Abstract: The integration and compound analysis of multi-resolution remote sensing image is one of the most important techniques in remote sensing image processing. For large scale land covering classification, it is common that low resolution data leads to worse performance due to the mixed-pixel problem, and high resolution data with wide covering range has more limitations such as long period of acquiring cycle, high data and processing cost. Facing these problems, an enhanced new compound classification method based on multi-kernel non-linear regression model was proposed, which could improve the description abilities of the traditional compound classification method based on linear regression model. Then, some texture information was introduced to improve the classification accuracy further more. Finally, the classification accuracies of the two regression models were compared through real data experiments based on the ground truth. The experimental results show that the classification accuracy is greatly improved by using this new method, and can be further improved with the texture information.