Caries is one of the most prevalent oral diseases worldwide. Terahertz spectral imaging technology has the advantages of strong broadband spectral analysis ability, high spatial resolution, and low ionizing radiation and is expected to be a new technical means for caries diagnosis. In this study, tooth slices containing dentin caries were used as the research object, and reflective terahertz spectral scanning was performed. The spectral data of the samples were reconstructed using two-dimensional imaging with the frequency-domain amplitude as the parameter, and several terahertz spectral images of caries were obtained at different frequencies. To solve the problems of small dynamic range, low contrast, and ambiguous edges, the detailed position of the terahertz image was obtained using a single parameter. Using the fusion method of wavelet gradient domain reconstruction, the larger gradient amplitudes of several images are concentrated in one image, and a new image with clearer and more complete detailed features is obtained. The experimental results show that the information entropy, average gradient, and contrast of the fused images are improved compared with those of the pre-fused images, and the discrimination effect between different tissues is more significant.