Traffic flow monitoring is essential to take effective measures on traffic jams. Wide-ranging monitoring is still a challenging task, because the conventional observation methods by roadside detectors only observe the location where the costly sensors are installed. In this study, we propose a wide-ranging traffic flow monitoring method utilizing remote sensing by small satellites, which is a new approach of remote sensing. Compared with the conventional remote sensing satellites, small satellites have a significant advantage: they can conduct frequent observation (e.g., few hours) by using dozens of satellites simultaneously. On the other hand, a severe disadvantage is that their sensor is inferior due to the size limitation, and thus existing vehicle detection methods are not applicable to small satellite imagery. In order to solve this issue, we develop a convolutional neural network-based image recognition algorithm that estimate traffic density. The proposed method is applied to satellite imagery with 93 cm resolution, which corresponding to small satellite imagery. The accuracy of the proposed method was RMSE 11 veh/km, sufficient to recognize traffic situations.