On image analysis for harvesting tropical fruits

Intelligent harvesting of tropical fruits requires image analysis on a natural background, which more complicated than that on a unique color background. In some case, we can easily distinguish fruit areas in a natural background image by color. However, finally, we have to use shape analysis for identification and to obtain exact boundary positions. This research aims to detect position of fruits by shape and by using elliptic Fourier descriptors to describe shape. Then, we deform the typical shape in the spatial frequency domain by scaling, rotating, and phase shifting and matching with the image to obtain the maximum likelihood. However, the matching process takes a long time owing to the inverse Fourier series and parameter deformation. To optimize these problems, we use FFT for inverting the typical shape and a genetic algorithm for searching the maximum likelihood because this is a problem dependent on the natural environment. The genetic algorithm can decrease position-searching time.