The invention discloses a high-accuracy detection method of real-time temperature data fusion of multiple sensors, which comprises the following steps of: step 1, remissness error removal based on relativity function sequence: (1) acquiring temperature data, (2) judging whether the number of acquisition times is larger than or equal to a set value C1 or not, if NO, continuing to acquire the temperature data, if YES, judging whether the number of the acquisition times is larger than the set value C1 or not, if NO, computing the variance and the mean value initial value of sensors, if YES, computing the variance and the mean value of the sensors in a recurrent way, (3) computing the fusion degree of the sensors, (4) computing the support degree of the sensors, and (5) selecting valid sensors, and deleting the temperature data obtained from invalid sensors; step 2, orthogonal neural network based temperature information fusion of the multiple sensors: (1) neural network training and weight vector recurrence, (2) computing neural network output, and (3) computing a temperature fusion value of the multiple sensors; and step 3, correcting variance and mean value of the invalid sensors. By the method, not only can the temperature detection accuracy and credibility be improved, related systems can be conveniently processed in real time. The method has the advantages that the stability is good, the compute is simple, the implementation is easy, and the like.