Abstract: Due to the problem of large amount of data collected by multi-sensors in wireless body area networks and redundant data types, it is difficult to effectively perform multi-dimensional data fusion.Although traditional manifold and decomposition data fusion methods (Isomap, MDS, PCA, etc.) have the ability to generate a reasonable rejection gradient at a small distance and are less affected by anomalies, the data dimension reduction effect for wireless multi-sensor volume domain networks is notideal.Therefore, this paper proposed the T-SNE algorithm based on manifold learning to fuse multi-sensor body area network data.The T-SNE algorithm first converts the Euclidean distance between a high-dimensional data point and its corresponding low-dimensional data point into a conditional probability matrix, then iterates the processed low-dimensional probability set a finite number of times, and finally updates the low-dimensional probability matrix to make the distance more reasonable exclusion gradient be generated between the small points, and a multi-dimensional body area network data fusion model is constructed.Experimental results show that compared with the traditional manifold reduction algorithm and traditional decomposition dimensionality reduction algorithm under a specific body area network data set, the T-SNE algorithm has a precision of 1.11 times that of Isomap, 1.33 times that of MDS, and 1.21 times that of PCA, achieves a better data dimensionality reduction effect.

Key words: Body field network, Data fusion, Manifold learning, Sensor