Remote sensing has been increasingly used in monitoring and analyzing agricultural activities. Crop identification based on analysis of individual pixel response, without considering neighboring pixels, may lead to poor results. In this paper, we investigate the use of superpixels generated by the Simple Linear Iterative Clustering (SLIC) algorithm to delineate homogeneous regions in images to help crop identification. The proposed classification strategy consists of combining pixel-level classification probabilities, estimated using the pixel spectral response, with pooled probability values of the pixels located inside superpixels. Weighting both probability contributions produce a hybrid classification. We test this idea to map the interim-harvest of corn and cotton in an agricultural area in Mato Grosso State, Brazil, characterized by the presence of large farms. For this, we use a cloudfree multispectral image captured by the wide-field imaging camera onboard the China-Brazil Earth Resources Satellite 4 (CBERS-4), that acquires four bands in the visible and near-infrared with a pixel spatial resolution of 64 m/pixel. The two main crops in our study area where identified with an overall accuracy of about 85%. Encouraging results suggest that the proposed method may be used as part of a remote sensing-based crop identification system.