为准确学习飞行员操作手势的轨迹分布模式,提出一种改进的层次自组织映射方法。引入Wilcoxon秩和检验技术,结合编辑距离判断内部网的匹配程度,通过交叉验证使验证集获得误差最小,从而自适应取得判断异常的阈值。根据训练得到的轨迹分布模式检测操作过程中的局部异常,判断运动轨迹所表示的事件是否为异常事件,并预测手势将来行为轨迹。实验结果验证了改进方法的有效性。 In order to accurately study the trajectory distribution pattern of pilots’ operation gestures, an improved hierarchical self-organizing mapping method is proposed. The Wilcoxon rank sum test technique is introduced, and the match distance of the internal network is determined by the edit distance. The error of the verification set is minimized by cross-validation, so as to adaptively obtain the threshold for judging the abnormality. According to the trajectory distribution model obtained by training, a local anomaly during the operation is detected to determine whether the event represented by the trajectory is an abnormal event and the future trajectory of the gesture is predicted. The experimental results verify the effectiveness of the improved method.