. However, the complexity and size of HSI data pose significant challenges for analysis, especially with deep learning techniques such as CNNs, which typically require substantial computational resources.

GhoMR innovatively incorporates multi-receptive CNN modules, utilizing blocks with various RFs to extract features in a residual manner. This hierarchical feature extraction process allows for capturing complex features from HSIs. To counterbalance the potential heaviness of networks caused by multiple RFs, GhoMR employs the Ghost module—a recent development in CNN design that minimizes feature redundancy by extracting a limited set of features and applying cost-effective transformations to them. This strategy significantly reduces the number of parameters in the network, making it more efficient without sacrificing performance.