糖心原创

Dr. Assaf Harel, 糖心原创

Training expertise in scene recognition

Abstract:

Visual analysis of complex real-world scenes is essential to聽a variety of professional contexts, ranging from defense and intelligence聽to聽architecture and urban planning. Expertise in recognizing information-rich聽yet聽highly variable scenes is putatively achieved through experience, yet聽little is聽currently known about how skills in scene recognition are formed and evolve聽during learning, and what underlying neural mechanisms support their聽acquisition. The present study is a first attempt at addressing these聽questions,
quantifying the behavioral changes associated with the acquisition of scene聽expertise. We assembled a rich stimulus-set consisting of high-resolution聽color聽scene images varying across five dimensions: Viewpoint聽(aerial/terrestrial), Naturalness聽(manmade/natural), and three hierarchical categorization levels:聽Basic-level,聽Subordinate, and Exemplar. For instance, the category 鲁deserts虏 contained聽three聽deserts types (Sandy, Shrub and Rocky), and each desert type contained ten聽individual images of specific deserts. Critically, each individual scene聽was聽presented both in an aerial and聽terrestrial viewpoint, to assess generalization across viewpoints. We聽trained 15 participants to categorize these scenes for a total of 12聽hours. Each聽individual training regimen was comprised of six sessions; participants聽trained聽on half of the stimuli for five sessions, and in the sixth session they聽viewed聽he other half of the scenes. To assess the efficiency of training, we聽employed聽two behavioral metrics: (1) within-set learning (i.e. learning across the聽five聽sessions), and (2) generalization (i.e. transfer of learning). Learning聽occurred within the five sessions (evident in a monotonic decrease in聽reaction聽times and increase in accuracy), and notably, we also found transfer of聽learning, as performance in the sixth session was pronouncedly better than聽performance in the first four training sessions. Together, these results聽suggest that expertise in scene recognition can be trained in the lab and聽will聽form the basis for future studies on the neural substrates of scene聽expertise.