It is important to detect cow oestrus in time and accurately. According to the characteristics of increasing activity, shorter repose time and increasing animal heat during cow oestrus, vibrational sensor and temperature sensor were used to detect the above parameters. LVQ neural network model was built, which taking walking steps, repose time and animal heat as input, and behavior characteristics as output. The results showed that, with the proposed algorithm, the correct detection rate was up to 100%, and prediction rate for oestrus was more than 70%.