| Pepper(Zanthoxylum bungeanum Maxim)has the reputation of “Chinese spices” and it is one of the traditional “eight condiments” and also is one of the important economic crops.As the origin country of pepper,China is the country with the largest area and production of pepper.Jiangjin District in Chongqing is mainly planted with “Jiuyeqing”pepper,which is the largest “Jiuyeqing” pepper planting base in China.Its pepper industry has covered 25 towns(streets),involving 220,000 households and 610,000 pepper farmers.It is one of the important pillar industries and also is a key industrial industry with high-quality agricultural products,which is an important way to get rid of poverty.Therefore,timely understanding of the cultivation area and spatial distribution of pepper,to provide reliable data for the investigation and decision-making of the pepper industry,sequentially realize the information management of the pepper industry.It has a great significance to adjust the development of the pepper industry and assist the relevant departments in the formulation of agricultural guidance policies,and improve the status and sustainable development of the dominant region of pepper.The traditional information on pepper planting information is obtained by using agricultural statistical reports,and this method is reported and summarized through the field survey data of grassroots personnel.This method is greatly affected by environmental and human factors in the data acquisition process,and it is difficult to timely update and release in the reporting process.Compared with the traditional methods,using satellite remote sensing technology to monitor the spatial distribution information of crops has lots of advantages such as strong economic benefits,strong timeliness,strong macroscopicity,wide coverage and large amount of information.In recent years,with the continuous updating and advancement of remote sensing space technology,multi-sensor,multi-time resolution,multi-spatial resolution and multi-source remote sensing image data have been widely used in the research and application of various fields of agricultural remote sensing.However,the special environmental conditions such as climate,topography,hydrothermal and arable land distribution in mountainous hilly areas have brought many challenges to remote sensing scientific research and its application.Therefore,actively exploring the research of crop identification in remote sensing technology in mountainous and hilly areas and promoting the application of remote sensing technology in mountainous and hilly areas has become a scientific problem that needs to be solved at present.This study takes the xianfeng Town of Jiangjin District as the research area,together with the present situation of land use map and the field investigation to obtain sample data of the distribution of typical pepper.The 6 vegetation index data(Normalized Difference Vegetation Index,Weighted Difference Vegetation Index,Soil Adjusted Vegetation Index,Green Normalized Difference Vegetation Index,Atmospherically Resistant Vegetation Index,Global Environmental Monitoring Index)are calculated by the cloudless coverage Sentinel-2 satellite images on April 14,July 23,and October 31,2017.Through the digital elevation model to generate elevation,slope,aspect,slope length,slope height,terrain wetness index and other auxiliary topographic data,use the maximum entropy model(MaxEnt)method to identify the pepper in xianfeng Town of Jiangjin District.The results of the study were applied to the identification and estimation of the planting area of the pepper in Jiangjin District.The main research contents and results are as follows:(1)The performance comparison of multi-temporal remote sensing data with single-temporal remote sensing data and auxiliary terrain data in the model construction.The performance evaluation indexes(AUC)of the multi-temporal remote sensing data and the auxiliary terrain data combination construction model were respectively 0.890 for the training AUC and 0.866 for the test AUC,both higher than the AUC values of the single-temporal remote sensing data(April 14,July 23 and October 23)and the model constructed by the auxiliary terrain factors.The results show that the performance of multi-temporal remote sensing data and auxiliary terrain data is better.(2)The impact of model parameter settings on model performance.Regularization Multiplier(RM)is set to 1,1.5,2,2.5 and 3,using multi-temporal remote sensing data and composition of auxiliary topography as environmental variables,to compare the impact of different Regularization Multiplier parameter values on model performance.The results show that when the parameter value of Regularization Multiplier parameter is 1,the AUC values of training and test reach the highest,which are 0.890 and 0.866 respectively.With the increase of Regularization Multiplier parameter,the AUC value of training and test showed a decreasing trend.(3)The dominant factors affecting the geographical and spatial distribution of pepper were screened and their characteristics were analyzed.The selected environmental variables include multi-temporal remote sensing vegetation index and topographic factor,with a total of 24 variables.The influence degree of each environmental variable on the spatial and geographical distribution of pepper was tested and evaluated by the relative contribution rate.Finally,the environmental variables with a cumulative contribution rate of over 95% were selected as the dominant factor of the impact model.The results showed that the dominant factors were: the normalized vegetation index on July 23,the weighted difference vegetation index and normalized vegetation index on April 14,the weighted difference vegetation index on October 31,and the elevation,slope height,aspect,slope and terrain wetness index.According to the response curve of the dominant factor,its characteristics can be obtained as follows:,normalized difference vegetation index on July 23 is about 0.1-0.6,weight difference vegetation index on April 14 is about 0.20-0.35,the normalized difference vegetation index on April 14 is about 0.38-0.78,the elevation is about 210-430 m,slope highet is about 10 to 170 m,the weight difference vegetation index is about 0.20-0.33 on October 31,aspect is about 0-270 °,slope is about 0-39 °,terrain wetness index is about 4.2-9.0,when the dominant factor in characteristics within the scope of pepper probability of the existence of the above 0.5.(4)Comparison of classification accuracy under different threshold rules.Using MaxEnt software gives the commonly used threshold rules of XianFeng Town pepper prediction probability graph for binary classification,and through XianFeng Town 25% of pepper samples survey distribution(47)and the not-pepper samples(140)classification accuracy verification,validation data set results show that when the threshold selection rules set to Equal training sensitivity and specificity,to use threshold of 0.361,the highest classification accuracy,the overall accuracy of 85.03%,the Kappa coefficient is 0.61.(5)Remote sensing identification of pepper in Jiangjin District.Based on the spatial distribution data of typical pepper in Jiangjin District and the 9 dominant factors screened,the maximum entropy model method was used to identify the pepper in Jiangjin District.The training model AUC value was 0.897 and the test AUC value was 0.896.The Equal training sensitivity and specificity threshold selection rule was selected,and the corresponding threshold was 0.319.The probability map of the pepper in the Jiangjin District was binarized,and 25% of the pepper samples(116)and not-pepper pepper samples(648)were used.The verification data set was used to evaluate the classification accuracy.The overall accuracy and Kappa coefficient were 81.81% and 0.48 respectively.At the same time,the spatial distribution prediction of pepper in Jiangjin District was carried out,and the planting area of pepper was estimated to be 539,900 mu by grid calculation. |
PDF Full Text Request