present work, we used two ANNs. The task of the first ANN
was to segment the image of interest (Fig. I) in two regions,
that is, a region of white pixels (which comprises plastic
border pixels and insect pixels), and non-white pixels (which
comprises gray plastic pixels and background pixels). The
inputs for the both ANNs were: (a) the intensity of the target
pixel; (b) the mean of the intensity of 8-connected pixels (8
pixels surrounding the target pixel); (c) the mean of the
intensities of the pixels in a 3x3 kernel for the first ANN and
5x5 for the second ANN; and (d) the standard deviation of the
intensities of the pixels in a 3x3 kernel for the first ANN and
5x5 for the second ANN. The items (a) and (b) are related to
the training strategies reported in [8]. Since the image of
interest is more complex than that previous work, we provided
statistical information (c) and (d). Levenberg-Marquardt was
the algorithm used in the training phase [9]. The stopping
criterion was set to 10000 epochs or an MSE for the training
phase lower than 10-10•