船舶运载量检测是港口运煤船作业中的重要环节。现有的船舶运载量检测工作通常依赖于人工读取船舶吃水值，并通过查表计算得出最终的运煤量。然而，受到环境天气等因素的影响，人工观测存在主观性、安全隐患等问题，同时，读数人员需要长期的观测经验，无法短期内替代，导致港口作业效率低下，可能引发运载量的争议。针对以上问题，本文利用无人机自动化采集水尺图像，研究了基于深度学习的船舶运载量检测方法，主要研究工作如下：

(1) 针对港口采集水尺图像的动态范围广，采集到的夜间样本少等问题，首先对夜间图像采用增强型的伽马变换进行亮度增强；然后通过分析夜间图像的特征，将白天图像转为夜间图像以扩充夜间图像，解决夜间图像样本少的问题；最后分析算法流程，对不同模型采用不同标注方法，构建水尺数据集。

(2) 针对水尺图像中字符目标小、船体受损，光照不均匀等因素造成的漏检和误检问题，提出一种融合位置信息的轻量化目标检测算法。算法以YOLOv8为基准，首先引入融合位置的通道注意力机制；接着增加小目标检测层；然后通过替换检测头原卷积为深度可分离卷积轻量化检测头；最后修改原损失函数为基于动态非单调聚焦机制的边界框损失函数，完成了对原目标检测网络的优化。除此之外，利用水尺字符的固有特性对输出结果进行优化，并且通过检测到的水尺标志信息进行辅助拍摄水尺画面。通过实验表明，改进后的目标检测网络在水尺字符的识别任务中，与基准网络相比各类别平均精度均值提高了1.3%，每秒处理帧数提高了15 FPS。

(3) 针对采集时间段不一致、采集角度差异大以及波浪大小变化，导致吃水线提取困难和水尺标志倾斜，进而引发吃水值计算误差增大的问题，提出一种改进的语义分割算法U-Net。首先，将原网络中的主干网络替换为特征提取效果更好的VGG16；其次，在编码与解码的跳跃连接中增加混合注意力机制；最后，修改损失函数为焦点损失函数。此外，利用语义分割的结果对水尺图像进行矫正，并对矫正后的结果进行吃水值计算。实验结果显示，改进后的语义分割模型使字符分割的交并比提高了2.23%，水体分割的交并比提高了0.46%，并且在图像矫正方面也优于传统方法。在单帧图像的吃水值计算中，当字符识别正确时，误差保持在3 cm以内。

针对实际应用场景，研发出一套船舶运载量检测系统，采用波峰波谷求均值的方法对连续帧的读数进行处理，实现了对船舶运载量的准确计算。该系统已在港口实际投入使用，具有广阔的应用前景。

Vessel load detection is an important part of the operation of coal carriers in harbor. Existing ship's load detection usually relies on manual reading of ship's draught value and calculating the final coal load by checking the table. However, affected by factors such as environmental weather, manual observation has problems such as subjectivity and safety hazards, and at the same time, the readers need long-term observation experience, which cannot be replaced in the short term, leading to inefficient port operations and possibly triggering disputes over the amount of cargo carried. Aiming at the above problems, this paper utilizes UAVs to automatically collect water gauge images, and studies the deep learning-based ship carrying capacity detection method, and the main research work is as follows:

(1) Aiming at the problems of wide dynamic range of water gauge images collected in the port and few samples collected at night, firstly, the night images are enhanced with enhanced gamma transform for brightness enhancement; then, by analyzing the characteristics of night images, the daytime images are converted to night images to expand the night images and solve the problem of few samples of the night images; finally, we analyze the algorithmic process and use different annotation methods for different models to construct the water gauge data set.

(2) Aiming at the leakage and misdetection problems caused by small character targets, damaged hulls, and uneven illumination in water gauge images, a lightweight target detection algorithm with fused position information is proposed. The algorithm takes YOLOv8 as the benchmark, and first introduces the channel attention mechanism of fused position; then adds a small target detection layer; then replaces the original convolution of the detection head with a depth-separable convolutional lightweight detection head; and finally modifies the original loss function to a bounding box loss function based on the dynamic non-monotonic focusing mechanism, which completes the optimization of the original target detection network. In addition, the output results are optimized by using the intrinsic characteristics of water gauge characters, and the detected water gauge logo information is used to assist the water gauge picture in shooting. Experiments show that the improved target detection network improves the average accuracy of each category by 1.3% and the number of frames processed per second by 15 FPS compared with the baseline network in the recognition task of water gauge characters.

