The interpretation method of GPR (Ground Penetrating Radar) detection image by human eye is easily affected by the subjectivity and experience of interpreters. In order to avoid this shortcoming, a spectral energy interpretation method based on Counterlet contour transform and K-Means++ clustering analysis is proposed. Based on actual highway tunnels, the original detection data of unfavorable geological bodies were obtained through on-site detection. IDSP detection data analysis software was used to generate the original image, and time domain and frequency domain preprocessing such as background removal and filtering were implemented to improve the signal-to-noise ratio. Based on subband distribution coefficient, Counterlet was used to decompose and reconstruct the preprocessed image and K-Means++ algorithm was used to convert the frequency information in the reconstructed image into color features. The color features were extracted by MATLAB, and a sample library of color features of unfavorable geological bodies was established accordingly. The original detection image was matched and compared with the sample library to realize automatic interpretation. Facts and figures show that it is feasible to decompose and reconstruct multi-directional, multi-resolution and multi-scale GPR images by using Counterlet contour transformation, the curve edge approximation effect is good and the reconstructed images have no information loss. K-Means++ algorithm can realize the conversion of energy to frequency to color in the gray-scale image of GPR, and the converted image has prominent and intuitive color. The matching and comparison of spectral energy can realize automatic interpretation more accurately and quickly and avoid individual subjectivity better.