| With the increase of airport flow,airport personnel management has gradually become a short board limiting civil aviation transportation.The service management of personnel has not been developed accordingly,which leads to the increase of the working intensity of staff,and the frequent occurrence of various accidents caused by insufficient management.The problems are becoming increasingly serious.In recent years,with the booming of deep learning,this paper improved and optimized the fast detection and recognition algorithm of airport face in the case of target occlusion and multi-scale target.This paper mainly studies the application based on deep learning about rapid face recognition and detection technology in the airport,and uses the improved algorithm to detect and recognize the airport face quickly.This paper designed the hardware and software parts of the system according to the task requirements of the system.The hardware part mainly includes the airport security check system and airport monitoring management system composed of camera,local terminal,switch and ID card information reader.The software part mainly includes the preparation work before model training,the improvement of face detection and recognition algorithm under complex conditions and the design of fast detection and recognition algorithm.Before model training and test,we should preprocess the collected images first.To solve the problem that the number of airport face samples is small,we should collect and make the airport face sample set under complex circumstances,and generate the data source in accordance with caffe framework format.An improved algorithm for airport face detection and recognition under occlusion is designed,and a regional cross entropy loss function algorithm is proposed.The global average pooling layer is used to replace the full connection layer to increase the learning of local features,so that the training model is more sensitive to the "insignificant" region.A soft non-maximum suppression algorithm was proposed to obtain the best face recommendation box to improve the detection rate of the airport face fast detection and recognition system for occluded face targets.An improved airport face detection and recognition algorithm was designed for multi-scale targets.Based on VGG16 fine-tuning network,the number of network layers was deepened and multi-scale training method was introduced to improve the poor positioning accuracy and poor recognition effect of airport small-scale face targets.In order to solve the problem of deep layers and too many weight parameters in the training model,the singular value compression algorithm is introduced.SVD is used to trim the weight parameter matrix of the full connection layer and decompose it into two smaller matrices,which greatly reduces the computational complexity during model training and improves the speed of the model to detect the target face.Security check system and airport monitoring management system are added to the application software of airport face rapid detection and recognition system,to meet the needs of specific airport sub-areas.After a long time of experimental verification,the airport face rapid detection and recognition system works steadily and has good performance.This paper designed the overall scheme based on the airport face rapid detection and recognition system,collected and made airport face samples,studied the shaded,multi-scale face detection and recognition algorithm,studied fast algorithm,trained system model and then designed a more perfect airport detection and recognition system.Through the test on the face image collected in the airport,the effect basically meets the requirements. |
