本文提出一套有别于常规模板匹配算法的流程，不仅能够提供被识别地震的时间、空间、强度信息，还可根据震相关联提取每个地震对应的震相到时资料供后续研究使用。该流程基本原理为：地震事件发生会使得震相记录在时间上趋于丛集，因此可利用密度聚类方法对震相信息进行提取。具体实现步骤为：①对采用相同降采样和滤波处理的模板事件和连续波形进行互相关系数计算，挑选相关性超过阈值的到时信息形成离散时间序列；②对整个序列进行DBSCAN聚类分析并对每个聚类簇开展自洽性检验，获取单个地震的震相关联，进而提取震相到时；③通过去重复事件、震级测算与定位获得最终的结果。将该方法应用到龙门山断裂带南段地震空区，结果显示，2013年5月至2016年12月，35个模板共获得了93个确切的地震信息，是所使用模板的2.65倍，包含390条P波震相以及791条S波震相。绝大部分模板并没有识别出自身的重复/相似地震，而被识别出的地震是由少量模板匹配成功的，主要分布于双河和安顺附近。精定位结果显示，双石-大川断裂倾向NW，同时在安顺以西可能存在某条不知名的、浅部直立、深部倾向NW的"铲状"断裂，震源深度与芦山地震序列类似，存在自SW至NE变深的现象，可能反映了龙门山断裂带南段不同段落具有类似的构造特征。测试结果表明，本文设计的流程可提供包括震源参数和每个地震震相到时信息在内的完整震相报告，在模板识别领域可作为一套计算机自主处理的可行方案。

In this paper we present a set of processes different from the conventional template-matching algorithm,which can provide the spatial,temporal and strength information of the identified earthquakes and extract the corresponding phase arrival time data of each earthquake based on phase association for subsequent research. The basic principle of these processes is that seismic events will make the seismic phase records cluster in time so that the density-based clustering method can extract the seismic phase information. The specific implementation steps are as follows:①calculate the normalized cross-correlation coefficient(NCC)of template events over continuous waveforms after the same down sampling and filtering processing,select the arrival time information whose NCC exceeds the threshold value to form a discrete-time sequence;②apply the DBSCAN clustering analysis on the whole sequence and carry out self-consistency-test on each cluster to obtain the phase association and further the arrival times of a single earthquake;③obtain the final results after the duplicate-removing,magnitude measurement and the earthquake locating. The above processes are applied to the seismic gap in the Longmen shan fault zone's southern segment in this paper. We obtained 93 identified events of 35 templates from May 2013 to December 2016,which is 2.65 times the template used,including 390 P phases and 791 S phases. Most of the templates did not recognize their own repeated/similar earthquakes,and the identified earthquakes were successfully matched by a few templates,mainly distributed near Shuanghe and Anshun. The precise location results show that the Shuangshi-Dachuan fault tends to NW,and there may be an unknown "shovel-like" fault in the west of Anshun. The focal depth is similar to that of the Lushan 7.0 earthquake sequence,2003,and there exists a phenomenon of deepening from SW to NE,which may reflect the similar structural characteristics of different sections of the southern Longmen shan fault zone. The test results show that the processes designed in this paper can provide a complete phase report,including source parameters and arrival time information of each seismic phase. It is a feasible scheme for independent computer processing in the field of template recognition.

P315

中国地震局“地震科技星火计划”（XH21026Y）、国家重点研发项目（2018YFC150330501）和“2020年地震大形势研究”共同资助