In the large-scale wireless sensor networks, which need to transfer large amounts of data, there must be some problems, such as, transmission reliability, congestion and energy consumption. High-efficiency data fusion technology can effectively solve these problems. In this paper, we combined the characteristics of clustering routing algorithm, and adopted two-layer data fusion technology. The first layer is about the data fusion between cluster node and cluster-head node. Cluster node determines whether to send data according to threshold. In terms of received data, cluster-head conducts consistency check to the data, excluding abnormal data. At second layer, BP neural network algorithm is adopted to fuse head-cluster node and base station. Experiment results show that data reliability is high after fusion, which greatly decreases transmission quantity and redundancy of data, reduce energy loss, and thereby improves the performance of entire network.