基于深度学习的多波束海底地质数据异常值检测方法

随着陆地资源不断开发,可用资源减少,人类将资源的开发转移到海洋领域,此时能够收集大量海底数据的多波束测深系统起着重要作用。但未经检测和过滤的、包含异常数据的多波束测深系统会给海洋开发带来危害,因此需要对异常数据进行检测。常用的检测异常值的算法有截断最小二乘估计异常值检测算法、基于改进GA异常值检测算法等,但这些算法的检测精度均较低。随着深度学习不断发展,许多异常值检测的算法均基于深度学习进行改进。提出一种新的异常检测方法——深度支持向量检测算法,与之前方法相比在检测出更多异常值的同时,能减少误判和漏判的情况且提高了检测精度。

ANOMALY DETECTION METHOD FOR MULTIBEAM SEABED GEOLOGICAL DATA BASED ON DEEP LEARNING

With the continuous development of land resources and the reduction of available resources,human beings transfer the development of resources to the marine field.The multibeam sounding system which can collect a large amount of seafloor data plays an important role.However,the undetected and filtered multi-beam sounding system containing abnormal data will bring harm to marine development,so multiple abnormal data are needed for detection.The commonly used outlier detection algorithms include truncated least square outlier detection algorithm,outlier detection algorithm based on improved GA and other algorithms,but the detection accuracy of these algorithms is low.With the continuous development of deep learning,many outlier detection algorithms are based on deep learning.In this paper proposes a new anomaly detection method——depth support vector detection algorithm.Compared with the previous method,this algorithm not only detects more outliers,but also reduces misjudgment and omission,and it improves the detection accuracy.