| With the rapid development of industrialization and urbanization in our country,the amount of urban solid waste is increasing,which seriously affects the urban natural environment and the comfort of residents.Therefore,it is imperative to promote efficient and orderly monitoring and investigation of solid waste storage yards.Accurate detection of the location of solid waste storage yards is an important prerequisite for the investigation work of solid wastes.And the assessment of the environmental impact of solid waste storage yards helps to implement the management plans such as solid waste inspection and cleaning,and construction of landfills in an accurate and orderly manner.However,the urban solid waste yards are wide in sources,large in number,and diverse.The traditional methods based on manual investigation are not only time-consuming and laborious,but also have problems such as inconsistent data calibers,incomplete coverage,and low efficiency,which are difficult to meet the urgent needs of urban environmental governance.The rapid development of geospatial technology provides new methods for regional solid waste yard detection and environmental impact assessment.Firstly,high-resolution remote sensing image data sources are becoming more abundant,and computer automated image processing methods are developing rapidly.The use of automated methods to process high-resolution remote sensing images can achieve wider,faster,and more objective remote sensing image object detection,which can greatly complement the manual methods.And the deep learning method effectively solves the limitations of manual design features by automatically learning image features.However,compared to the detection of general ground objects,the detection task of solid waste yard has more complicated characteristics: 1)The scale of the solid waste yard on the remote sensing image is variable and the difference is huge.A fixed-size sliding window can hardly take into account multiple targets with such a huge difference in scale at the same time.In addition,when the window slides in a large target area,the complete yard will often be split.2)The formation of solid waste yards is closely related to other features in the scene.For example,domestic solid waste often appears with residential buildings,and industrial solid waste and construction waste are mostly piled up on bare land or grassland.Therefore,the fully reasoned of global information of the image can effectively enhance feature expression.In view of the above characteristics of solid waste yard detection tasks,this thesis proposes a solid waste yard detection model based on deep learning and global reasoning,and designs a multi-scale optimization strategy for actual detection tasks.In particular,the research content and results of solid waste yard detection in this thesis are summarized as follows:(1)This thesis constructs a multi-scale solid waste yard detection model based on a deep learning model and a global reasoning unit.This model can make full use of the advantages of the multi-scale deep learning model and the global reasoning unit’s ability to mine global information in the scene.Experimental results show that the F1 score of the model constructed in this thesis can reach 0.7334,and the AP50 can reach 70.18%.Compared with the original object detection model,the F1 score and AP50 of this method are increased by 0.4 and 4%respectively.In addition,compared with other deep learning classic target detection models,the method in this thesis also shows great advantages.(2)Aiming at the solid waste yard detection task in the research area,this thesis designs a multi-scale optimization strategy,which is specifically divided into two steps: multi-scale cropping and boxes merging.The experiment proves that the multi-scale optimization strategy proposed in this thesis can effectively solve the problem of large differences in the scale of the solid waste yard and the problem of complete yard being split.Combined with the detection model constructed in this thesis,it can effectively deal with the multi-scale solid waste yard detection task.Secondly,studies in the field of environment management mostly studied the relationship between emissions and socio-economic factors from the perspective of time series,while ignoring the spatial distribution of solid wastes.Considering that the solid waste yard is closely related to various surrounding environmental features in space,this thesis combines multisource geospatial data such as remote sensing images,point of interest,water,and road network,and constructed environmental impact indexes from the perspectives of nature and social economy through the establishment of buffer zones and overlay analysis to quantitatively analyze the environmental impact of the solid waste yard in the central city of Xuzhou.Results show that:(1)The overall distribution of solid wastes in central city of Xuzhou is sparse,and the relatively concentrated areas are mainly located near the West Third Ring Road,the East Third Ring Road,the railway line on the southeast side,and wholesale market.(2)The distributions of impact levels of different features show some correlation,but also show large difference.In general,solid wastes in areas with high vegetation coverage and close to water bodies have higher natural environmental impact.The social and economic impact of solid wastes near the city center and transportation lines is higher.(3)The areas with higher levels of the comprehensive environmental impact index of solid wastes are mainly located near the Old Yellow River on the east of the city center,the railway junction on the north side of the wholesale market,and the Yunlong Lake.In comparison,the environmental impact index of the solid waste on the east of the city center and near the Old Yellow River is relatively higher.Relevant departments can prioritize the solid waste investigation work according to the environmental impact index.In addition,areas with a low index level are more suitable for the construction of landfills and other facilities. |
