Facial emotion recognition plays a crucial role in human-computer interaction and psychological research. Early emotion recognition techniques using visible images could be easily tampered as emotion can be faked on the physical level. So, thermal imaging-based methods were considered to capture the natural and spontaneous intensity of emotions. Currently, only a handful of research studies are being performed using thermal cameras to detect emotions. This article proposes a deep-learning approach to identify and classify seven basic human emotions from the KTFEv2 thermal dataset, a novel version of the original KTFE. The results obtained by our method surpass the current existing work on this dataset in terms of accuracy, precision, f1- score and support. The overall accuracy achieved was 84.12%.