Research On Unsafe Behavior Detection Of Ship Officer On Watch Based On CNN

The shipping industry has experienced rapid development,resulting in an increase in maritime vessel traffic.During their duty,ship drivers must ensure the safe operation of their ships by paying attention to the marine environment and taking appropriate measures.However,prolonged duty can lead to unsafe behaviors such as fatigue,stress,and incorrect decisionmaking,which can result in maritime traffic accidents and casualties.Therefore,effectively monitoring the duty behavior of ship drivers and preventing unsafe behavior has become a significant challenge for the shipping industry.Detecting unsafe behaviors,such as fatigue driving,distracted driving,and illegal driving,is not easy to monitor and recognize using traditional methods,necessitating the need for intelligent technology for monitoring and early warning.The detection of ship driver behavior involves multiple challenges,including collecting and processing ship driver behavior data effectively,classifying and labeling ship driver behavior,and developing an efficient behavior detection model.This article proposes the use of deep learning algorithms to detect unsafe behaviors of ship drivers during duty,to assist shipping companies and maritime management departments in supervising the behavior of ship drivers and implementing corresponding safety measures.Additionally,this study provides a new approach and technical means to support the intelligent and automated management of the shipping industry.This study aims to use deep learning algorithms to address the challenge of detecting unsafe behaviors of ship drivers during duty.The proposed approach can provide technical support and method reference for the intelligent and automated management of the shipping industry.The main work is as follows:1.Establish a dataset for detecting unsafe behavior during the duty hours of ship operators.Based on the duty hours of ship operators,a dataset containing various unsafe behaviors is established for subsequent behavior detection model training.By labeling and classifying the data,we established a dataset containing thousands of images and corresponding labels.These images cover various unsafe behaviors,such as falling,smoking,and leaving the post.The establishment of this dataset provides a foundation for algorithm improvement and evaluation in the future.2.Build a behavior detection model and improve it.We selected the YOLO-v4 algorithm as the basic model and introduced technologies such as Mobilenet-v3 and HALF to improve it.Specifically,we used Mobilenet-v3 to accelerate the operation of YOLO-v4 and used HALF to reduce the computational complexity of the model.These improvements enable our algorithm to have significant improvements in computational consumption and recognition speed while maintaining detection accuracy.3.Evaluate the algorithm,comparing computational consumption,recognition accuracy,and recognition speed.We conducted experiments on multiple datasets,comparing the improved algorithm and the original YOLO-v4 algorithm in terms of computational consumption,recognition accuracy,and recognition speed.The experimental results show that the improved algorithm can significantly reduce computational consumption while maintaining accuracy and has faster recognition speed,which can be used for real-time behavior detection tasks.These research results provide effective technical means for the safety management and training of ship operators and provide new ideas and methods for the application of deep learning technology in the shipping industry.In this thesis,the unsafe behavior of ship drivers on duty is detected by computer vision technology,and the data set of unsafe behavior of ship drivers on duty is constructed in the process of research.These research results not only provide effective technical means for safety management and training of ship drivers,but also provide new ideas and methods for the application of deep learning technology in the field of shipping.