| With the rapid development of coastal industrial and agricultural production and the booming tourism industry,floating marine debris pollution has gradually become a common marine pollution phenomenon in various coastal waters of the world.The emergence of floating marine debris has a maj or impact on the sea area.In addition to causing water pollution,visual pollution,it may also threaten the safety of navigation,and even cause the lack of oxygen in the water body to cause fish and other aquatic organisms to die,and ultimately threaten humans health through the food chain.As a kind of floating marine debris monitoring tool,webcam monitoring can continuously obtain the dynamic change process of floating marine debris in a certain key area for a long time,so it is suitable for monitoring the near-shore floating marine debris at a fixed point.Therefore,using image captured from webcam monitoring equipment installed at Haiyang Lou building and Xiamen SongYu harbour,the study developed research on automatic identification technology of floating marine debris,and quantitatively estimate the composition and area of floating marine debris flowing into Xiamen Bay.The research results are expected to provide an important scientixfic basis for the monitoring and management of floating marine debris pollution in Xiamen.The main research results were as follows:Firstly,as an example,this paper took the original image captured from webcam monitoring equipment installed at Haiyang Lou building of Xiamen University from September 2016 to September 2017,using OTSU method based on grayscale stretching,designing a set of mask files in combination with the change of tidal level resulting in the exposure of the interference towards those relatively fixed interference such as Yanwu Bridge,tree,drain pipe,etc.The mask file is read according to the height of the sea level from the camera interpolated from the tide data of Xiamen Bay,used to exclude such interferences.Interference with unfixed position,such as ships,the K-means clustering method is used to classify the ship and the water into two types on the image after using the mask file,and the target water body containing the floating marine debris is obtained.Finally,the appropriate linear stretching multiple is selected according to the water color difference,and the floating marine debris pixels is extracted by the OTSU method.The results showed that this method had higher extraction accuracy in the case with a large amount of floating marine debris in extreme weather events,with a kappa coefficient of 0.7042.We also analyzed the change of floating marine debris after the typhoon Morandi on September 18,2019.The results indicated that extreme weather events had a significant impact on the amount of floating marine debris in Baicheng Beach.Secondly,as an example,taking the image of floating marine debris in SongYu harbour from July to December in 2018,a semantic segmentation network based on vgg16 was constructed,and four training schemes were designed to discriminate the orange_strips(may be formed by a mixture of suspended sediment and foam produced by ship discharge)and wood chips in the image.The results showed that the higher proportion of pixels to be classified in the total pixel points,and the pixels can obtain more training,resulting in higher classification accuracy.Based on this idea,the image of synthesis of the wood chip pixels and the orange strip material were sent to the semantic segmentation network for training,and the classification accuracy is 0.9265 and 0.8768 respectively.This method can still obtain better extraction effect when the amount of floating marine debris is relatively small,and is generally suitable for solving the problem that the interference in the previous chapter cannot be completely excluded and susceptible to the change of sea conditions.lt can effectively reduce the confusion between the pixel of the interference and the floating marine debris,which has strong practicability.Finally,the initial internal and external parameters of the camera and the hourly height of the sea level from the camera were sent into the pinhole model to obtain the area of each pixel corresponding to the physical object hourly.Moreover,the coverage area of floating marine debris of each picture is calculated.The calculation result has an error of 15%.The principle of the method can be understood widely.This method is simple and convenient,being suitable for non-wide-angle cameras which not required high precision. |