(3) Aiming at the problems of inconsistent collection time period, large differences in collection angles and changes in wave size, which lead to difficulties in draught line extraction and tilting of the water gauge sign, and then trigger the increase of the error in the calculation of draught value, an improved semantic segmentation algorithm U-Net is proposed. Firstly, the backbone network in the original network is replaced with the VGG16 with better feature extraction effect; secondly, the hybridization is added to the jump connection of encoding and decoding in a mixed attention mechanism; finally, the loss function is modified to a focus loss function. In addition, the results of semantic segmentation are utilized to correct the water gauge image, and the draft value is calculated for the corrected results. The experimental results show that the improved semantic segmentation model improves the intersection and concatenation ratio by 2.23% for character segmentation and 0.46% for water body segmentation, and it also outperforms the traditional method in image correction. In the calculation of draft value of single frame image, the error is kept within 3 cm when the characters are recognized correctly.

For actual application scenarios, a ship load detection system has been developed, which adopts the method of peaks and valleys to find the mean value to process the readings of consecutive frames, and realizes the accurate calculation of ship loads. The system has been actually put into use in the harbor and has a broad application prospect.

[1] 姜莉, 姜海航, 王行正, 等. 船舶水尺计重应用研究[J]. 中国口岸科学技术, 2021, 3(12): 40-44.

[2] 杨宝军. 水尺计重作业有了新“利器”[J]. 中国检验检疫, 2019(05): 48-51.

[3] 郑灼辉. 一种新型智能化船舶水尺检测仪[J]. 工业计量, 1999(06): 38-40.

[4] 管利广, 祁亮, 徐小雯. 内河船舶吃水及装载状态监测系统研究与设计[J]. 中国水运, 2015, 15(12): 82-84.

[5] 吴俊, 丁甡奇, 余葵, 等. 船舶底部纵剖轮廓线扫描测量方法[J]. 交通运输工程学报, 2014, 14(02): 62-67.

[6] 孙国元, 毛奇凰. 自动检测船舶吃水和稳性参数的方法探讨[J]. 中国航海, 2002(02): 30-32.

[7] 盖志刚, 赵杰, 杨立, 等. 一种新型激光智能水位测量系统的研制[J]. 光电子.激光, 2013, 24(03): 569-572.

[8] 郭顺福. 正确勘划水尺标志和精确测量船舶吃水[J]. 造船技术, 2009(04): 26-28.

[9] 罗婧, 施朝健, 冉鑫. 一种视频图像船舶吃水线自动检测方法[J]. 船海工程, 2012, 41(01): 30-32+37.

[10] 郭方. 基于视频的船舶吃水线检测方法的研究[D]. 大连海事大学, 2010.

[11] 刘丹. 基于图像处理的散货船港航交重计量系统[D]. 大连海事大学, 2012.

[12] 吴海. 基于机器视觉的船舶吃水线检测系统研究[D]. 燕山大学, 2016.

[13] 林王峰. 基于视频图像的船舶水尺自动测量系统的设计与实现[D]. 集美大学, 2017.

[14] 安鸿波. 轮船吃水线动态视觉识别方法[D]. 中国矿业大学, 2020.

[15] 李锦峰. 基于图像的船舶水尺智能检测技术研究[D]. 哈尔滨工程大学, 2021.

[16] 朱学海, 张帅, 张东星, 等. 基于机器视觉与深度学习的船舶水尺智能识别技术研究与应用[J]. 检验检疫学刊, 2019, 29(02): 101-104+110.

[17] 吴禹辰. 基于图像的船舶水尺自动检测系统设计[D]. 哈尔滨工程大学, 2019.

[18] 薛银涛. 基于卷积神经网络的船舶吃水线检测算法[D]. 燕山大学, 2019.

[19] 张钢强. 基于深度学习的船舶水尺读数识别系统研究[D]. 浙江理工大学, 2021.

[20] 肖禹辰, 臧奇颜, 张键. 一种稳健的船舶水尺检测与识别方法[J]. 江苏海洋大学学报, 2023, 32(01): 81-87.

[21] Krizhevsky A, Sutskever I, Hinton G. ImageNet Classification with Deep Convolutional Neural Networks[C]// NIPS. Curran Associates Inc. 2012: 1097-1105.

[22] Girshick R, Donahue J, Darrell T, et al. Rich Feature Hierarchies for Accurate Object Detection and Semantic Segmentation[C]// Proceedings of the IEEE conference on computer vision and pattern recognition. 2014: 580-587.

[23] Girshick R. Fast R-CNN[C]// Proceedings of The IEEE International Conference on Computer Vision. 2015: 1440-1448.

[24] Ren S, He K, Girshick R, et al. Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks[C]// Advance in neural information processing systems. 2015: 91-99.

