Comparing the accuracy of different image processing methods to estimate sugar beet canopy cover by digital camera images

In this study, digital photography was used to estimate the amount of sugar beet’s canopy cover. For this purpose, a dataset of visible images of sugar beet crops, during the growing season, in 2018, under drought and nitrogen stress were taken in a greenhouse at the ETH research station for plant sciences in Lindau Eschikon, Switzerland. The treatments of this research included two levels of irrigation stress (low water and sufficient water) and three levels of fertilizer stress (20, 40, and 80 kg/ha nitrogen). Image discrimination and threshold algorithms are applied to perform segmentation on the images in Python. Compound segmentation methods using Excess Green, Excess Green minus Excess Red discrimination vegetation indices (plant from soil and background), and without discrimination index and manual input thresholding and Otsu and Triangle automated algorithms were used. Therefore, nine different compound methods including discrimination and thresholding algorithms used to estimate the canopy cover under different stresses. Results showed that compound methods of Excess Green minus Excess Red vegetation index and manual input thresholding and Excess Green Index and Otsu have the highest accuracy, 94.69 and 87.52 percent, respectively. The method without discrimination index and triangle thresholding which has 53.18 percent accuracy was the least accurate method.