On the network structure optimization problem of hyperspectral data depth mapping network, combined with the insufficient application of existing network structure to the learning and expression of hyperspectral line features, a depth kernel mapping network optimization method based on structure adaptation is proposed from the perspective of adaptive optimization of network structure. A hyperspectral data classification method based on multiple optimized depth kernel mapping network is presented. Firstly, the kernel structural and learning parameters of the depth kernel mapping network are optimized. This method can adaptively adjust the network structure according to the distribution characteristics of hyperspectral data and improve the performance of hyperspectral image classification. Secondly, the accounting sub and network node optimization methods proposed above are applied to the network to improve deep kernel learning mapping network in hyperspectral image classification. The experimental results show that improving the depth kernel mapping network from the perspectives of kernel operator, mapping network node, and network structure can effectively improve the feature extraction and classification performance of hyperspectral data.