地理传感器时间序列具有复杂动态的语义时空相关性和地理时空相关性。尽管已经开发了各种深度学习模型用于时间序列预测,但很少有模型能专注于捕捉地理传感器时间序列内的多类型时空相关性。此外,同时预测多个传感器在未来某一时间步的...地理传感器时间序列具有复杂动态的语义时空相关性和地理时空相关性。尽管已经开发了各种深度学习模型用于时间序列预测,但很少有模型能专注于捕捉地理传感器时间序列内的多类型时空相关性。此外,同时预测多个传感器在未来某一时间步的值非常具有挑战性。为了解决上述问题,提出了一种自回归模型与深度神经网络的联合模型(Joint model of Autoregression and Deep Neural Network,J-ARDNN),用于处理地理传感器时间序列的多目标预测任务。在该模型中,空间模块用于捕捉不同序列间多类型空间的相关性,时间模块采用时间卷积网络来提取单个序列内的时间依赖关系。此外,还引入自回归模型来提高预测模型的鲁棒性。为了验证J-ARDNN模型的有效性和优越性,在不同领域的真实时间序列数据集上进行了充分的实验,结果表明,J-ARDNN模型的预测性能优于对比方法。展开更多
The difficulty of multiple targets tracking is how to quickly fulfill the target matching from one flame image to another and fix the position of the target. In order to accurately choose target feature information fo...The difficulty of multiple targets tracking is how to quickly fulfill the target matching from one flame image to another and fix the position of the target. In order to accurately choose target feature information for reliable matching, simplify operations under the reliable precondition, and realize precise moving objects tracking, an approach based on Kalman prediction and feature matching was proposed. The position of the target in next frame image was predicted by Kalman, and then the moving objects of two adjacent frames were matched by the centroid and area methods. When occlusion occurs, the best matching result was found to realize tracking by matching matrix algorithm. The simulation results show that the proposed method can achieve multiple targets tracking accurately and in real-time under complicated motion movements.展开更多
文摘地理传感器时间序列具有复杂动态的语义时空相关性和地理时空相关性。尽管已经开发了各种深度学习模型用于时间序列预测,但很少有模型能专注于捕捉地理传感器时间序列内的多类型时空相关性。此外,同时预测多个传感器在未来某一时间步的值非常具有挑战性。为了解决上述问题,提出了一种自回归模型与深度神经网络的联合模型(Joint model of Autoregression and Deep Neural Network,J-ARDNN),用于处理地理传感器时间序列的多目标预测任务。在该模型中,空间模块用于捕捉不同序列间多类型空间的相关性,时间模块采用时间卷积网络来提取单个序列内的时间依赖关系。此外,还引入自回归模型来提高预测模型的鲁棒性。为了验证J-ARDNN模型的有效性和优越性,在不同领域的真实时间序列数据集上进行了充分的实验,结果表明,J-ARDNN模型的预测性能优于对比方法。
基金Project(61172089) supported by the National Natural Science Foundation of China
文摘The difficulty of multiple targets tracking is how to quickly fulfill the target matching from one flame image to another and fix the position of the target. In order to accurately choose target feature information for reliable matching, simplify operations under the reliable precondition, and realize precise moving objects tracking, an approach based on Kalman prediction and feature matching was proposed. The position of the target in next frame image was predicted by Kalman, and then the moving objects of two adjacent frames were matched by the centroid and area methods. When occlusion occurs, the best matching result was found to realize tracking by matching matrix algorithm. The simulation results show that the proposed method can achieve multiple targets tracking accurately and in real-time under complicated motion movements.
