LAUSR

Draft reading is a common means to measure the amount of cargo loaded on a ship in the maritime industry, and waterline detection is a key technology in this regard. A variety of factors limits traditional edge detection-based waterline detection methods. They are challenging to apply to different situations. Existing deep learning-based methods do not consider the specificity of the task and have disadvantages such as complex processing flow and insensitivity to the waterline. This article proposes a novel column-based selection method that treats waterline detection as a classification problem. Compared with the current pixel-wise segmentation, the proposed method can effectively reduce the computational cost. Based on the proposed method, we also introduce a structural loss. By introducing spatial constraints, the accuracy of the location selection is further improved. Moreover, we propose a simple architectural unit to accomplish draft mark detection, which effectively enhances the lightweight in draft reading. The experimental results on the real-world datasets demonstrate the state-of-the-art performance of the proposed waterline detection method. In addition, the average error on the draft reading task is 0.27 m, and the processing speed achieves 25+ frames per second with a resolution of $544 \times 960$ .