Conventional THz spectral imaging enhancement methods mainly use Local Linear Embedding (LLE) local linear embedding algorithm to complete complex high latitude projection, which is vulnerable to the influence of local distance mapping relationship of samples, resulting in high Root Mean Square Error (RMSE) root mean square error. Therefore, a new Terahertz (THz) spectral imaging enhancement method needs to be designed based on Graph Neural Network (GNN). That is, the terahertz spectral imaging image is preprocessed, the terahertz spectral imaging enhancement model is constructed using graph neural network, and the terahertz spectral imaging enhancement algorithm is designed to achieve terahertz spectral imaging enhancement. The experimental results show that the designed THz spectral imaging enhancement method based on graph neural network has low RMSE root mean square error of different images after enhancement, which proves that the designed THz spectral imaging enhancement method has good enhancement effect, reliability, and certain application value, and has made certain contributions to meet the complex and changeable spectral imaging application scenarios.