[25] HE K, GKIOXARI G, DOLLÁR P, et al. Mask R-CNN[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2020, 42(2): 386-397．

[26] 吉训生, 李建明.一种改进的Faster-RCNN电路板字符检测方法[J]. 小型微型计算机系统, 2020, 41(06): 1291-1295.

[27] Redmon J, Divvala S, Girshick R, et al. You Only Look Once: Unified, Real-Time Object Detection[C]// Proceedings of the IEEE conference on computer vision and pattern recognition. 2016: 779-788.

[28] Liu W, Anguelov D, Erhan D, et al. SSD: Single Shot MultiBox Detector[C]// European Conference on Computer Vision. 2016: 21-37.

[29] Redmon J, Farhadi A. YOLO9000: Better, Faster, Stronger[C]// Proceedings of the IEEE conference on computer vision and pattern recognition. 2017: 7263-7271.

[30] Redmon J, Farhadi A. Yolov3: An Incremental Improvement[J]. arXiv preprint arXiv: 2018, 1804. 02767.

[31] Bochkovskiy A, Wang C Y, Liao H Y M. Yolov4: Optimal Speed and Accuracy of Object Detection[J]. arXiv preprint arXiv: 2020, 2004. 10934.

[32] Li C, Li H, et al. YOLOv6: A single-stage object detection framework for industrial applications[J/OL]. arXiv: 2022, 2209.02976.

[33] Law H, Deng J. Cornernet: Detecting objects as paired keypoints[C]// European Conference on Computer Vision(ECCV), 2018: 734-750.

[34] Duan K, Bai S, et al. Centernet: Keypoint triplets for object detection[C]// Proceedings of the IEEE/CVF international conference on computer vision. 2019: 6569-6578.

[35] Zhou X, Zhuo J, Krahenbuhl P. Bottom-up object detection by grouping extreme and center points[C]//Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2019: 850-859.

[36] Duan K , Xie L , Qi H ,et al. Corner Proposal Network for Anchor-free, Two-stage Object Detection[J]. 2020:399-416.

[37] Long J, Shelhamer E, Darrell T. Fully Convolutional Networks for Semantic Segmentation[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2015, 39(4): 640-651.

[38] Ronneberger O, Fischer P, Brox T. U-Net: Convolutional Networks for Biomedical Image Segmentation[J]. Springer, Cham, 2015: 234-241.

[39] Badrinarayanan V, Kendall A, Cipolla R. Segnet: A deep convolutional encoder-decoder architecture for image segmentation[J]. IEEE transactions on pattern analysis and machine intelligence, 2017, 39(12): 2481-2495.

[40] Chen L C, Zhu Y, Papandreou G, et al. Encoder-decoder with atrous separable convolution for semantic image segmentation[C]// Proceedings of the European conference on computer vision (ECCV). 2018: 801-818.

[41] QIN X, ZHANG Z, HUANG C, et al. U2-Net: going deeper with nested U-structure for salient object detection[J]. Pattern Recognition, 2020, 106: 107404.

[42] Zheng S, Lu J, Zhao H, et al. Rethinking Semantic Segmentation from a Sequence-to-Sequence Perspective with Transformers[C]//Computer Vision and Pattern Recognition. IEEE, 2021: 6881-6890.

[43] Babakhani P, Zarei P. Automatic gamma correction based on average of Brightness [J]. ACSIJ Advances in Computer Science, 2015, 4(6): 156-160.

[44] 袁昊. 基于深度学习的交通标志检测识别与跟踪及应用研究[D]. 长安大学, 2023.

[45] CHEN S, ZOU X, ZHOU X, et al. Study on fusion clustering and improved YOLO v5 algorithm based on multiple occlusion of Camellia oleifera fruit[J]. Computers and Electronics in Agriculture, 2023, 206: 107706.

[46] 夏衍, 罗晨, 周怡君等. 基于Swin Transformer轻量化的TFT-LCD面板缺陷分类算法[J]. 光学精密工程, 2023, 31(22): 3357-3370.

[47] 张万枝, 曾祥, 刘树峰, 等. 基于改进YOLO v5s的马铃薯种薯芽眼检测方法[J]. 农业机械学报, 2023, 54(9): 260-269.

[48] WANG J D, SUN K, CHENG T H, et al. Deep high-resolution representation learning for visual recognition [J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2021, 43(10): 3349-3364.

[49] YU C Q, GAO C X, WANG J B, et al. BiSeNetV2: bilateral network with guided aggregation for real-time semantic segmentation [J]. International Journal of Computer Vision, 2021, 129(11): 3051-3068.

[50] LI R, ZHENG S Y, ZHANG C, et al. ABCNet: attentive bilateral contextual network for efficient semantic segmentation of fine-resolution remotely sensed imagery [J]. ISPRS Journal of Photogrammetry and Remote Sensing, 2021, 181: 84-98.