Cultivated Land Information Extraction And Fragmentation Degree Evaluation In Longkou City

Cultivated land is the most precious resource and the most basic means of production in agriculture.The fragmentation of cultivated land in our country reduces the utilization efficiency of cultivated land resource,limits the use of advanced equipment in agricultural production,causes the increase of agricultural cost,affects the large-scale agricultural production,and hinders the development of modern agriculture.Accurate and efficient extraction of cultivated land information,analysis of regional cultivated land fragmentation degree,improvement of cultivated land utilization efficiency,easing the contradiction between cultivated land protection and urban construction,has important practical significance for national food security,national economic development and overall social stability.This study took Longkou City as the research area,extracted cultivated land information by combining Gaofen-2 remote sensing image and deep learning method.On this basis,from the perspective of landscape pattern,the evaluation index system of cultivated land fragmentation was constructed from three aspects: scale fragmentation,shape fragmentation and spatial agglomeration fragmentation,to evaluate the comprehensive degree of cultivated land fragmentation in Longkou City.Finally,spatial auto correlation analysis was used to analyze the spatial agglomeration characteristics of cultivated land fragmentation and reveal the spatial distribution law of cultivated land fragmentation.The main research contents and results are as follows:(1)Cultivated land information extraction in Longkou City based on deep learning.Based on the 2020 Gaofen-2 image data of Longkou City,the image was manually labeled and the deep learning data set was constructed.Finally,the data set containing 12711 images and corresponding label files was constructed.Combined with the characteristics of cultivated land in Longkou City,the deep learning model Dense Net+Deeplab V3+ was constructed to train and extract cultivated land information in Longkou City.The results show that the model accuracy rate is 95.23%,accuracy rate is 93.03%,recall rate is 91.32%,and FI value is 0.92.The deep learning model completed by training was applied to the whole Longkou city,and125.34 km2 of cultivated land was extracted.The accuracy of the extraction results was evaluated,and the Kappa coefficient was 95.18%,which proved that the cultivated land extraction method was accurate and effective.(2)Evaluation of cultivated land fragmentation degree based on landscape pattern index in Longkou City.On the basis of cultivated land information extraction,9 landscape pattern indices were selected from the three aspects of cultivated land scale fragmentation,cultivated land plot shape fragmentation and cultivated land spatial agglomeration fragmentation based on village scale to construct an evaluation index system of cultivated land fragmentation degree,and a multi-factor comprehensive evaluation model was used to evaluate the cultivated land fragmentation degree in Longkou City.The results showed that the areas with high comprehensive fragmentation degree in Longkou were mainly distributed in the southern region,especially in Xiading,Qijia and Shiliang towns,which accounted for 8.63% of the total area of cultivated land.The regions with low comprehensive degree of fragmentation mainly concentrated in the northern and northwestern regions,accounting for 40.55% of the total area of cultivated land.(3)Analysis of spatial characteristics of cultivated land fragmentation in Longkou City based on spatial autocorrelation.Based on the results of cultivated land fragmentation evaluation,the global autocorrelation and local autocorrelation were used to analyze the spatial agglomeration characteristics of cultivated land fragmentation degree in Longkou City.The results showed that the comprehensive fragmentation index of cultivated land,the scale,shape and spatial agglomeration fragmentation index of cultivated land in Longkou City had positive spatial correlation.The global autocorrelation analysis of fragmentation index showed that Moran’s I index was 0.1824、0.0925、0.1272 and 0.2175。in order.In the P-value hypothesis test,the P-value values of the four degree indexes of fragmentation are all less than 0.05.According to the local autocorrelation analysis of the fragmentation index,the slope of the Moran scatter plot of the comprehensive fragmentation index of Longkou City is positive,and the sample points of administrative villages are mostly scattered in the thirdquadrant,showing a positive correlation on the whole.