Wavelet-based_flow clustering_method

we proposed a wavelet-based clustering approach to address the case of geographical flow within linear feature. We take a decomposed perspective on flow clustering and decompose flow into flow sequences, which are constructed by ordering flows with the same length based on their spatial position relationships. Flow sequence makes it possible to consider flow quantity and flow length separately in flow clustering, thus reducing the effect of the inhomogeneity of flow length. We use wavelet transform to identify anomalous flow in flow sequence by analyzing their pattern from spatial-frequency domain. Our method is divided into three steps. Firstly, decomposing flow into flow sequence. Secondly, wavelet transform is applied to identify anomalous flow in the flow sequence. Thirdly, clustering these anomalous flows into flow clusters.