Epilepsy is a common neurological disease, and its diagnosis mainly relies on the analysis of EEG signals. In recent years, deep learning-based methods have been widely used in epilepsy detection, but these methods usually rely on a single feature extraction technique and mostly ignore the spatial domain features of EEG signals. In order to capture the spatial domain features of EEG signals, researchers have tried to introduce the graph representation of EEG and combine it with GNN model for modeling. However, the graph representation of existing methods usually requires each vertex to traverse all other vertices to build the graph structure, resulting in high time complexity and difficulty in meeting the needs of clinical real-time diagnosis. In response to the above challenges, this study proposed CNG structure, which reduces redundant edges by dynamically selecting neighbor nodes, significantly reducing the time complexity while retaining key information. On this basis, we further proposed a dual-view input-based automatic epilepsy detection and classification framework, DV-SeizureNet. This framework can simultaneously learn the time, frequency, and spatial domain features of EEG signals to achieve epileptic abnormality detection and seizure classification. Experiments on the TUSZ dataset show that DV-SeizureNet achieves an accuracy of 91.4% in epilepsy detection tasks, which is 2.1% better than the existing state-of-the-art methods. In the classification task, the average classification accuracy of the model for four types of epileptic seizures is 82.8%, and the F1-score is 81.2%. DV-SeizureNet uses a dual-view learning framework to comprehensively extract and fuse the spatiotemporal and frequency domain features of EEG signals, and performs well in epilepsy abnormality detection and seizure classification tasks, providing a reliable auxiliary tool for clinical diagnosis.