Automatic aircraft type recognition from remote sensing images can give critical information for rapid battlefield analysis, military decision-making and civil applications such as emergency airplane searches. It is challenging to collect labeled data for deep learning-based type recognition systems because the bottleneck for such systems is the necessity of enormous volume of labeled data. Labeling types of aircraft images is a procedure that must be carried out by field specialists and requires a significant amount of time when performed to a large number of images. Due to the lack of benchmark datasets up till 2020, there is a substantial research gap in aircraft type recognition in remote sensing, which hinders the advancement of research in the field. The intent of this thesis is to build a framework based on deep active learning, explored for the first time in the field, that achieves better accuracy performance than the previous research studies. This thesis contributes vastly to tackling annotation issue for aircraft type recognition in remote sensing images for the first time, where the framework achieves high accuracy by using the fewest annotated samples possible, unlike previous studies. Moreover, this thesis adds to the field by constructing a new dataset that is considered the largest yet to be reported for this field with a total of 50 aircraft types with 22,233 images. The dataset constructed, which is the largest and most challenging dataset is created by integrating two most recent datasets that had not yet been investigated in the field using image processing approaches. Furthermore, this thesis provides a strong base upon which future research may be constructed by exploring and evaluating two brand-new publicly accessible datasets for the first time. This is accomplished by analyzing the results of the top competent deep learning algorithms on these datasets.