Objective Brain network analysis is a powerful tool to study the topological properties of epileptic patients, however traditional feature fusion methods of brain networks ignore interactions between features. To address this issue, a feature fusion method for brain networks based on Discriminant correlation analysis (DCA) was proposed to improve the classification of patients with negative temporal lobe epilepsy. Methods The resting-state functional magnetic resonance (rfMRI) data of 20 patients with negative temporal lobe epilepsy and 20 healthy people were preprocessed, and a functional connectivity matrix based on Pearson correlation was constructed with AAL90 template; The sparsity of the constructed brain function network ranges from 0.08 to 0.4 with a step size of 0.01; Five local topological properties were calculated and the area under the curve (AUC) was used as a metric; The minimum redundancy maximum correlation (mRMR) algorithm and sequence search algorithm were combined to construct the optimal subset of five local topological properties; DCA was used to achieve feature fusion. KNN was used as a classifier and leave one cross validation (LOOCV) was used to evaluate the classification results; Results The classification accuracy of the feature set fused by DCA is 95%, which is 5% higher than that of the traditional series combination feature set, and 5% to 10% higher than that of the single topology feature set. Conclusion The feature fusion strategy proposed in this paper effectively combines the information of different topological properties, improves the classification performance, and provides a new method for the diagnosis and detection of negative temporal lobe epilepsy.