In recent years, Convolutional Neural Network (CNN) based frameworks are being applied to polarimetric synthetic aperture radar (PolSAR) image classifications and achieved improved results. However, the performance of CNN for PolSAR image classification is greatly dependent upon the selection of polarimetric features. Although CNN automatically extracts abstract high level features from the data, it is still beneficial to incorporate additional hand-crafted features to enhance the classification results. In this research, to incorporate spatial information of the pixels in classification process, first, a profile of the PolSAR image is constructed by using superpixel algorithm. Then, the constructed superpixel profile is fed into a CNN model for classification. The experiment conducted on three real PolSAR datasets highlights the utility of superpixel profiles. For all the three datasets, our proposed method demonstrates a consistent improvement of at least 3% in classification accuracy in comparison to the state-of-the-art CNN model.