Research On Brain Response Representation Of Visual Expert Based On EEG

In image processing,artificial intelligence technology provides strong technical support for automatic target detection and recognition,but the ability to detect and recognize targets in open and complex environments is far from the accuracy and robustness of manual interpretation,making it difficult to meet the needs of practical applications.Brain-machine hybrid intelligence technology(BHIT)offers a new way of solving this problem.BHIT is a new intelligent fusion technology based on brain-machine interface.BHIT forms an intelligent mode of multi-level integration of brain and machine through two-way information perception,analysis and understanding between brain and machine,which achieves the full interconnection of biological intelligence and machine intelligence.At present,an urgent problem in implementation of BHIT for improving the detection and recognition of targets in complex environments is the accurate feature extraction of expert brain response.Around this problem,this thesis investigates the mothods of characterizing the brain response of visual experts during image interpretation,as a way to provide technical support for the application of BHIT in image interpretation.Studies have shown that visual experts in different domains have similar characteristics in behavior and central nervous activity.As human beings are experts in face recognition,in this paper,faces were used as expert objects to explore the specific brain response representation of visual experts in image processing from two different perspectives based on EEG data.First,the neuronal oscillatory couplings in the brain of visual experts were investigated.Second,the dynamic changes in brain network topology of visual experts were investigated.In addition,visual experts in various fields are scarce,so it is difficult to recruit visual experts and conduct scientific research in a short period of time.To address this problem,this paper proposed a visual training scheme based on the semantic and category characteristics.The effectiveness of the training scheme was verified by using synthetic aperture radar image target recognition.The methodology used,the main results obtained and the innovation points of this thesis are specified as follows:(1)This paper investigated the underlying neural mechanisms of visual expert(faces)and explored the specific brain response representations from the perspective of neuronal oscillatory couplings.It has been shown that face processing is a process of information interaction between specific brain regions,and this process involves oscillatory coupling activities between neural assembles in related brain regions.However,current EEG-based face processing studies have mainly focused on the activity of neural assemble in individual brain regions(e.g.,spectral analysis,etc.),which do not comprehensively consider the activity between related brain regions during face processing and cannot accurately characterize the specific responses.In this study,EEG data of faces and no-face objects(bonsai)were recorded under a perceptual task.The neuronal oscillatory coupling activity in the brain at three scales,including whole brain,short-range,and long-range,was assessed by using phase lagged index(PLI),and the brain activity of the two types of stimuli was analysed at three scales within each frequency band based on subject-level analysis.Finally,based on the above analysis,the related PLI values were selected as the specific brain response representations of visual experts.The findings suggested that there were four options of the specific brain response of visual experts.The first was neuronal oscillatory coupling activity in the whole brain over the theta frequency band.The second was shortand long-range neuronal oscillatory coupling activity in brain regions associated with the frontal lobe over the theta and alpha frequency bands.The third was short-and long-range neuronal oscillatory coupling activity in brain regions associated with the ventral visual pathway over the gamma frequency band.The fourth was neuropil oscillatory coupling activity in the occipito-temporal association over the alpha frequency band.(2)The dynamic brain networks method was used to investigate the potential mechanisms of visual expert(faces)and explore the specific brain response representations of visual experts in both temporal and spatial dimensions.To this end,EEG signals of faces and non-faces(ketch)were recorded under a perceptual task,and the functional brain networks were constructed in different frequency bands and different time windows of EEG data.The topology of dynamic brain networks corresponding to the two types of stimuli was analysed at the trial level.Based on the analyses,the relevant network measures were identified as specific brain response representations of visual experts,and the classification ability of the identified representations was validated using machine learning method.The results suggested that the specific topology of the brain network reflected the potential neural mechanism of face processing and the measures of the minimum spanning tree were the specific brain response representations of visual experts(faces).(3)To address the problem that visual experts in related fields are difficult to recruit in a short period of time,in this paper,a training and testing scheme was proposed based on semantic and category information?It has been presented that systematic training can make a normal person to be an visual expert of a certain type of object.Such type of object need to satisfy: morphological similarity and the ability to distinguish individuals within a category by sub-level features of the category.Based on this,this study validates the effectiveness of the scheme on the MSTAR image set.It was found that after training,1)A stable expert effect emerged on the subject's behavior,such as their speed and accuracy of target recognition were greatly improved;2)the N170 component appeared in the relevant electrodes over the right occipito-temporal regions of the brain;3)The topology of dynamic brain network of the subject showed an expert effect similar to that in the previous study when recognizing targets.These results suggested that the proposed scheme was effective for visual expert training.In conclusion,in this study,a series of studies were conducted with the aim of extracting specific brain response representations of visual experts based on EEG collected during expert object(face)and non-expert recognition.The results not only deepen the understanding of neural mechanisms of visual expert specificity,but also clarify the extraction method of visual expert specific brain response representation and effective visual expert training scheme,which provide the research basis and technical support for the implementation of BCHI in image processing.