Figure 3.(a) Actual profile of the sheet; (b) Processed profile obtained from the image.This signal contains the information on the deformation of the sheet. Therefore, there is a correspondence between the significant points of this signal (Figure 3b) and the significant points of the sheet deformation (Figure 3a). These points are:Point 0: start point of the imperfection. At this point there is a change of curvature in the sheet (Figure 3a), changing the slope sign. In the processed profile (Figure 3b) it results in an inflection point.Point 2: point of maximum height of the defect (Figure 3a). In the processed profile (Figure 3b) it is equivalent to an inflection point at which the gradient is maximum. The information from this gradient is related to the height or severity of the imperfection.

Point 4: end point of the imperfection. At this point there is a change of curvature in the sheet (Figure 3a), changing the slope sign. In the processed profile (Figure 3b) it results in a new inflection point.4.2. Profile ObtainingTherefore, an algorithm has been implemented that takes into account the specific characteristics of the profiles obtained from the image. This algorithm processes the profile signal, detecting the maximum, the minimum, and the inflection points and it stores the information. The steps followed by the algorithm are:Image filter: for the direction in which the profile is going to be analyzed, a box is set in the image and a media filtering is performed around this direction with a width between 5 and 10 pixels. This way the effects of the illumination system are reduced.

Signal filter: to eliminate the noise in the signal obtained from the profile, a filter based on wavelets is applied. It is a smooth filter based on a heuristic variant of the Stein risk principle with a rescaled threshold depending on the noise level of the signal. The problem with this filter is that the filtered signal is affected Carfilzomib by the length of the signal, especially if a big length is taken in which the initial and final ends do not contain information of the imperfection and generate a high component of noise. To avoid this problem, the signal filter divides into two stages:(a)In a first stage the signal is filtered by the wavelet filter and the start and end points limiting the useful information of the profile are
Generally, the quality of oil palm fruits is categorized based on the texture, shape and color of the fruit [1].

Currently in Malaysia, the human expert grading approach is used to inspect the maturity of oil palm FFB and classify them for harvesting. Typically, the color of the surface of the fruit and the number of loose fruit drops from bunches are the two main factors that guide the judgement of human experts [1,2]. In practice, this type of grading method often leads to mistakes where there is high potential to grade the fruit wrongly.