In order to better support the upgrading of transportation supervision capabilities in the Qiongzhou Strait, the main existing problems in the ro-ro passenger transportation were analyzed. The risk factors of cargo carried by ro-ro passenger ships were analyzed, and a visual identification classification method was proposed. Relying on the capabilities of the PaddlePaddle deep learning platform, a technology framework for carrier risk identification is proposed. Four main application scenarios are selected for mixed flow of passengers and vehicles, people falling into the water, people falling, and people gathering. The vehicle license matching, vehicle securing, and vehicle Application testing was carried out in three main application scenarios of spontaneous combustion. Through application testing, it was found that the risk identification of cargo carried by ro-ro passenger ships based on the deep learning framework can reduce personnel workload to a certain extent, improve the coverage of safety supervision, and improve the safety management level of ro-ro passenger ships.