Target classification of imagery from Unmanned Aerial Vehicles (UAVs) is increasingly important for military reconnaissance and surveillance. A promising technique to improve target classification of UAV imagery is Deep Neural Networks (DNNs). However, DNNs may consist of a very large number of parameters, which makes it difficult for operators to understand what image features DNNs use for target classification. This lack of transparency is a challenge for military applications of DNNs for target classification since operators are ultimately responsible for all decisions due to the high risks of weapon engagements. Operators therefore also need explanations of DNN classifications to assess their reliability. This report describes an experiment where participants performed a target classification task of military vehicles in low-altitude UAV imagery. The objective of the experiment was to evaluate whether support of DNN classifications and support of saliency map explanations of DNN classifications, which highlight the most important features for DNN classifications, improve accuracy in target classifications. Saliency map explanations were generated with the Randomized Input Sampling for Explanation (RISE) method. Participants performed the target classification task in three different conditions: without support of DNN classifications, with support of DNN classifications, and with support of RISE saliency map explanations of the DNN classifications. The results show that, contrary to expectations, participants' accuracy in target classification decreases with support of DNN classifications and it decreases even further with support of RISE saliency map explanations. Participants' lower accuracy in target classification with support of DNN classifications and RISE saliency map explanations is likely due to a combination of two reasons: reliance on automated decision aids and difficulty in assessing DNN reliability. The results show that participants under-rely on DNN classifications when they are correct and over-rely on DNN classifications when they are incorrect. The conclusion of the experiment is that it is not trivial to present DNN classifications and explanations of DNN classifications that actually support operators' target classification. Additional experiments are required of how to present information from DNN classifications and whether other promising XAI-approaches improve operators' target classification.