Nowadays, classification of satellite images is one of the main subjects discussed and studied in remote sensing topics. POLSAR images are one of the types of SAR images that due to the use of different polarizations (horizontal and vertical), these images have potential to extract more accurate information from the scattering mechanism of the desired environment. However, this can also make feature extraction and classification operations more complicated. In this paper, convolutional neural networks (CNNs) are used to extract features and classify images. In addition, by using morphological profiles and one of target decomposition methods called H / A / alpha decomposition, we try to extract the spatial and scattering features separately and stack them with the main POLSAR data, to use as input to the convolutional network. Also we use the SVM classifier to increase the limited training samples through obtaining an initial classification map. The result is improvement of classification.