Automatic SAR Target Detection Based On Brain-computer Interface

At present,Synthetic Aperture Radar(SAR)image automatic target recognition is a hot research topic in the application of SAR.It is of great significance for military applications such as perception of battlefield environment and precise strike.However,the performance of existing methods can not meet the needs of practical applications in the face of SAR target detection in complex background.Artificial interpretation based on visual recognition experts is still an important means of image analysis and decision making of SAR.Visual object recognition capabilities are important for human survival,socialization,and development.The effective training of a particular visual field feature can help the trained object become a recognition expert in the relevant field.In military applications,vision experts in SAR image have accumulated rich experience through long-term,high-intensity training for targets of SAR image in an open environment.Their behavioral characteristics are significant and they can quickly and effectively determine military targets under the complex backgrounds,whose performance is much better than existing computer algorithms.The brain-computer interface(BCI)is a direct connection between the human or animal brain and the external device.It realizes the interaction between the human brain and the machine through decoding the electroencephalography(EEG)and translating it into instructions that the machine can read.According to the existing research conclusions,this paper believes that SAR image vision experts should have a specific neural basis to support their specific visual recognition ability.Based on the above facts and assumptions,this paper adopts the human-machine combination scheme,based on the BCI technology to design the information calculation mode of “human-in-loop”,and study the target-specific brain response information of SAR image by studying the SAR image interpretation expert model.An automatic SAR image target detection method based on BCI is constructed,which realizes intelligent automatic detection of SAR ground targets in open complex environment.After a series of screenings,this paper enrolled 10 healthy subjects,4 of whom(4 males,average age 27±2.3)have long been engaged in the research of SAR image field,and another 6 subjects(male 6)Name,average age 25 ± 1.2)is never involved in any form of medical imaging training or scientific research in the field of SAR images.For the large-format SAR image,this paper firstly uses the hexagonal search path to segment the large-format SAR image into a series of small-format SAR images,which reduces the computer processing requirements and ensures the accurate representation of each frame image;Combined with the results of manual marking,1386 target experimental images and 12658 background experimental images were selected.Secondly,the Odd Ball experimental paradigm was determined for the number of target experimental images in SAR images far less than the characteristics of background experimental images.In addition,through the behavioral measurement experiments based on the Odd Ball experimental paradigm of 10 healthy subjects,the experimental results show that for the SAR image,four of them have long been engaged in the research work of SAR image field.The interpretation accuracy rate was significantly higher than the other 6 subjects,and it was regarded as the SAR image interpretation expert model.Then,based on the Odd Ball experimental paradigm,a series of background images and little target image is presented to the test sequence by Rapid Serial Visual Presentation(RSVP).Detect P300 neural signals evoked by visually recognized event and construct target detection of human-machine model of the EEG based on the SAR imagesIn order to build a fast,efficient and accurate EEG-based automatic SAR target method,how to optimize the SAR image target detection model parameters has become one of the focuses of this paper.This paper attempts to study whether experimental design,feature extraction and classification methods have different effects on the accuracy of SAR image detection model from experimental design,channel selection,feature extraction and classification.In the aspect of experimental design,the optimal parameters were found by comparing the effects of different background and image ratios,image presentation time and number of repeated images on the accuracy of EEG interpretation.In terms of channel selection,GaussJordan method was used for each test.The corresponding EEG signals are selected by principal elements to screen out the necessary EEG signal channels.In terms of feature selection,long-short memory neural networks are used to analyze time-domain EEG data with long-term dependence.Try to compare the classification effects of multiple classifiers and choose the appropriate classifier.After several sets of experimental tests and data analysis,aiming at the target detection problem of synthetic aperture radar image,this paper finds that 1:4 target and background image ratio,150 ms image presentation time and support vector machine classification method corresponding to the synthetic aperture radar image target detection model having the best interpretation accuracy.This paper completes the automatic target detection function of SAR in complex background by synthesizing SAR image to interpret brain response information and computer vision in the cognitive process of expert model.Based on the development tools such as Matlab and Python,combined with the designed EEG-based SAR image target detection method and optimized experimental parameters found in previous studies,this paper uses the Odd Ball experimental paradigm with a target-to-background ratio of 1:4,the presentation time of each frame image is 150 ms,Common Space Pattern(CSP)extracts EEG features,Support Vector Machine(SVM)is the classifier to construct an automatic SAR target detection method based on BCI.The method can process four-frame small-format SAR images in one second,and the target interpretation rate of online SAR can reach 79.17%.The research results in this paper may provide a new solution for automatic target detection of SAR images.