Pre-harvest Sugarcane Yield Estimation Using UAV-Based RGB Images and Ground Observation

Sugarcane supply can vary according to the cultivation area, climatic condition, and disease. Although there are several scientific simulation models for sugarcane yield estimation, they are not widely and efficiently used due to a large number of data requirements. The success of yield estimation using remote sensing and aerial observation was limited due to the insufficient spatial requirement and spectral requirement. This study introduces a technique to use UAV-acquired RGB images coupled with ground information for reliable and fast estimation of sugarcane yield for two popular varieties (KK3 and UT12) in Thailand. The first challenge was to discriminate sugarcane and non-sugarcane pixels in the images. For this purpose, both object-based image analysis (OBIA) and pixel-based image analysis techniques were investigated. The results revealed that OBIA technique (GLCM) could determine sugarcane pixels with 92 and 96% accuracy while pixel-based ExG method had accuracy of 84 and 88% for both the varieties. After identification of sugarcane pixels, the numbers of stalks, average height, and weight data were collected from 30 random sample points (size of 2 m × 2 m) from each variety. Using natural break method sampling, classes were created based on pixel value and number of stalks. The yield was finally estimated from sugarcane pixels using ground data and compared with harvested yield of both varieties. The object-based method produced the best result followed by pixel-based and traditional technique to estimate the yield. The very high spatial resolution of UAV image and advanced image classification of OBIA demonstrate significant potential for the farmers and related industries to predict yield before harvest.