The widespread deployment of wireless sensor networks and the burgeoning internet of things (IoT) are enabling devices to be connected in wider and denser ecosystems in myriad of applications, making wireless security of paramount importance. Radio Frequency (RF) fingerprinting has the potential to enhance the security and with increasing popularity of deep learning, RF fingerprinting approaches have attracted increased attention. In this paper we propose a graphical deep learning approach for RF fingerprinting which can be paired with either convolutional or dense neural network architectures for device identification. With the proposed approach captured raw sample sequences are first transformed into image inputs, which are then used to train and test the deep learning models. The use of images, as opposed to raw samples sequences, provides a compact way to capture signal characteristics without impacting the structure and the complexity of the subsequent deep learning processing. We evaluate the performance using a two-stage process consisting of coarse and fine evaluation, carried out using synthetic and over-the-air captured datasets respectively. The coarse evaluation enables us to swiftly verify the performance without involving hardware implementation for data collection, whereas signals captured over-the-air are used for fine evaluation, with the proposed approach achieving near 100% identification accuracy.