Shape-based target recognition using wavelet neural network

A novel shape-based target recognition method using the wavelet neural network (WNN) with tunable time-frequency parameters is developed. The WNN integrates the ability of local singularity analysis of continuous wavelet transform (CWT) into the structure of artificial neural network (ANN). Firstly, the maxima distribution across scale levels of CWT modulus of the object boundary normalized curvature representation is used to detect singular points along the boundary. Then feature extraction, selection and classification procedures are implemented by WNN automatically. Experimental results show that the classification performance of WNN, even in case of high noise, outperforms that of the wavelet neural network with fixed time-frequency parameters, conventional ANN and traditional Fourier descriptors-based minimum distance classifier.