Urban Black And Odorous Water Body Recognition Based On Improved U-net Network Remote Sensing Semantic Segmentation

The essence of black and odorous water bodies is caused by the imbalance between environmental protection and urban development.The treatment of urban black and odorous water bodies is one of the key tasks of the current ecological environment in China.At present,the use of remote sensing to monitor urban black and odorous water bodies is still in the early stage of development.The determination of urban black and odorous water bodies is still mainly based on field measurements,and the cost of manpower and material resources is relatively high.According to the results of field data monitoring,continuous measurements cannot be performed in time.Space cannot be measured on a large area,which makes water resources managers unable to quickly understand the black and odor of water bodies from time and space.This article takes advantage of the wide range of remote sensing monitoring,short cycle time,and low cost of manpower and material resources.It uses GF-2 satellite imagery as a data source to monitor urban black and odorous water bodies to make up for the shortcomings of high cost and inconsistent spatial and temporal data of urban black and odorous water bodies.Defects,so as to achieve dynamic monitoring of urban black and odorous water bodies,comprehensively understand the distribution characteristics of black and odorous water bodies,and provide scientific decision-making basis for the treatment of black and odorous water bodies in China.This article takes the urban built-up area of Qingdao as the research object,collects the spectral sampling data of the water body in the research area,analyzes the spectral difference between the black and odorous water body and the general water body based on the GF-2 image band characteristics,and constructs three black and odorous water body recognition algorithms.The optimal algorithm was selected to identify the black and odorous water bodies in Qingdao in 2019,and the distribution characteristics of the black and odorous water bodies in the study area were analyzed.The process identification method proposed in this paper mainly covers the following four aspects:(1)The research in this paper mainly studies the water body inside the urban built-up area.In the preliminary work,the scope of the urban built-up area needs to be determined in order to facilitate the next step.Through comparison,it is found that the BCI index can distinguish the three types of land cover vegetation,impervious surface and soil.Based on the impervious surface information,the density of impervious surface is calculated,and then the urban built-up area is determined.In the process of calculating the BCI index,the GF-2 remote sensing image needs to be transformed with a spike.Because different satellite sensors have different band settings,the spike conversion parameters are also different.At present,there are few studies on the conversion parameters of domestic high-score satellites.The IKONOS satellite image and GF-2 image band parameter settings are similar,so this paper uses the IKONOS ear-cap transformation matrix to transform the GF-2 image,and achieved good results.(2)This paper proposes an improved U-Net network semantic segmentation method for water extraction,and uses the decoding structure of the classic U-Net network to improve the network.(1)The VGG network is used to shrink the path to extract features;(2)In the expansion path The low-dimensional feature information is strengthened in the middle,and the feature map of the upper layer of the shrinking feature pyramid and the feature map of the corresponding expansion path of the next layer are fused to improve the segmentation accuracy of the extraction results;(3)The introduction of conditional random fields in the post-classification processing To refine the segmentation results.In the case of training,verification and test data sets remain unchanged,Seg Net,classic U-Net network and the improved U-Net network in this paper are used for control experiments.The test results show that the improved U-Net network structure is higher than the Seg Net and classic U-Net networks in Io U,accuracy and Kappa coefficient indicators: compared with Seg Net,the three indicators have increased by 10.5%,12.3% and 0.14,respectively;Compared with the results of the classic U-Net network,various indicators have increased by 5.8%,4.4% and 0.05,respectively.(3)Analyze the ground measured spectral data to construct an urban black and smelly entity recognition algorithm.The reflectance of urban black and smelly water in the visible range is lower than that of ordinary water,and the difference in the peak position of the green wave band is more obvious.In the wavelength range of 550-650 nm,the reflectance slope of black and odorous water bodies is smaller than that of ordinary water bodies.Based on these differences in spectral characteristics,three black and odorous water body identification algorithms were constructed: single-band threshold method,normalized black and odorous water body index method(NDBWI)and black and odorous water body classification index method(BOCI).From the recognition results,the following conclusions can be drawn: the single band threshold method is significantly lower in recognition accuracy than the NDBWI method and BOCI method,and cannot be used in the identification task of black and odorous water;the accuracy rate of the recognition by the NDBWI index method is 77.78%;Has the highest recognition accuracy rate,reaching more than 87.04%.(4)According to the BOCI index,the optimal algorithm for identifying black and odorous water bodies,the black and odorous water bodies in Qingdao were identified.Among the 262 sections of the river within the built-up area,a total of 25 black and odorous water bodies were identified.The study found that urban black and odorous water bodies are mainly distributed in the tributaries of the main rivers.The densely populated areas of various urban areas,the vicinity of factories and the areas where the river flow is not smooth are the high incidence areas of black and odor.Because the same river flows through different regional environments,the black and odor conditions of the water bodies in different river sections will also be different.