Advanced satellite imagery to classify sugarcane crop characteristics

Techniques that provide a rapid and widespread assessment of crop properties equip industry decision makers with knowledge to improve their farming environment, both tactically and strategically. An interdisciplinary approach that links the fields of hyperspectral remote sensing, statistical data mining and sugarcane systems was undertaken to establish new relationships to determine variety type and crop age of sugarcane plants. In contrast to commonly used sensors such as those occupied by Landsat satellites, images captured by hyperspectral sensors can provide a more detailed assessment of crop status. Appropriate statistical analysis methods are needed to decode the multifaceted information recorded in these hyperspectral images. A range of statistical approaches have been applied for analysis of an EO-1 hyperion hyperspectral image from a major sugarcane growing region in Australia. Two relatively new classification methods - support vector machines and random forests - demonstrated superior performance in classifying sugarcane variety and crop cycle, e. g. the number of times that the plant has grown back after harvest, when compared against traditional statistical methods. Assignment results were further enhanced when classifications of pixels within sugarcane paddocks were aggregated to paddock classifications using paddock boundary information. Whilst the analysis methods of the hyperspectral data have been tested for the classification of variety and crop cycle, the potential application arenas for this type of imagery is both extensive and relatively unexplored. This type of data coupled with appropriate analysis methods will play a vital role in futuristic sustainable agriculture practices as this imagery becomes more accessible and as land managers and researchers become more aware of the types of decisions that hyperspectral remote sensing data can aid.