海量传感数据分类可视化重构算法设计

Design of visual reconstruction algorithm for massive sensing data classification

由于海量传感数据多维属性和空间维度的存在,传统的可视化重构算法可能无法展示所有的特征信息,因此,研究一种能够有效展示多维属性和空间维度的可视化重构算法。根据海量传感数据的不同属性进行特征提取后,再从空间维度的角度对传感数据进行转换,并将每个传感数据时间序列转换为时间序列图的形式。接着,使用扩张卷积神经网络(DCNN)对转换后的时间序列图进行分类任务。基于学习分类矩阵和微图池化方法对分类后的序列图进行可视化重构,将重构后得到的输出值映射到一组簇中,并计算节点得分。最后,将节点得分排名靠前的节点作为重构的基点,实现传感数据的重构。测试结果表明,所设计的算法能够以较高的灵敏度实现对数据的准确分类处理。

Due to the existence of multidimensional attributes and spatial dimensions in massive sensor data,traditional visualization reconstruction algorithms may not be able to display all feature information.Therefore,a visualization reconstruction algorithm that can effectively display multidimensional attributes and spatial dimensions is studied.After feature extraction based on the different attributes of massive sensing data,the sensing data is transformed from a spatial dimension perspective,and each sensing data time series is transformed into a time series graph.Next,an extended convolutional neural network(DCNN)is used to classify the transformed time series graph.Based on the learning classification matrix and micro graph pooling method,the classified sequence graph is visually reconstructed,and the reconstructed output values are mapped to a set of clusters,and node scores are calculated.Finally,the nodes with the highest score ranking are used as the basis for reconstruction to achieve sensor data reconstruction.The test results show that the designed algorithm can achieve accurate classification and processing of data with high sensitivity.