Multi-scale spatial and spectral feature fusion for soil carbon content prediction based on hyperspectral images

Soil carbon content prediction based on hyperspectral images can achieve large-scale spatial measurement, which has the advantages of wide coverage and fast information collection, and is more suitable for field data collection. However, existing studies on soil carbon content prediction using hyperspectral images mainly focus on feature extraction of spectral information, while ignoring spatial information, and cannot well reveal the intrinsic structural characteristics of the data. Aiming at the lack of spatial features consideration, this paper proposes a multi-scale feature fusion method for soil carbon content prediction from hyperspectral images. While extracting spectral features, spatial information of hyperspectral images is introduced for the first time, and a multi-scale spectral and spatial feature network (SpeSpaMN) is designed. In SpeSpaMN, a multi-scale spectral feature network (SpeMN) is constructed to extract spectral features, and a multi-scale spatial feature network (SpaMN) is built to extract spatial features. The two networks are fused by using the complementary relationship between different scale features to realize soil carbon content prediction based on multi-scale feature fusion. The results showed that SpeSpaMN achieved the best prediction performance, followed by SpeMN. The RPD of Inland, Aoshan Bay and Jiaozhou Bay samples based on SpeSpaMN were increased by 47.36%, 37.96% and 4.30% respectively compared with the full-spectrum method. This method can effectively solve the problem of deep fusion of spatial and spectral features in soil carbon content prediction using hyperspectral images, so as to improve the accuracy and stability of soil carbon content prediction.

Ecological Indicators