Diabetic foot disease is the main cause of disability and death in diabetic patients. Infrared thermography can detect the metabolic differences caused by peripheral neuropathy and vascular dysfunction in the early stage of diabetic foot, which is an effective means to achieve early screening of diabetic high-risk foot. In this study, a feature fusion-based early screening method for the diabetic foot was designed, and a multi-scale temperature feature fusion module was proposed to enhance and fuse foot features of different scales. Furthermore, we achieved automatic segmentation of plantar infrared heatmaps by segmenting color image-registered infrared heatmaps and enhanced the discriminability of image features using JET mapping and gamma correction. The accuracy of this method is 97.66% and 97.01% on the experimental dataset and the public dataset, respectively, which shows the advancement of this method.