基于改进SSD的航天发动机试验报警方法

Method of aerospace engine test alarm based on improved SSD

航天发动机是导弹和火箭的推进装置,其可靠性直接影响航天器的性能。发动机试验是航天发动机的研发、应用过程中不可或缺的一环,其中发动机试验影像是判断发动机性能的重要数据。针对发动机点火故障影像中出现火星等异常情况的突发性和隐蔽性,提出一种基于改进单发多框检测(SSD)的航天发动机试验预警方法。该方法在SSD目标检测算法的基础上,使用算法性能更强的深度残差网络替代原有基础网络,以提高目标检测的精度。同时,针对SSD目标检测算法对小尺度样本的漏检特性,提出基于先验尺度调整的改进措施,提高算法在此种应用环境下的小尺度样本的检测精度。实验表明,该算法适用于航天发动机试验预警应用,且在准确性和实时性上表现良好。

Aerospace engine which is the propulsion device of missile directly affects the performance of spacecraft.In order to verify the performance of the engine,engine test which is an indispensable work in the process of development and application is carried out on the test bench.An early-warning method of space engine test based on the improved SSD is proposed to deal with the abnormal situation which is sudden and hidden.Based on the SSD target detection algorithm,this method uses the depth residual network with stronger algorithm performance to replace the original basic network,so as to improve the accuracy of target detection.At the same time,aiming at the missing detection characteristics of SSD target detection algorithm for small-scale samples,this paper proposes the improvement measures based on prior scale adjustment to improve the detection accuracy of small-scale samples in this application environment.The experimental result shows that the algorithm is suitable for the early warning application of space engine test,and gets better accuracy and real-time performance.