The Construction And Recognition Of Equipment Operating Key Sample Set Based On Interactive Intelligent Annotation

The operation training of new complex equipment is an important part of daily training,and virtual equipment operation training is an important trend of current development,in which complex new equipment usually consists of multiple subsystems.Therefore,the identification of new equipment operating keys is a necessary means for virtual equipment operation training and auxiliary actual equipment operation.However,at present there are many factors and problems that often can not meet the requirement of the live-fire operation in practical training,such as equipment column loading quantity is little,can only meet the needs of a few personnel training at the same time,equipment space is usually small,expensive equipment value,and military equipment teaching takes place not only at the stated time and place,The modern equipment teaching will occur in extreme environment with high frequency.The appearance of these extreme conditions will cause great interference to the preset guiding teaching equipment.To address the aforementioned issues,this thesis presents an interactive intelligent system tagging based equipment operation training sample set construction and recognition to replace manual on-site teaching by remote operation.The following are the primary research findings:(1)In this thesis,the Faster RCNN network is adopted to identify equipment keys.However,because the sample size of equipment keys is very small,it is even harder to obtain samples in practical application.In view of such small sample size,the optimized Faster RCNN network is proposed in this thesis.By changing the backbone network and optimizing the region generation network and region of interest.Compared with the result of training accuracy of traditional Faster RCNN network,a small number of samples through the optimized Faster RCNN network have a good training effect.(2)In this thesis,homomorphic blurring and region growth preprocessing methods are proposed to reduce the interference of complex background and reduce the pressure of small sample set.Simultaneously,in order to further increase the model’s identification accuracy,this thesis proposes a method to expand the sample set.On the basis of establishing the small sample set manually,an interactive automatic sample annotation and sample set expansion mechanism is proposed.That is,the target key recognition and tracking,at the same time automatically cutting and labeling pre-sample,through the intelligent interactive way of manual positive sample set screening.Finally,the sample set of positive samples is expanded through image enhancement and data enhancement,which solves the challenge inherent in getting a large number of complex samples with equipment keys.(3)In this thesis,a preprocessing method of input image restoration and illumination compensation under complex background is proposed to reduce the interference of complex environment and reduce the pressure of small sample set.Simultaneously,in order to further increase the model’s identification accuracy,this thesis proposes a method to expand the sample set.On the basis of establishing the small sample set manually,an interactive automatic sample annotation and sample set expansion mechanism is proposed.That is,the target key recognition and tracking,at the same time automatically cutting and labeling pre-sample,through the intelligent interactive way of manual positive sample set screening.Finally,the sample set of positive samples is expanded through image enhancement and data enhancement,which solves the challenge inherent in getting a large number of complex samples with equipment keys.The research content of this thesis has better solved the difficulty of obtaining samples for military equipment operation and teaching and training system,and has better identification effect even in extreme environment.Simultaneously,it enables the utilization of distance teaching to replace manual teaching,It not only resolves the issue less equipment and insufficient space.