In order to improve the precision of grassland classification from remote sensing images, we analyze the characteristics of image features extracted from convolutional neural networks (CNNs), and propose a remote sensing image feature extraction method based on feature-integrated depth neural networks. Firstly, PCA whitening is performed on the remote sensing image to reduce the correlation between data and accelerate the learning rate of neural networks. Secondly, both low-level features and high-level features are bilinearly integrated to enhance and optimize the integrated features. Finally, the remote sensing data is trained. As the introduction of effective information in new features, both feature expression ability and the grassland classification accuracy are improved. Experimental results show that the proposed algorithm can effectively improve the accuracy of grassland classification. The classification accuracy reaches up to 94.65%. Compared with the traditional convolutional neural network, BP neural network and SVM algorithm, our accuracy is increased by 4.3%, 10.39% and 15.33% respectively.