| Soil erosion is one of the serious ecological problems facing China,and the stability of the soil environment is a prerequisite for human development.Although the country has carried out a series of ecological restoration projects to alleviate the soil erosion problem,soil erosion still poses a threat to the ecosystem.With the help of unmanned aerial vehicle(UAV)remote sensing technology,artificial intelligence,data analysis and other technical means,scientific and effective identification of soil erosion intensity,objective analysis of the spatial distribution characteristics of soil erosion intensity and comprehensive evaluation of soil erosion sensitivity play a vital role in preventing and controlling soil erosion and protecting the ecological environment.The paper took the southern region of a coastal area as the research object,and carried out soil erosion intensity identification,spatial distribution characteristics study,and comprehensive evaluation and analyzed of soil erosion sensitivity based on high-resolution UAV multispectral data and convolutional neural network(CNN)model,in order to provide a scientific basis for soil erosion control in the study area and similar regions.The main research content and results of this paper were as follows:(1)Data collection and pre-processing in the study area.The thesis is based on the southern region of a coastal area,and the study area is an ecological restoration area.The study used UAV technology to collect high-resolution multispectral images,and used DJI Terra software and GIS technology to pre-process the multispectral data such as stitching and cropping to obtain panoramic images of the study area;collected data such as rainfall in the study area,and processed and calculated them;and used UAV and RTK technology to collect soil erosion sample data,and processed and labelled them.Based on panoramic images,elevation data and land use data of the study area were obtained using Pix 4D and ENVI,and the land use types of the study area contained forest land,grassland,shrubs,other land,construction land and water.(2)CNN-based soil erosion intensity identification model construction.CNN models can handle high-dimensional data,express complex nonlinear relationships,and have advantages and potential when applied to the identification of soil erosion intensity that cannot be achieved by traditional soil erosion models.Based on the collected highresolution UAV multispectral images and other related data such as rainfall,soil erosion factors were extracted,and the CNN model was used to capture the complex nonlinear relationships among the erosion factors to construct a soil erosion intensity identification model,which was trained and validated.The results showed that the CNN model performed better in the training test phase after 50 iterations;compared with the traditional erosion model(RUSLE)and other machine learning(support vector machine and feedforward neural network)methods,the CNN performed better in four evaluation metrics: accuracy,recall,precision and F1-score.Therefore,the model constructed based on CNN was feasible and can be used for the identification and prediction of soil erosion intensity in the southern region of a coastal area.(3)Analysis of spatial distribution characteristics of soil erosion intensity in the study area.The CNN identification model constructed was used to obtain the overall spatial distribution data of soil erosion intensity in the study area,and GIS technology was applied to analyze and study the distribution characteristics of soil erosion intensity under different land use types,vegetation cover and slope conditions respectively.The results showed that: the study area was mainly dominated by slight erosion,accounting for 73.05% of the overall;soil erosion intensity changed with different land use types,construction land and water bodies basically did not have soil erosion,and the soil erosion intensity of forest land was lower than that of grassland;there was a correlation between vegetation cover and soil erosion intensity in the study area,and overall,the higher the vegetation cover,the weaker the soil erosion,but In the area of plantation forest,due to its simple understory vegetation structure,there was still some moderate erosion;the slope of the study area was mostly concentrated in 0°～15°,and the terrain was relatively flat,but under the influence of human activities,there were more microtopography and temporary piles of loose soil in the area,which were more susceptible to soil erosion,and in a comprehensive view,the slope and soil erosion intensity were positively correlated.(4)Comprehensive evaluation and analysis of soil erosion sensitivity in the study area.Based on the actual situation of the study area,six factors,namely soil erodibility factor(K),topographic factor(LS),cover management factor(C),conservation practice factor(P),nitrogen reflection index(NRI)and normalized difference salinity index(NDSI),were selected as soil erosion sensitivity factors.The weights of each factor were determined based on the entropy weighting method(EWM)to construct a sensitivity index that can comprehensively reflect the sensitivity of soil erosion in the coastal area,and the sensitivity of soil erosion in the study area was evaluated and analyzed.The results showed that soil salinity,vegetation and topography in the study area had significant effects on the distribution of soil erosion sensitivity,and NDSI was the main factor affecting the distribution of soil erosion sensitivity in the southern region of a coastal area;the area of non-sensitive areas accounted for 19.83%,the area of mildly sensitive areas accounted for 29.86%,and the area of areas above mildly sensitive accounted for 50.31%of the total area of the study area.Combining the spatial distribution characteristics of soil erosion intensity and soil erosion sensitivity,the paper proposed targeted countermeasures for soil erosion control. |
PDF Full Text Request