Improving the accuracy of separation the crops cultivated area using integration of multi-temporal radar and optical sentinel images and machine learning algorithms

Mostafa Kabolizadeh ¹ ( RS and GIS department, Earth science faculty, shahiud Chamran university of Ahvaz )
Kazem Rangzan ² ( RS and GIS department, earth science faculty, Shahid Chamran university of Ahvaz )
klhalil habashi ³ ( MS.C COMBAT desertification )

Submited date : 2022-10-24 Accepted date : 2022-12-05

Keywords: crops classification, Radar images, Khuzestan province, remote sensing, image integration,

Abstract :

In order to achieve food security, timely, accurate and repeatable monitoring the cultivated areas is necessary. In this regard, the present research aimed to improving accuracy of separation the crops cultivated area in the northeastern region of Ahvaz. To achieve the research goal, based on the available data, three groups of time series combination were created. The first group includes the combination of Sentinel 1 and 2 time series along with the NDVI index for the entire period, the second group of Sentinel 1 and 2 time series combinations based on the peak greenness and the third group were single images of Sentinel 1 and 2 in the peak greenness period. Then, images were classified using ML and SVM algorithms. Finally, crops cultivated area maps were prepared. the accuracy of the obtained results was evaluated using overal accuracy indices and Kappa coefficient. Based on the obtained results, it was found that the combination of the time series of Sentinel 1 and 2 images along with the NDVI index for the entire period (Combination No. 3) using the SVM method to extract the cultivated areas of the study area has the highest overall accuracy and Kappa coefficient, which is 91.22 and 0.89 percent respectively. Also, the obtained results indicated that the SVM algorithm has the highest overall accuracy and kappa coefficient for time series combinations, and for single image methods, the ML algorithm has the highest overall accuracy and kappa coefficient. Based on the findings, it is concluded that the combination of the time series of Sentinel 2 images and the SVM template for extracting cultivated areas have high accuracy compared to the single image method, and combining the VH polarization of Sentinel 1 to the time series of Sentinel 2 improved the accuracy about 5%.