Given the rise of social media, detecting social bots is crucial yet challenging. Despite state-of-the-art graph-based methods, bots can mimic real users by following many authentic accounts, significantly undermining the effectiveness of detection efforts. To tackle this, we propose NDE-GNN, a Neighborhood Difference-Enhanced Graph Neural Network based on Hypergraph. Utilizing hypergraphs, we extract user neighborhood representations, differentiate them with node features for distinct characteristics, and enhance these via Graph Neural Networks. Enhanced hypergraph features are fused with original graph features using a self-attention mechanism and feature fusion module. Fused features pass through a linear layer and softmax for detection. Experimental validation on real-world Twitter bot datasets confirms the superiority of NDE-GNN.