| In recent years,the rapid and blind expansion of cash crop cultivation has become a global phenomenon,resulting in the replacement of a large number of original forest land and arable land.While bringing economic benefits to the local area,this has also caused ecological and environmental problems such as increased soil erosion,reduced soil fertility,weakened carbon sequestration capacity,and reduced biodiversity.There is an urgent need for effective regulation to realize the sustainable use of land resources and the efficient protection of the ecological environment.In this context,how to quickly extract information about the spatial distribution of cash crops and their dynamic changes,accurately analyze the motivation and process mechanism of the rapid expansion of cash crops,and effectively predict the spatial pattern of future expansion in different scenarios has become the core problem currently facing.The method of automatic extraction of high-precision feature information based on remote sensing technology is the key to solving these problems.Mountain orchards have the characteristics of large spatial scale differences,irregular shapes,scattered distribution,and closeness to the spectral characteristics of the surrounding forests,and are easily affected by topography and the environment.As a result,phenomena such as the same spectrum of foreign matter,different spectrum of the same matter,and mixed pixels are more common in remote sensing images.It also leads to problems such as poor model accuracy and insufficient spatial scale adaptability when using traditional deep learning models to extract orchard spatial information.Therefore,it is necessary to explore the high-precision automatic extraction method of mountain orchards to solve the technical problems of low accuracy and efficiency of orchard spatial information collection under complex terrain conditions.Only in this way can it provide reliable data support for further research on the driving force mechanism of orchard expansion and simulation of its future development trend.It can also provide scientific guidance for optimizing regional land use planning,formulating sustainable development policies for the fruit industry,and promoting the quality and efficiency of planting industry.Fengjie County,Chongqing City is located in the hinterland of the Three Gorges Reservoir area,with a long history of orchard planting and has now developed into one of the local pillar industries.This paper takes Fengjie County as the research area,based on GF series satellite imagery,Landsat series satellite imagery,and various data sources such as DEM,economy,and population.First,according to the spatial morphology and image imaging characteristics of Fengjie orchard,a multi-scale cavity convolution module is designed and implanted into the traditional U-Net structure.It is proposed to construct a multi-scale hollow convolution U-Net model,which can further expand the receptive field and improve the ability of multi-scale feature recognition and fusion.Combine the feature selection algorithm of the RF classification model to complete the orchard spatial information extraction from 2000 to 2020,and compare the accuracy of the multi-scale cavity convolution U-Net with the RF,U-Net,Seg Net,Res Net and other models;Then,design a conceptual model of the driving force of the spatiotemporal evolution of the orchard.Construct a driving factor system framework covering the three dimensions of nature,social economy,and policy,and two spatial scales of pixel and township.Use the RF regression model to quantitatively analyze the relative importance and marginal dependence of driving factors,and explore the driving force mechanism of the spatiotemporal evolution of orchards in the past 20 years;Finally,three scenarios were set up,and the Markov-FLUS coupling model was used to simulate the development trend of Fengjie Orchard in 2030,and corresponding policy recommendations were put forward.The main findings are as follows:(1)Multi-scale cavity convolution U-Net improves the accuracy of orchard spatial information extraction.In this study,500 spots were randomly selected from the orchard extraction results in 2020 for field verification and proof sampling,and a field verification sample data set was constructed.Combined with the orchard label data set,the orchard extraction accuracy of the multi-scale cavity convolution U-Net model and the RF classification model are compared and analyzed.The results show that in 2013-2020,the ACC and Recall indicators of the multi-scale cavity convolution U-Net model for 6 years are better than the OA and PA indicators of the RF model.In the field verification accuracy in 2020,the Kappa coefficient and precision P of the multi-scale hole convolution U-Net model are 0.9214 and 91.69%,which are higher than the 0.9143 and90.98% of the RF classification model,respectively.At the same time,the accuracy of the multi-scale cavity convolution U-Net is compared with the traditional U-Net,Seg Net,Res Net and other deep learning models,and the accuracy of the model and the accuracy of field verification are respectively measured.The results show that multi-scale hole convolution U-Net has certain advantages.It can be seen that the multi-scale cavity convolution U-Net achieves higher accuracy in the orchard extraction process.(2)During 2000-2020,Fengjie Orchard expanded at a high speed.The 2013-2020 orchard spatial data extracted based on the multi-scale cavity convolution U-Net and the2000-2012 orchard spatial data extracted based on RF are integrated to construct the longterm orchard spatial distribution data from 2000 to 2020.On this basis,analyzes of the temporal and spatial evolution of Fengjie orchards in the past 20 years,the overall growth rate is about 87.47%,and the average annual growth rate is about 4.37%.From the perspective of spatial dimensions,Fengjie Orchards are mainly distributed along the Yangtze River and the main water systems of Meixi River,Daxi River,Caotang River,Zhuyi River and other major water systems,and are distributed in the first and second terrace areas on both sides of the river valley.The orchard gradually expanded along the direction of the basin.Among them,Anping Town,Zhuyi Town,Yongle Town,Baidi Town,Kangping Township and other townships along the Yangtze River occupy a dominant position and are the core producing areas of Fengjie Orchards.From the perspective of time dimension,the growth rate during the 2000-2006 period was relatively slow,at 9.93%;the growth rate during the 2006-2012 period increased,reaching 10.99%;2012-2020 has the fastest growth rate,reaching 53.66%,accounting for74.84% of the total growth scale,which belongs to the rapid development stage of Fengjie Orchard.