农业机械轨迹作业行为模式识别是一项多变量时间序列分类任务,旨在利用轨迹数据的时空特征识别农机的行为模式。针对已有方法未能从频率角度挖掘农机轨迹的全局特性以及识别精度不足的问题,提出了一种面向农机轨迹行为模式识别的频域注...农业机械轨迹作业行为模式识别是一项多变量时间序列分类任务,旨在利用轨迹数据的时空特征识别农机的行为模式。针对已有方法未能从频率角度挖掘农机轨迹的全局特性以及识别精度不足的问题,提出了一种面向农机轨迹行为模式识别的频域注意力和U型残差网络FARNet。该网络包含两个不同网络分支,用于全面挖掘农机轨迹的依赖信息。其中一个分支搭载了基于频域注意力的Transformer(transformer based on frequency attention,FAT)来挖掘农机轨迹在频域空间的全局时序依赖;另一分支部署了基于正交约束的U型残差网络(U-shaped residual network based on orthogonal constraints,URNet),其以ResUnet作为骨干网络提取轨迹特征图在不同感受野的深层语义信息,探索轨迹特征间的局部空间依赖。最后设计了一种特征对齐学习模块(feature alignment learning module,FA)来融合并对齐两个分支的输出特征,全面调节农机轨迹在全局和局部不同范围下的上下文信息,提高算法的识别性能。为验证所提方法的有效性,在真实轨迹数据集上进行了实验,结果表明,所提方法相比现有的SOTA模型在水稻和小麦收割机轨迹数据集上的F1-score提高了13.94和11.47个百分点。展开更多
As a core part of the electronic warfare(EW) system,de-interleaving is used to separate interleaved radar signals. As interleaved radar pulses become more complex and denser, intelligent classification of radar signal...As a core part of the electronic warfare(EW) system,de-interleaving is used to separate interleaved radar signals. As interleaved radar pulses become more complex and denser, intelligent classification of radar signals has become very important. The self-organizing feature map(SOFM) is an excellent artificial neural network, which has huge advantages in intelligent classification of complex data. However, the de-interleaving process based on SOFM is faced with the problems that the initialization of the map size relies on prior information and the network topology cannot be adaptively adjusted. In this paper, an SOFM with self-adaptive network topology(SANT-SOFM) algorithm is proposed to solve the above problems. The SANT-SOFM algorithm first proposes an adaptive proliferation algorithm to adjust the map size, so that the initialization of the map size is no longer dependent on prior information but is gradually adjusted with the input data. Then,structural optimization algorithms are proposed to gradually optimize the topology of the SOFM network in the iterative process,constructing an optimal SANT. Finally, the optimized SOFM network is used for de-interleaving radar signals. Simulation results show that SANT-SOFM could get excellent performance in complex EW environments and the probability of getting the optimal map size is over 95% in the absence of priori information.展开更多
文摘在自组织映射(Self-organizing Map,SOM)模型的训练过程中,不同类数据对权重矩阵的更新有不同作用,某一类数据对权重矩阵的更新会对其他类获胜神经元特征向量产生偏离其数据特征的影响,从而降低算法聚类精度。针对以上问题,提出一种改进的基于置信度SOM模型(Improved Confidence-based SOM Model,icSOM)。样本数据首先由K-means算法初步分类,为模型训练提供更多的数据信息;然后将预分类后的数据分别训练相互独立的SOM模型,以消除不同类之间的影响;最后在传统SOM模型基础上提出置信度矩阵概念,通过综合判断获胜神经元的置信度及其与输入数据间的欧氏距离最终得到置信神经元,根据置信神经元所属类别给数据分配聚类标签。在鸢尾花数据集(Iris)及葡萄酒数据集(Wine)上利用icSOM进行聚类分析,实验结果表明,所提算法可以更好地处理样本数据,取得了较好的聚类效果。
文摘农业机械轨迹作业行为模式识别是一项多变量时间序列分类任务,旨在利用轨迹数据的时空特征识别农机的行为模式。针对已有方法未能从频率角度挖掘农机轨迹的全局特性以及识别精度不足的问题,提出了一种面向农机轨迹行为模式识别的频域注意力和U型残差网络FARNet。该网络包含两个不同网络分支,用于全面挖掘农机轨迹的依赖信息。其中一个分支搭载了基于频域注意力的Transformer(transformer based on frequency attention,FAT)来挖掘农机轨迹在频域空间的全局时序依赖;另一分支部署了基于正交约束的U型残差网络(U-shaped residual network based on orthogonal constraints,URNet),其以ResUnet作为骨干网络提取轨迹特征图在不同感受野的深层语义信息,探索轨迹特征间的局部空间依赖。最后设计了一种特征对齐学习模块(feature alignment learning module,FA)来融合并对齐两个分支的输出特征,全面调节农机轨迹在全局和局部不同范围下的上下文信息,提高算法的识别性能。为验证所提方法的有效性,在真实轨迹数据集上进行了实验,结果表明,所提方法相比现有的SOTA模型在水稻和小麦收割机轨迹数据集上的F1-score提高了13.94和11.47个百分点。
基金supported by the National Natural Science Foundation of China(61571043)the 111 Project of China(B14010)。
文摘As a core part of the electronic warfare(EW) system,de-interleaving is used to separate interleaved radar signals. As interleaved radar pulses become more complex and denser, intelligent classification of radar signals has become very important. The self-organizing feature map(SOFM) is an excellent artificial neural network, which has huge advantages in intelligent classification of complex data. However, the de-interleaving process based on SOFM is faced with the problems that the initialization of the map size relies on prior information and the network topology cannot be adaptively adjusted. In this paper, an SOFM with self-adaptive network topology(SANT-SOFM) algorithm is proposed to solve the above problems. The SANT-SOFM algorithm first proposes an adaptive proliferation algorithm to adjust the map size, so that the initialization of the map size is no longer dependent on prior information but is gradually adjusted with the input data. Then,structural optimization algorithms are proposed to gradually optimize the topology of the SOFM network in the iterative process,constructing an optimal SANT. Finally, the optimized SOFM network is used for de-interleaving radar signals. Simulation results show that SANT-SOFM could get excellent performance in complex EW environments and the probability of getting the optimal map size is over 95% in the absence of priori information.
