Crosswind Measurements through Pattern Recognition Techniques

Optical devices currently used for crosswind measurements are calibrated and results interpreted on the basis of theoretical predictions. However, under strong turbulence conditions, the experimental observations do not compare well with the theoretical predictions. This deficiency can be overcome by use of a learning machine which utilizes a pattern recognition technique. Basically this approach substitutes observed experience for a detailed knowledge of the physical model. The technique utilizes a minicomputer to process observed spectral profiles. In the training phase spectral features, which are weakly intercorrelated but strongly correlated with the crosswind, are selected and classified with known winds. These data are stored in the machine. During the execution phase, the observed spectral features are compared with the previously stored classifications to yield a best estimate of the crosswind. Experimental results are presented which quantify the performance of this method.