(3)Orchard expansion is the result of the combined effects of natural,social,economic,and policy factors,and policy factors have played an important guiding role.Unlike some previous studies,this study adds a policy dimension on the basis of the two dimensions of nature and social economy.Three quantitative policy factors that special funds for poverty alleviation in the navel orange industry,special funds for modern agricultural citrus,and funds for technology projects in the navel orange industry were introduced.According to the analysis results of the RF regression model,the relative importance of the 10 driving factors included that the distance from water source,the distance from urban area,soil p H,the slope aspect in pixel scale and the average slope,the average elevation,the annual output value of the fruit industry,the special funds for poverty alleviation in the navel orange industry,the special funds for modern agricultural citrus,the funds for science,the technology projects in the navel orange industry in township scale are ranked in the forefront,and they play a major role in the expansion of the orchard.The relative importance of factors such as the total population at the end of the year,the standard deviation of the slope,and the distance from the national highway are relatively low,and their influence on the expansion of the orchard is not obvious.The relative importance of the three policy quantitative factors ranks 5th,6th,and 8th respectively,which proves that the policy factors play an important role in the expansion of the orchard.Appropriate policies need to be formulated to guide the healthy development of the fruit industry.(4)The RF regression model has powerful marginal dependence quantitative analysis capabilities,which can more accurately explain the driving mechanism of driving factors on orchard expansion.According to the results of the marginal effect analysis of the top 10 driving factors in relative importance,orchard expansion is affected by multiple factors,These effects exhibit significant spatial heterogeneity at different levels,and also have complexity and nonlinear characteristics.In general,orchard expansion tends to occur in areas with an average slope of 15°-25°,an average elevation of 200-450 m,a distance of less than 3km from the water source,and a soil p H of 5.5-7.6,with financial support and good location conditions.In this study,20% of the orchard sample data was used to compare the accuracy of the RF regression model with traditional models such as multiple linear regression and logistic regression.The AUC-ROC accuracy indicators were 0.82,0.79,and 0.80,respectively.This shows that the RF regression model has higher accuracy and stronger explanatory power.(5)The industry-ecological balance model is relatively more suitable for the actual needs of Fengjie’s development.Set up three scenarios of industry priority,ecological priority,and industry-ecological balance,and predict and simulate the spatial pattern of Fengjie orchard expansion in 2030 based on the Markov-FLUS coupling model.Analyzes the risk of ecological environment and the results show that under the scenario of industry-ecological balance,orchard production is expected to increase by 60-80 thousand tons,the orchard’s areas with extremely easy expansion account for 4.01% and easy expansion account for 4.54%,and the orchard’s areas is not easy to expand zone and the reverse expansion zone accounts for 91.45%.Orchards mainly expanded slightly along the Yangtze River and its tributaries,and their overlapping areas with high-risk areas of soil erosion and areas prone to geological disasters increased slightly.With the gradual increase in the economic benefits of the fruit industry,the ecological environment risks are within the controllable range,which will not pose a threat to the sustainable development of Fengjie and the Three Gorges Reservoir.In the three scenarios,it is relatively more in line with the actual development needs of Fengjie.In summary,this research focuses on the extraction method of orchard spatial information and the analysis of expansion driving factors.The innovations are mainly reflected in two aspects:(1)On the basis of improving the traditional U-Net model,a multi-scale cavity convolution U-Net deep learning model is proposed,which improves the accuracy of orchard extraction.Aiming at the problems of poor spatial scale adaptability and poor accuracy of the traditional U-Net model,a multi-scale cavity convolution module is designed.It can be implanted between the encoder and the decoder to expand the convolutional receptive field without increasing the number of volume nuclear parameters.The orchard spatial features at different scales in the image are extracted by combining the hole convolutions with different void ratios,and the extracted multi-scale features are merged,and finally the extraction accuracy of the orchard spatial information is improved.By comparing the accuracy with the RF model and the traditional U-Net,Seg Net,Res Net and other deep learning models,it shows that the multi-scale cavity convolution U-Net model has significant advantages in the accuracy of orchard extraction.(2)Three quantitative policy-type driving factors are introduced to participate in the analysis of the driving force of orchard expansion,and it is found that the relative importance of policy-type factors is at the forefront and is an important driving factor.In the conceptual model of the driving force,the policy dimension was added,and three policy quantitative factors were added: the special fund for poverty alleviation in the navel orange industry,the special fund for modern agricultural citrus,and the fund for the navel orange industry science and technology project.Based on the RF regression model,the relative importance of 27 driving factors was measured and ranked.The results showed that these 3 policy factors were ranked in the top eight and played an important role in guiding and promoting the expansion of the orchard.The multi-scale cavity convolution U-Net model and the conceptual model of driving force proposed in this study can provide reference for the research on the driving force of remote sensing extraction and spatiotemporal evolution of similar cash crops in mountainous areas.It can also have important scientific significance and application value for regional land and space planning,natural resource monitoring and management,and sustainable development policy research. |
PDF Full Text Request