There are still quite a few problems in river ice environment monitoring by remote monitoring network, such as the increase of the various sensor nodes and network capacity overload due to the dense deployment, and the decrease of the life expectance of the network. In view of this, the improved K-means data fusion algorithm was proposed to fuse a variety of data by a novel self-developed sensor. Furthermore, the resilient BP network was employed to build a data model database, so as to identify the overall river ice environment at the back-end remotely in real time. Simulation verification and comparison test results show that the proposed algorithm can achieve efficient and accurate data transmission by various sensors and the existing sensor network during the real-time monitoring of the river ice environment.