【摘要】 高原鼠兔的行为分析可为高原鼠兔的生态学研究与鼠害的防治提供依据。为提高高原鼠兔短期行为预测精度,通过序列图像的时空特征,定义了高原鼠兔的基本行为模式集合;依据自然生态环境下每一个行为模式都会持续一段时间的原则,对视频序列进行降采样从而获得行为时间子序列,以减小行为预测的数据处理量;采用小波神经网络方法建立高原鼠兔短期行为预测的数学模型,确定小波神经网络的参数,用所设计的小波神经网络对高原鼠兔的短期活动行为进行预测。试验结果表明:相对于BP神经网络,小波神经网络的预测准确率高,达到94.94%。预测误差小,均方误差仅有0.012 9。采用小波神经网络可对高原鼠兔的短期行为进行预测。该研究可为高原鼠兔的行为研究提供一种新的信息化思路。更多还原

【Abstract】 Behavior analysis can provide references for ecology research and rodent control for Ochotonacurzoniae.In order to improve the precision of short-term behavior prediction,basic behavior patterns were defined based on the time and space characteristics of sequence images.According to the persistence of behavior patterns,video sequence was down-sampled to obtain time sub-sequence,thus,the data needed to be processed can be reduced.Short-term behavior time series prediction model for Ochotona curzoniae was established with wavelet neural network,which can be used to forecast the behavior of Ochotona curzoniae with determined parameters.The results show that wavelet neural network has better prediction precision for Ochotona curzoniae than BP neural network,which can reach 94.94%,and cause less prediction error,the mean square error is 0.012 9.This study can provide a new and informatization thinking for behavior researches of Ochotona curzoniae.更多还原