摘要
由于部队弹库导弹出入库人工登统计的工作现实,导致了导弹出入库过程中浪费大量的人力及时间成本。利用深度学习的方式,训练一个可供弹库使用的文本识别模型。首先实地拍摄训练样本集,根据所拍摄的样本集,人工合成更多数据集以供训练。通过实验,发现CRNN算法识别目标导弹上的编号准确率以及LOSS均达到较好的性能。因此,结合Django框架,调用训练好的CRNN模型,设计实现部队导弹目标编号识别系统,经过实地测试,该系统可满足部队弹库实际使用。
Because of the reality of the manual log-in statistics of troop missiles entering and leaving the warehouse,a lot of manpower and time costs are wasted in the process.Using deep learning technology,training a text recognition model that can be used by the missile warehouse.First,the training sample set is shot on the spot,and according to the sampled set,more data sets are artifically synthesized for training.Through experimental comparison,it is found that the accuracy and loss of CRNN algorithm to identify the number on the target missile have achieved good performance.Therefore,in combination with the Django framework,the trained CRNN model is called to design and realize the military missile target number recognition system.After field testing,this system can meet the actual use of the military warehouse.
作者
丛林虎
何伟鑫
CONG Linhu;HE Weixin(College of Coastal Defense,Naval Aviation University,Yantai 264001)
出处
《舰船电子工程》
2021年第8期88-92,共5页
Ship Electronic Engineering
基金
国家自然科学基金项目(编号:51605487)资助。
