2026-10-12T10:31:59Z https://api.figshare.com/v2/oai

oai:figshare.com:article/33812185 2026-09-15T17:42:52Z category_15 category_734 category_106 category_122 portal_5 item_type_3 month_year_09_2026

<p>Parameter settings for each classification model.</p> Chenxiao Yue (24948814) Jiannan Chi (24948817) Jiahui Liu (724342) Yixu Wang (5898245) Cong Zhang (582922) Ming Cao (731917) Neuroscience Biological Sciences not elsewhere classified Science Policy Mental Health individual internal goals efficient feature extraction distinguishing healthy individuals deep learning models capture latent high different task conditions two task scenarios two task conditions two task paradigms human gaze behavior directed search compared task paradigms gaze behavior two factors directed search fixation behavior xlink "> visual attention transgaze model significantly enhanced semantic features scene characteristics results show relevant applications natural scenes guiding effects free viewing external stimuli driven jointly dimensional features differential effects <div><p>In natural scenes, human gaze behavior is driven jointly by external stimuli from scene characteristics and individual internal goals. To investigate the influence of these two factors on gaze behavior, this paper systematically analyzes eye movement behavior under two task paradigms: free viewing and goal-directed search. First, the guiding effects of different hierarchical visual features on fixation behavior under the two task conditions are analyzed. Second, the differences in eye movement trajectory characteristics between the two task paradigms are compared. Finally, the CNN-TransGaze model is proposed to distinguish eye movement behavior patterns under different task conditions. The results show that shape, location and orientation features exhibit strong fixation attraction under both task paradigms, while semantic features are significantly enhanced in goal-directed search compared with free viewing. There are significant differences in eye movement behavior characteristics between the two task scenarios. Moreover, compared with traditional statistical feature extraction methods, deep learning models can capture latent high-dimensional features more effectively, thereby enabling accurate discrimination of distinct eye movement patterns. Efficient feature extraction from diverse eye movement patterns using deep learning enables the implementation of relevant applications, such as distinguishing healthy individuals from patients with abnormal eye movement patterns caused by neurodegenerative diseases.</p></div> 2026-09-15T17:42:34Z Dataset Dataset 10.1371/journal.pone.0357289.t001 https://figshare.com/articles/dataset/_p_Parameter_settings_for_each_classification_model_p_/33812185 CC BY 4.0