Crop Classification Using Polarimetry of the Scattered Waves1
Peerapong Uthansakul and Chatchai Waiyapatanakorn2
Department of Electrical Engineering, Graduate School,
Chulalongkorn University. Bangkok, Thailand
..
There have been a significant number of researches on the use of scattered waves polarization states for crop classification. However, most of the researches have their specific techniques for data analysis, which are neither standard nor amenable to other work. Thus a polarimetric measurement system is developed and comparative study of much analysis technique are carried out. A novel classification index is also proposed herein.
The system developed is ground-based quasi-monostatic and operates in the range from 3.3 - 4.2 GHz. Four combinations of polarization states can be achieved, i.e. HH, HV, VH, and VV. The system is used to collect data from 7 types of crops: corn; sorghum; soybean; mungbean; sesame; sunflower; cotton. The measure data are calculated to form the four components of scattering matrix useful for differentiating all crops. Five data analysis techniques are employed : 1) mean values and variance of each component of the scattering matrix 2) mean values of two components of the scattering matrix 3) degree of polarization 4) polarization state on Poincare sphere and 5) co-polar and cross-polar gap. The last technique is the new one proposed in this work.
Results from data analysis show that all 5 techniques are capable of classifying all types crops. Data associated with each crop tend to cluster together as a group. Hence it is convenient to identify one from another. It is apparent that the use of co-polar and cross-polar gap yields the best result in term of classification capability. However, this technique requires data of the full frequency band. For the other 4 techniques, it is found that use of the mean of SHH and SVH gives the best result, but it does not provide information on the scattered waves behaviour compared with the use of degree of polarization or the polarization state which has information on polarization properties of the scatter waves. It is found that the performance of all classification techniques range from 47 out of 70 to 70 out of 70 measured spot. It is, therefore, concluded that the polarimetric measurement system developed functions satisfactorily and effectively. In particular, It is worth noting that the co-polar and cross-polar gap is a very interesting novel classification index.
Engineer