Abstract: Emission factors are an effective means to quantify pollutants emissions from road mobile sources and to predict emissions at a certain time or area. To address the drawback that the COPERT model cannot be used directly on on-board diagnostic(OBD) data to obtain accurate emission factors, we propose a two-stream network based on the fusion of time-series features and time-frequency features to amend the COPERT model. First, to gauge the instantaneous emission factors for nitrogen oxides (NOx)-emission by heavy-duty diesel vehicles, data gathered from multiple driving segments, which were collected in the course of actual driving, were used; from these data, we select the monitored attributes having a high correlation with the emission factors of NOx, and deploy Spearman rank correlation analysis based on the data volume to obtain the final dataset constituted by these attributes and emission factors; then, on the final dataset, a historical information matrix is constructed to ascertain the influence of historical information on emission factors, each time series attribute is converted into a time frequency matrix using the continuous wavelet transform (CWT), and a multi-channel time frequency matrix is formed by superimposing the time frequency matrix with each attribute of this historical information matrix; finally, to complete the fusion of time-series features and time-frequency features, the historical information matrix and the time-frequency matrix are input into the two-stream parallel model comprising the ResNet50 and convolutional block attention module (CBAM) modules. The reliability and accuracy of the proposed method in obtaining emission factors on the OBD dataset after modification of the COPERT model were verified by comparing with the current mainstream models.