基于K-means动态聚类的鸢乌贼角质颚模式识别

Pattern of recognition beaks in Sthenoteuthis oualaniensis based on K-means dynamic clustering

中文关键词: 鸢乌贼  角质颚  模式识别  曼哈顿距离  欧氏距离

英文关键词: Sthenoteuthis oualaniensis  Beak  Pattern of recognition  Manhattan distance  Euclidean distance

中文摘要:

      本研究采用K-means动态聚类算法，对2014—2019年间采集于西北印度洋、热带东太平洋、中国南海的鸢乌贼(Sthenoteuthis oualaniensis)样本的角质颚进行识别。基于K-means动态聚类算法能够很好地区分来自3个海区的鸢乌贼，对数据进行z-score标准化后，任选2维角质颚形态学参数以曼哈顿距离和欧氏距离进行K-means动态聚类分析，总正确区分率分别为86.7%和88.7%。K-means动态聚类算法对于鸢乌贼角质颚的识别有很大的参考价值，后续改进优化K-means算法使其具有普适性，将会提高鸢乌贼种群的识别能力。

英文摘要:

      Cluster analysis has been widely used for pattern recognition, machine learning, and in other fields. The K-means dynamic clustering algorithm is simple and efficient, which is why it is one of the most commonly used methods of cluster analysis. The beak of cephalopods, comprising hard tissue, has been widely used to determine species and identify populations owing to its stable structure, corrosion resistance, easily observed growth lines, and abundant characteristic information, causing it to have great application prospects. In this study, the K-means dynamic clustering algorithm was used on 150 pairs of Sthenoteuthis oualaniensis beaks within the mantle length range of 120~200 mm. Samples were collected from the northwest Indian Ocean, the tropical eastern Pacific Ocean and the South China Sea from 2014 to 2019. The results showed that S. oualaniensis from the northwest Indian Ocean had the largest beaks, followed by the tropical eastern Pacific Ocean, and those in the South China Sea. The K-means dynamic clustering algorithm showed that S. oualaniensis from the three areas can be well distinguished. We used z-scores to normalize the data the created a 2D beak morphological parameter matrix to randomize the data before we conducted a K-means dynamic clustering analysis with Manhattan distance and Euclidean distance. The total correct discrimination rate was 86.7% and 88.7%, respectively. This study also identified that the geographic regional differences in beak morphology are unlikely to be due to sampling bias. From the location of the clustering center, we concluded that the Manhattan and Euclidean distance algorithms and outlying points will generate deviations from the clustering center. The K-means dynamic clustering algorithm for beaks of the S. oualaniensis has great reference value. We identified improvements that optimize the K-means algorithm to expand capability for universal use. These improvements and a retrieval system will improve our capabilities to identify S. oualaniensis species.