Identification of weed/corn using BP network based on wavelet features and fractal dimension

Abstract

The aim of this study was to investigate weed/corn Back-propagation (BP) network discrimination method based on wavelet feature parameters and fractal dimension of young weed/corn image. In-filed images were taken under natural sunlight and various backgrounds, and five common weed species located corns fields were considered in this research. The obtained images were converted into gray level images on a black background by a color index (ExG – ExR). Energy values were calculated from wavelet coefficients by using two-level wavelet decomposed gray level images. Then the obtained seven energy parameters were used as input vector to construct BP network classifier. The results showed that monocotyledon and dicotyledon could be totally separated with 100% accuracy, whereas weed/corn could not be effectively separated. To improve identification accuracy, the fractal dimension of weed/corn image was added to the original input vector. The results of this experiment demonstrated that BP network classifier associated with seven wavelet energy parameters provided 77.14% recognition rate (correctly identify weeds and corns), whereas BP network classifier associated determined by wavelet energy parameters and fractal dimension achieved a better recognition rate 94.28%. Key words: Corn seedling, weeds, wavelet transform, energy, fractal dimension, identify.