© 2022, The Author(s), under exclusive licence to Springer-Verlag London Ltd., part of Springer Nature.Learning an object category from only a few samples is generally inadequate for the correct classification of large-scale problems. One needs many training samples to obtain a classifier that generalizes well and has reasonable success rates for an operator. However, collecting many annotated data is not always possible in several applications, including target detection from ground penetrating radar (GPR) data. Furthermore, GPR images of target or clutter objects show a nonlinear dependence on soil permeability and permittivity. Therefore, even if enough training data were available to train a good classifier for one soil type (such as dry sand), the success of this classifier does not translate well if the soil type is changed (say, to wet sand). To decrease this domain gap, in this work, we propose to do a multi-model knowledge transfer (KT) for improvised explosive device detection from GPR data and investigate how effective it is to pass the models learned from known environments to models trained for unknown environments. We show that (1) knowledge transfer from multiple sources (i.e., multiple types of sand) generates better results than single-source transfer, and (2) as little as three training data from the unknown source increases the detection rates by 10% for single KT and 4% for multiple KT on simulated data. Further, we show that adapting the models for each type of source is a better approach than just combining all the training data in a single model. These results demonstrate that a multi-source domain adaptation approach significantly reduces data collection and manual annotation efforts and increases detection rates in unknown environments.