Study On Recognition Of Mangrove Communities Based On Deep Learning And UAV Monitoring In Jiulongjiang Estuary,Longhai

Mangrove ecosystem has important ecological and economic value.However,due to the impact of human activities,climate change and invasive alien species in recent years,the mangrove area has been steadily decreasing and the ecosystem function has been declining.Therefore,it is imperative to strengthen the protection and management of mangrove,and the spatial distribution of mangrove communities are the basis and premise of mangrove communities.The realization of high-precision classification of mangrove communities with similar spectral and spatial characteristics based on UAV images is of great significance for the management of mangrove ecosystems.In this paper,the mangrove communities in Jiulongjiang Estuary was taken as the research object.On the basis of field investigation,the multi-temporal high spatial resolution images were obtained with the help of unmanned aerial vehicle(UAV)low altitude remote sensing platform,the spectral and spatial characteristics of mangrove communities in different phenological periods were fully explored,and the image recognition mechanism of mangrove communities was deeply analyzed.The main conclusions are as follows:(1)Establish a recognition model of mangrove communities based on deep learning classification method.The classification network evolutionmethod is used to segment the UAV images,the stepwise discriminant analysis method is used to select the feature bands of the images,and then the recognition model of mangrove communities was constructed based on the deep learning classification method.The overall accuracy was 90.62%,and the producer's accuracy and user's accuracy of Kandelia candel,Aegiceras corniculatum and Avicennia marina were 93.43%,92.40%,91.46% and 91.39%,93.12%,93.32%,respectively.(2)The supervised and unsupervised classification methods were used to map the identification and distribution of mangrove communities in the study.There was little difference in the classification accuracy between the two type classification methods,and the overall accuracy was around 60%.Meanwhile,there were more mixing phenomena among different communities.The main reason is that the spectral characteristics of mangrove communities are similar to those of Spartina alterniflora,woodland and grassland,and the traditional supervised and unsupervised classification methods do not make full use of the texture characteristics of the mangrove communities in the classification process.(3)Based on the classification method of deep learning,the spatial distribution results of the mangrove communities in the study area showed that the community of Kandelia candel was the largest mangrove communities,mainly distributed in the inland part of the study area.The area of Aegiceras corniculatum was the next most important,which accounts for 3.66%.And it was mainly distributed at the edge of community of Kandelia candel.The community of Aegiceras corniculatum were the least,accounting for only 0.17% of the total area of the study area,it was distributed sporadically in the western part of the study area and mixed with Spartina alterniflora.The study on the identification of mangrove communities in Jiulongjiang Estuary based on deep learning and unmanned aerial vehicle monitoring can not only provide data support for the management of mangrove ecosystem in Jiulongjiang Estuary,but also provide reference for the identification of mangrove communities in other areas.