The best phase and optimal identification method for identifying pomegranate trees

Abstract：【Objective】It is of great significance in production and management of fruit industry to quickly and accurately realize the planting information (i.e. planting region, planting area and planting structure) of various fruit trees in large areas by using the remote sensing technology. The present paper tries to find out the best phase and optimal method for identifying pomegranate trees by using monthly GF1-WFV images collected in 2015, 2017 and 2018 (a total of 42 images),in order to provide a theoretical basis for the remote sensing monitoring of pomegranate trees in Guanzhong and other fruit growing regions in China in the future.【Methods】Firstly, the images of each period were preprocessed (including image space clipping, image radiation calibration, image atmospheric radiation correction, image geometry correction, image mean filtering, etc). Then, in order to find out the best identification phase and method, the identification efficiencies of six methods (comparison of reflectance spectra of ground objects, bands difference or ratio analysis, analysis of spectral index identification, spectral index change tracking, image compound and analysis of multi-indices combined use) were tested separately by usingROI (Region of Interest) data, and these data were collected from the sample plots of 13 main crops (including pomegranate trees, grape vines, peach trees, cherry trees, mulberry trees, persimmon trees, apricot trees, pear trees, apple trees, walnut trees, plum trees, star fruit trees, winter wheat or summer maize). Finally, the identification efficiency of exploration results was verified in global image.【Results】(1) Because pomegranate trees were in the early stage of new shoot growth in mid-April, their leaves were small and narrow, and their canopy coverage and leaf area index were both small, the pomegranate trees had relatively low NDVI values during this period. The NDVI thresholds could be used to distinguish pomegranate trees from other 9 types of fruit trees and wheat. However, NDVI thresholds could not distinguish pomegranate trees from grape vines, peach trees, urban and rural green vegetation, abandoned land with grass, young fruit trees, etc. (2) In mid-May, the pomegranate trees that were in full bloom had a relatively high 1/Rb1-1/Rb3 value. Using 1/Rb1-1/Rb3 thresholds, not only the pomegranate trees could be distinguished from most other fruit tree species and wheat, but also urban and rural green vegetation, abandoned land with grass, and young- fruit trees could be correctly identified and largely eliminated. However, influenced by many unfavorable factors (For example, the complex combination of many ground objects, the existence of mixed pixels, the existence of the phenomenon that different ground objects have the same reflectance spectrum, the limited accuracy of image radiation correction, the lower spatial resolution of the used images, and the tendency of threshold setting), a few of ground objects like fruit trees (such as apple trees, pear trees and so on) and non-fruit trees were confused with pomegranate trees. If the NDVI threshold of this period was used, some non- pomegranate tree features can be removed. (3) The accuracy of identifying pomegranate trees was more ideal when the above two phase of images were compounded and triple thresholds of (1/Rb1-1/Rb3) mid-May, NDVI mid-May and NDVI mid-April were used. The correct recognition rate of pomegranate trees could reach 86.28%, and the overall classification accuracy could reach 85.37%. (4) The best phase to identify pomegranate trees from remote sensing images was in mid-May, followed by mid-April. Fruit maturity is not the best time for identifying pomegranate trees. The reason for this is that there are not only pomegranate trees but also apple and persimmon trees in the studying area with green leaves and red fruits. And it is difficult to identify pomegranate trees due to the influence of leaf color transformation of some grape vines, kiwi vines and apricot trees at this time. In other periods except the above period, the reflectance spectra of pomegranate trees are not very different from those of other fruit trees, so it is difficult to distinguish them.【Conclusion】When satellite remote sensing images were used to monitor and identify pomegranate trees in Guanzhong fruit growing area, the remote sensing images in mid-April and mid-May (The spatial resolution of these images should be less than or equal to 16 meters, the spectral resolution of these images should be more than or equal to four bands, and these images have blue, green, red and near infrared bands.) could be purchased. On the basis of radiometric calibration, atmospheric radiometric correction, geometric correction, registration and recombination of these images, the accuracy of identifying pomegranate trees could be more ideal when the images of mid-April and mid-May were compounded and triple thresholds of (1/Rb1-1/Rb3) mid- May, NDVI mid- May and NDVI mid- April were used. The above results could be applied to other fruit production areas where crops planting structure and natural environmental conditions are similar to the studying area. Based on these results, the monitoring period should be brought forward or postponed and spectral index methods could be chosen when pomegranate trees are monitored by using the remote sensing images of the fruit production areas with simple planting structure and great difference in natural environment conditions from the studying area.