The invention provides a ship monitoring intrusion detection method based on a multistage early warning strategy, and the method comprises the steps: S1, estimating an intrusion target in a data set through employing a preceding-stage detector according to the data set prepared in advance; s2, training a post-stage detector according to a pre-prepared data set; s3, observing the scene to obtain an observation image; s4, the observation image is sent to a preceding-stage detector, and a plurality of potential invasion target sub-images are obtained; s5, sending the plurality of potential intrusion target sub-graphs into a post-stage detector, dividing the potential intrusion target sub-graphs into known alarm targets, known alarm-free targets and unknown targets, storing the known alarm targets and the unknown targets into a disk, and sending out a report; and S6, returning to S3, and circularly executing S3 to S6. The comprehensive detection accuracy and recall rate of the multi-level detector system can reach 95% or above on a self-built data set. The recall rate of a preceding-stage detector can reach 98.31%, the accuracy rate is 80.26%, and all potential targets in a scene can be detected; the precision rate of a post-stage detector is 97.23%, the recall rate is 90.25%, and candidate targets can be screened.
本发明提供一种基于多级预警策略的船舶监控入侵检测方法,包括S1、根据预先准备的数据集,采用前级检测器估计数据集中的入侵目标;S2、根据预先准备的数据集,训练后级检测器;S3、对场景进行观测,得到观测图像;S4、将观测图像送入前级检测器,得到若干潜在入侵目标子图;S5、将若干潜在入侵目标子图送入后级检测器,分为已知报警目标、已知无警目标和未知目标,将已知报警目标和未知目标存入磁盘,并发出报告;S6、返回S3,循环执行S3至S6。本发明多级检测器系统综合检测精确率和召回率在自建数据集上均可达到95%以上。前级检测器召回率可达98.31%,精确率为80.26%,能够检出场景中一切潜在目标;后级检测器精确率为97.23%,召回率为90.25%,能够筛选候选目标。
Ship monitoring intrusion detection method based on multistage early warning strategy
一种基于多级预警策略的船舶监控入侵检测方法
2023-08-11
Patent
Electronic Resource
Chinese
Ship monitoring intrusion detection method based on mode matching
European Patent Office | 2023
|Automobile door opening anti-collision early warning system and multistage early warning method
European Patent Office | 2022
|Ship monitoring intrusion detection method based on evolutionary self-learning
European Patent Office | 2023
|