Abstract: In order to overcome the large error and poor repeatability of traditional dark field image processing, a dark field spectrum analysis algorithm is proposed. This algorithm uses Matlab as the platform to accurately identify gold nanoparticles and calculate the brightness through image expansion and other methods in a dark field environment. At the same time, it takes into account issues such as spectral standardization and removal of contaminants and impurities, and then reconstructs the single-particle scattering peak and records its spectrum characteristic. The experimental results of gold nanorod detection show that this algorithm is more flexible and accurate in identifying particles than the traditional method based on brightness binarization. The particle recognition rate can reach 91.33%, and the accuracy of particle fitting has increased by 15.59%. This algorithm can also detect the spectral characteristics of particles before and after the combination in batches. The data results are consistent with the theoretical values, and the data processing effect is good. It provides a more flexible, fast and accurate new algorithm for the field of dark field image processing.