Real-time semantic segmentation method oriented to vessel cabin patrol area

Abstract

Disclosed is a real-time semantic segmentation method oriented to a vessel cabin patrol area, comprising the following steps: S1. acquiring a vessel cabin patrol area image, and preprocessing the vessel cabin patrol area image to obtain a processed vessel cabin patrol area image; S2. establishing a vessel cabin semantic segmentation model, training the vessel cabin semantic segmentation model based on the processed vessel cabin patrol area image, and obtaining a trained vessel cabin semantic segmentation model when a set loss function is converged; and S3. performing semantic segmentation on an actual vessel patrol area based on the trained vessel cabin semantic segmentation model. The model provided by the present disclosure not only alleviates a calculation amount of the model, but also improves an operation speed of the model, and improves accuracy of the model by focusing on cabin patrol area edge information.