Wavelet neural networks for supervised training of multispectral data and classification of soil moisture

Artificial neural networks (ANN) constitute a powerful class of nonlinear function approximates for model-free estimation. Neural network models are characterized by topology, activation function and learning rules. The wavelet neuron model is obtained by replacing an activation function with wavelet bases in the traditional neuron model. The wavelet is a localized function that is capable of detecting some features in signals. A wavelet basis function is assigned for each neuron and each synaptic weight is determined by learning. Wavelet neural networks are used in this study to process remotely sensed data and classify soil based on its moisture content. To evaluate the effectiveness of the wavelet neural networks, a soil moisture data set consisting of 750 vectors, each with three components (surface temperature, brightness temperature at L-Band (TB-L) and at S-Band (TB-S)) and some remotely sensed images are evaluated in the experiments. A comparison with Backpropagation networks is investigated for the supervised training of remotely sensed data and classification of soil moisture.