Signal processing methods of multi-sensor based on cross sensitivity repression

Multi-sensor signal detection methods are generally adopted in modern testing system. However, most sensors are sensitive to several parameters, and this phenomenon is called cross sensitivity. Consequently, signal processing technique needs to ensure the precise measurement by restraining cross sensitivity. Corresponding signal processing of multi-sensor realizes the regression procedure from output signals to input signals, namely constructing the inverse model. Considered the disadvantages of traditional reconstruction methods, three improved algorithms are proposed forward: Based on the local linearization strategy and Stone-Weierstrass theorem, Total Least Squares represses cross sensitivity by considering the noise of both input and output signals; Moving Least Squares can accomplish the restraining of cross sensitivity efficiently through researching the construction method and characters of interpolated function with selecting basis function and weight function reasonably; In case of small sample data, traditional methods will encounter the problem of overfitting and poor generalization capability, while structural risk minimization-based Support Vector Regression can reconstruct the input signals of multi-sensor accurately. The emulation results and theory analysis indicate that the proposed algorithms are more accurate and reliable for signal processing.