The application of 3D Octave convolution model in high spatial-hyperspectral image classification can improve the accuracy of multi-tree species classification tasks， which is of great significance to improve the refinement level of forest management. A 3DOC-SSAM model combining 3D Octave convolution and attention mechanism is designed. Through 3D Octave convolution and spatial-spectral attention mechanism， the operation efficiency and classification performance of the model are improved. The results show that ： （ 1 ） The overall accuracy of the 3DOC-SSAM model reaches 99.53 %， which is 13.86 %， 18.49 %， 12.90 % and 5.36 % higher than that of SVM， ELM， 2D-CNN and 3D-CNN， respectively. The average accuracy AA reached 99.38 %， and the Kappa coefficient reached 0.994 7. （ 2 ） In the case of small sample training， the overall accuracy and average accuracy can still reach 96.9 % and 95.52 %， which is higher than the comparison model. The research results provide an efficient and high-precision solution for multi-tree classification tasks， and have broad application prospects in forestry remote sensing， which is helpful to improve the scientificity and sustainability of forest resource management.