Existing classification and identification models for transgenic rapeseed oil based on single-spectrum analysis suffer from limited information and high data dimensionality， leading to low operational efficiency and inaccurate detection results. To address these issues， a novel classification method for transgenic rapeseed oil was proposed， leveraging terahertz fusion spectroscopy combined with an improved Fused Lasso model. Using two types of transgenic rapeseed oil and two types of non-transgenic rapeseed oil as research subjects， the terahertz time-domain spectroscopy （THz-TDS） system was employed to obtain the terahertz absorption spectra of the four rapeseed oil samples in the frequency range of 0.2 to 1.6 THz. Features were extracted from the absorption and derivative spectra using the successive projections algorithm （SPA）， and then fused. Introducing a regularization sparse model， Fused Lasso， which integrates feature selection and classification. This model was improved into a multi-class model using the one-vs-one （OVO） method， and Bayesian optimization （BO） was employed to optimize its regularization parameters. The results demonstrated that the BO-Fused Lasso model， based on fusion spectra， significantly outperformed the traditional Fused Lasso model based on a single absorption spectrum in classifying the four types of rapeseed oil. The accuracy rates for the training and testing sets were 96.88% and 95.00%， respectively. This study， therefore， presents a novel approach for accurately identifying transgenic and non-transgenic rapeseed oils and provides a valuable reference for the detection of other transgenic substances.