This paper proposes a novel extension of morphological attribute profiles (APs) for classification of remote sensing data. In standard AP-based approaches, an input image is characterized by a set of filtered images achieved from the sequential application of attribute filters based on the image tree representation. Hence, only pixel values (i.e. gray levels) are employed to form the output profiles. In this paper, during the attribute filtering process, instead of outputting the gray levels, we propose to extract both statistical and geometrical features from the connected components (w.r.t tree nodes) to build the so-called feature profiles (FPs). These features are expected to better characterize the object or region encoded by each connected component. They are then exploited to classify remote sensing images. To evaluate the effectiveness of the proposed approach, supervised classification using the random forest classifier is conducted on the panchromatic Reykjavik image as well as the hyperspectral Pavia University data. Experimental results show the FPs provide a competitive performance compared against standard APs and thus constitute a promising alternative.