 This paper proposes a novel approach to EEG-based seizure subtype classification using a slim, deep neural network called EEGNet. The network incorporates a temporal information enhancement module with sinusoidal encoding to improve the accuracy of the first convolution layer. Furthermore, a training strategy for automatic hyperparameter selection was developed to optimize the network's performance. Experimental results on two datasets showed that the proposed method outperforms other traditional and deep learning methods in both cross, subject and transfer learning scenarios. Additionally, the code and CHSC dataset are made available to the research community. This article was authored by Rumien Peng, Chen Mingxiao, Jun Jiang, and others.