The invention relates to the field of data fusion, and provides a method and device for measuring longitudinal cylindrical grinding surface roughness through a data fusion technology. According to the technical scheme, the method includes the steps of firstly, designing an orthogonal grinding test, wherein control parameters are grinding depth, workpiece rotating speed and moving speed of a working platform, and workpiece vibration signals and workpiece surface roughness need to be respectively measured; secondly, conducting the grinding test, wherein the test is conducted according to the scheme in the first step, the control parameters and a parameter measurement result are recorded, and the workpiece vibration signals are transmitted to an industrial personal computer through an eddy current sensor; thirdly, training and verifying a neural network, wherein a neural network model of the surface roughness is built, input parameters are the grinding depth, the workpiece rotating speed, the moving speed of the working platform and the workpiece vibration signals, output parameters are the workpiece surface roughness, and then test data are utilized to train and verify the network; fourthly, measuring the surface roughness, wherein after the neutral network is verified, the grinding surface roughness under a same process system can be online measured.