Night Vision Target Detection And Recognition Based On Deep Learning

Target detection and recognition of night vision images play an important role in battlefield situation awareness,night monitoring and inspection.However,due to the many disadvantages of night vision images,such as low signal-to-noise ratio of low light level images and few feature points of infrared images,the accuracy of image recognition in night vision is still low.Compared with traditional target recognition method based on artificial features,target recognition algorithm based on Deep Learning has stronger ability of feature extraction and noise immunity,thus has stronger robustness for night vision images.In this paper,we combine image processing with Deep Learning and propose two algorithms for target detection and recognition of night vision:(1)Image recognition based on Lossless-constraint Denoising-Auto Encoder(LD-AE).Traditional Auto Encode algorithm has strong ability of feature learning and data dimensionality reduction,but the generalization ability in noise data,especially low light level images,is vulnerable,which affects classification and recognition performance.In this paper,a lossless constrained Lossless-constraint Denoising(LD)method is proposed,which is adding a lossless constraint term to enhance the resistance and robustness of the encoder to noise.The application of LD-AE to the classification and recognition of high noise low light level image can effectively improve the recognition efficiency.(2)Infrared target detection and recognition model based on Single Shot MultiBox Detector(SSD).SSD is a target detection algorithm with high accuracy and high generalization performance.In this paper,we speed up the computation efficiency of the network by reducing calculate redundant,apply it to the infrared target detection,and propose the mini-SSD real-time infrared target detection algorithm.The mini-SSD has strong robustness to the infrared target under the low contrast or the target deformation.Based on the detection of mini-SSD,the recognition of specific target is realized by combin;ing with LD-AE.Experiments show that the target detection and recognition of night vision images based on Deep Learning method has stronger generalization ability and noise immunity than traditional methods,which still have high detection and recognition accuracy under complex background or large deformation.