The gradual opening of Arctic shipping routes will significantly impact the global shipping industry, but the severe sea ice conditions in the Arctic pose serious risks, with shipwreck incidents caused by collisions with sea ice occurring frequently. This paper focuses on the Yongxing shipwreck, a typical sea ice accident, and utilizes a deep learning model to classify sea ice images from the incident area, aiming to analyze the causes of the Yongxing accident. The model used in this study is based on the MobileNetV3 backbone, combined with the CBAM and FPN modules to enhance the network's feature extraction capabilities. Experimental results show that the overall classification accuracy of the proposed network reached 97.18%, with a kappa coefficient of 0.962. The classification accuracy for each of the four categories (seawater, new ice, gray-white ice, and thick ice) exceeded 95%. The analysis of the classification results indicates that the Yongxing ship mistakenly entered a thick ice zone in the Tatar Strait, leading to a collision with sea ice and subsequent sinking. The review suggests that acquiring high-accuracy sea ice maps through remote sensing can provide reliable decision-making support for Arctic vessels in planning safe routes to avoid sea ice hazards.