The keyphrase generation task aims to produce a series of representative keyphrases that encapsulate the core themes of a text. Text can be modeled as either a sequence or a word graph, each highlighting features of the text from distinct perspectives. However, existing methods typically employ either sequence-to-sequence or graph-to-sequence frameworks, focusing solely on capturing either sequence or word graph features, but fail to combine both features for improving keyphrase generation performance. In this study, we focus on enhancing keyphrase generation by integrating features from both the text sequence and the word graph structure. Specifically, we introduce SGKG, short for Sequence and word Graph feature fusion model for Keyphrase Generation. During encoding, the sequence encoder captures sequential features, while the graph encoder learns structure features. During decoding, the features from both encoders are integrated to generate keyphrases through the fusion generation mechanism and the multi-source copying mechanism. Experimental results demonstrate that SGKG outperforms existing methods, achieving significant improvements in keyphrase generation.