Research on recognition algorithm of impact vibration based on HOG feature of time–frequency spectrum

Abstract

The time–frequency spectrum of the vibration signal obtained based on the time–frequency transformation algorithm can fully reflect the energy distribution at any time–frequency coordinate. But this characteristic is not easy to be insight and generalized by the human eye, especially in the era of big data, in the face of massive data and rapid processing. During the research, we found that this information can be clearly expressed in the histogram of oriented gradient (HOG) feature of the time–frequency spectrum. Therefore, the time–frequency spectrum support vector machine classifier based on the HOG feature of the vibration signal has considerable feasibility. This article had conducted in-depth research on this idea, determined the advantages of wavelet time–frequency spectrum for signal recognition, and further selected wavelet basis and kernel function. A comprehensive test of the HOG parameters had finally achieved a good recognition effect. The comprehensive recognition accuracy of the algorithm for the impact and vibration signals collected in this paper was stable at about 96%. The research results will be applied to the future smart city vibration monitoring system. On the other hand, it may provide a new idea to improve the accuracy of the current battlefield target recognition system.