Research On Orchard Identification Based On HJ1A-HSI Hyperspectral Remote Sensing

With the development of the national economy and the improvement of the people's living standards,the orchard industry in China has developed rapidly.The area of orchards has continued to increase.The orchard industry has become an important pillar of the rural economy and has shown good prospects for development.Qixia City is an important orchard base in China.Qixia City,Yantai City,Shandong Province is an important orchard base in China.The area of orchards exceeds 20% of the total area of Qixia City.However,the location of the orchards is unclear and cannot be grasped in a timely manner.This is not conducive to the sustainable development of the fruit industry.In recent years,the rapid development of remote sensing technology can achieve the monitoring needs of the distribution of orchards.This study used domestic the HJ1A-HSI image to identify the orchard of Qixia City in order to provide a new technical method for extracting orchard distribution information.In this study,some townships in Qixia City,Shandong Province were used as research areas.Based on HJ1A-HSI and HJ1A-CCD,ENVI5.1 software was used to perform image preprocessing.The vegetation area of the study area in the HSI image was extracted by calculating the NDVI value of the CCD image.Based on ENVI and MATLAB platforms,the three feature selection methods,OIF,ASP+ABS,and SCP,were used to select the band of HSI data in the study area.Then the field survey data were used to extract the orchard distribution information in the study area using the MLC,NNC,and ISODATA classification methods.The accuracy of the recognition results was analyzed and evaluated.At the same time,based on the NNC classification method,the orchard information extracted from nonband selected images was extracted.Thus the necessity of band selection was proved and finally the orchard information extraction method was suitable for determining HJ1A-HSI data.The main research conclusions were as follows:(1)The vegetation area of HJ1A-HSI image was extracted.By calculating the NDVI image of HJ1A-CCD data,the vegetation area of the study area was extracted from the CCD image.The extraction of the vegetation area on the HJ1 AHSI image of the study area was realized,and the influence of the non vegetation region on the orchard recognition in the study area was eliminated.(2)After band selection,three new band sets were obtained.This study used OIF,ASP+ABS,and SCP to select bands for HJ1A-HSI data in the study area based on the ENVI and MATLAB platforms.The selection results of OIF were b1,b2,b109,b110,b111,b112,b113,b114,and b115;the selection results of ASP+ABS were b2,b3,b4,b46,b75,b76,b109,b110,and b113;the selection results of SCP were b4,b8,b15,b16,b41,b52,b84,b89,b112.(3)Hyperspectral image recognition of orchards were conducted in the study area.In this paper,the new images composed of the above three band selection methods were used to identify the orchards in the study area by using three classification methods: MLC,NNC,and ISODATA.The orchard distribution information were extracted,and the accuracy of the recognition results were analyzed and evaluated.The results showed that in these three classification methods,the NNC classification method had the highest accuracy in orchard recognition,the MLC classification method was followed,and the ISODATA classification method had the lowest orchard recognition accuracy.In the NNC classification method,the NNC classification effect based on the band selection method of SCP was the best,the overall accuracy was as high as 81.54%,and the Kappa coefficient was as high as 0.8006,which achieved a good classification effect.In the band selection method,the SCP band selection method had the highest recognition accuracy,the ASP+ABS band selection method was followed,and the OIF band selection method had the lowest recognition accuracy.Finally,the optimal method of HJ1A-HSI orchard recognition was determined as a combination of band selection method based on spectral characteristic parameter(SCP)and neural network classification method(NNC).The necessity of band selection for HJ1A-HSI orchard recognition were verified.By comparing the orchard recognition accuracy of the band-selected and non-bandselected HSI data,the orchard identification accuracy based on the three bands selection methods of SCP,ASP+ABS,and OIF were higher than that of the non-band selection.The overall accuracy and Kappa coefficient were increased by 13% and 0.156,8.46% and 0.101,3.85% and 0.049 respectively,and the improvement effect was more significant.This showed that the band selection methods in this paper improved the ability to classify and recognize of HJ1A-HSI hyperspectral images.The necessity of band selection for classification recognition using hyperspectral remote sensing was proved.