Maritime target detection using simulation-based human visual recognition experiments

Maritime target detection plays a vital role in security, transportation, rescue, and other domains. In recent years, automatic recognition algorithms for maritime target detection have become increasingly prevalent. These algorithms are often compared with human visual recognition performance in specific aspects. However, experimental studies on human-eye recognition of marine targets under stable optical conditions are rarely reported. This lack hinders comparative research between deep learning algorithms and human performance due to insufficient methodological and data support. To address these challenges, this study employed simulation methods to establish a visible light imaging dataset comprising three categories and nine types of marine vessel targets under both clear and foggy weather conditions. An experimental methodology analogous to deep neural network learning was developed for human visual recognition of maritime targets. The recognition accuracy rates for human-eye identification of maritime targets at different distances were calculated, and the distribution of misclassification types was analyzed.