Diabetic foot ulceration (DFU) is a severe and prevalent complication of diabetes mellitus, leading to high rates of amputation and mortality. Abnormal plantar foot temperature changes are an early sign of diabetic foot ulcer that can be detected using a thermal camera. This study presents a comprehensive approach to diabetic foot detection using thermal imaging and advanced machine learning techniques. Broadly, the proposed approach comprises a UNet-MobileNetV2 based deep learning architecture to segment the foot region followed by classification of the foot thermal patterns using thermal radiomics. Our approach was evaluated using five-fold cross-validation on both the datasets, addressing the limitations of previous studies that relied on a single dataset. The proposed segmentation model achieved Dice and Intersection over Union (IoU) scores of 97.1% and 96.34%, respectively. Using an XGBoost classifier, the proposed radiomics resulted in Area Under Curve (AUC) of 96.42% ± 1.4%, 96.76% ± 2.64%, and 92.18% ± 4.93% on Plantar, Standup, and Combined datasets, respectively. The consistently high AUCs across these datasets demonstrate the robustness of the proposed thermal radiomics approach.