In this work, we present an extended study of image representations for automatic target recognition (ATR). More specifically , we tackle the issue of the image resolution influence on the classification performances, an understudied yet major parameter in image classification. Besides, we propose a parallel between low-resolution image recognition and image classification in a fine-grained context. Indeed, in these two particular cases, the main difficulty is to discriminate small details on very similar objects. In this paper, we evaluate Fisher Vectors and deep representations on two significant publicly available fine-grained oriented datasets with respect to the input image resolution. We also introduce LR-CNN, a deep structure designed for classification of low-resolution images with strong semantic content. This net provides rich compact features and outperforms both pre-trained deep features and Fisher Vectors. We also present visual results of our LR-CNN on Thales near-infrared images.