Research On Dynamic Water Scale Identification Of Bulk Carrier Based On YOLO Algorithm

Draft survey is the most commonly used method to calculate the cargo load of bulk carrier in shipping industry.The efficiency of traditional artificial water gauge observation method cannot be guaranteed,and there are some problems in its accuracy.In bad weather conditions,the waterline observation personnel will have a certain risk when working,the waterline is affected by the wind and waves,the fluctuation is violent,the accuracy is difficult to guarantee.In order to improve the intelligence level of ports and whards and solve the shortcomings of traditional artificial water gauge observation methods,this paper proposes a research on the dynamic water gauge identification algorithm of bulk carriers based on the improved YOLOv5 neural network,which can not only provide accurate water gauge reading of bulk carriers,but also achieve real-time detection.According to real-time water gauge reading data of bulk carriers,The variation law of waterline under heavy wind and waves is summarized.At last,the results are corrected and stored to prevent cargo volume disputes.In this paper,the activation function and Neck layer structure of YOLOv5 neural network are improved,and the attention mechanism is added.The meta-ACON activation function was used to replace the Re LU activation function,which prevented the recurrence of Dead Re LU Problem,and enhanced the generalization and optimization effect of the model.Bi FPN structure was used to replace PANet structure to further strengthen the feature fusion and simplify the network model structure and speed up the network training.CBAM attention mechanism is added to improve the adaptive ability of the network through the combination of channel and spatial attention mechanism.Compared with the original network model,the accuracy of the improved network model is improved by nearly 10%,the recall rate is improved by nearly 3%,the number of prediction boxes with the confidence score above 0.5 is significantly more,and the overall training process is more stable.By means of border edge pixel correction method and tilt image correction method,the character recognition accuracy of bulk carrier waterline is further improved,and the accuracy of detection of bulk carrier waterline is further improved through tilt waterline reading study and waterline fitting method.The clustering algorithm is combined to remove invalid data,and the final result is optimized based on the change law of waterline under wind and waves.