Research And Application Of Key Technology Of Object Recognition In CPS

Cyber-Physical Systems(CPS)is a system that closely combines information space and physical space.It integrates information technology and automatic control technology to create a ”state awareness-real-time analysis-decision execution” data closed loop.Environmental state perception is the basis of CPS,and the physical objects to be perceived include people and various devices.With the development of deep learning technology,target recognition technology has gradually matured.Studying the application of target recognition technology in CPS will effectively reduce the security problems caused by manual duty and inspection negligence,also provides the possibility of transforming traditional systems into CPS.And it can be provided to CPS for multi-source information fusion,thereby improving the reliability,accuracy and security of decision-making.As a branch of target recognition technology,face recognition technology has been widely used.Face recognition based on surveillance cameras in CPS is different from the currently widely used scenes.There are large changes in face poses.Therefore it is impossible to require on-site personnel to cooperate with recognition.In order to improve the accuracy of multi-pose face recognition in surveillance scenarios,this paper proposes a face recognition algorithm based on pose estimation.After detecting the face from the image,the algorithm first classifies the face to the closest angle category in the sample library,and then performs classification and recognition in the sub-sample library.In response to the pose estimation needs of the algorithm,this paper designs a pose estimation algorithm based on face key points,which maps face key points to a binary map,and uses an improved AlexNet to train the angle classifier.Tested on CAS-PEAL's PD(Pose Looking Down)sub-library,the angle recognition accuracy rate reached 94.43 %,and the rank 2 adjacent gesture recognition accuracy rate reached 98.35 %.The recognition accuracy of the face recognition algorithm combined with pose estimation on the pose PD± 67(the face is a top-down angle,and the left and right deflection is 67 degrees)reaches83.44 %.When the FPR is 0.01,the TPR is increased from 88.8 % To 89.6 %.Finally,the above algorithm is integrated to realize the face recognition system deployed on the edge nodes.This paper also studies the application of target recognition technology in device status recognition,and proposes a device indicator position detection and state recognition algorithm based on monitoring images.This algorithm performs two-stage indicator position detection based on SSD and Hough transform,and designs a multi-frame fusion algorithm to improve detection based on the characteristics of relatively fixed camera angle and device position Rate.State recognition in the HSV(Hue Saturation Value,hue,saturation,lightness)color space uses the LeNet model.The algorithm has been tested on the self-built power equipment indicator data set.After four consecutive frames of fusion detection,the detection rate reached 100%,and the state recognition accuracy rate reached99.8%.Finally,the detection algorithm and recognition algorithm are jointly deployed and tested,which proves the effectiveness and accuracy of the algorithm.Finally,this paper discusses how CPS integrates the field personnel information and equipment status information obtained after fully perceiving the field environment,and performs simulation testing on the scene simulation data based on the decision tree algorithm.