Research On Sea Ice Identification And Parameter Calculation Based On Computer Vision

With the global warming,the Arctic channel is gradually opened,and the demand for polar ships is increasing.The sea ice parameters play an important role in the design and manufacture of polar ships.However,due to the harsh polar environment,it is very difficult to obtain the ice image data.How to use the limited data to efficiently obtain the sea ice parameters has become a hot topic in the field of polar marine engineering in recent years.This thesis focuses on the recognition of sea ice image and the extraction of parameter.By dividing into the physical meaning of different parameters and analyzing the influencing factors,we carry out the corresponding identification algorithm selection.With collated and labeled the polar sea ice data set,we carry out the machine learning of the sea ice instance segmentation.Finally,we discuss the classical theoretical model and parameter calculation algorithm,and introduce the new methods which is applicable to the polar sea ice identification and parameter calculation.Firstly,this article introduces the working principle and network structure of the instance segmentation model,including reasonable construction of data sets,effective extraction of image features,scientific setting of training parameters,and accurate testing of training results.It provides services for accurately identifying sea ice and extracting sea ice shape features.Computer vision and digital image processing are used to process the result of the instance segmentation,extract the region of interest and binarize the sea ice image,extract the circumferential cracks of the wedge ice,and store all the coordinate points on the cracks.Considering the different sizes of sea ice in the ship-based images,an algorithm suitable for the fitting of sea ice circumferential cracks is proposed.The fitting results of circumferential cracks and the result of calculating ice-breaking parameters are analyzed and evaluated.In addition,the data set and training parameters are optimized and improved according to the calculation requirements of sea ice concentration.The sea ice concentration is calculated and the sea ice size distribution is analyzed based on the actual image of the cold region,and the reliability of the algorithm is tested,which provides an auxiliary decision for the safe navigation of ships.