Crop identification with wavelet packet analysis and weighted Bayesian distance

This study proposes a novel approach for crop identification by using wavelet packet transform combined with weighted Bayesian distance based on crop texture and leaf features. Automatic processing for agriculture produces requires accurate identification of crops to target plants for treatment according to their needs. Wavelet analysis, which features spatial/frequency localization, data compression, denoising, and data analysis/data mining, is a good candidate for identifying crops. If the energy of wavelet packet coefficients is the sole identifying characteristic, however, results may vary significantly depending on factors such as weather, plant density, growth stage, and sunlight. To overcome these variables, the weighted Bayesian distance was introduced for an identification criterion, also referred to as the decision distance, where the weighting is based on the statistic of crop texture and leaf shape. By utilizing the decision distance under different climates within three consecutive days of photography, the crop identification can achieve an accuracy of 94.63%.