Massey Documents by Type

Permanent URI for this communityhttps://mro.massey.ac.nz/handle/10179/294

Browse

Has files: YesSubject: search.filters.subject.Apple inspection

Search Results

Now showing 1 - 2 of 2
  • Thumbnail Image
    Item
    An investigation into apple inspection in colour space : a thesis presented in fulfilment of the requirements for the degree of Master of Engineering in Industrial Automation at Massey University, Palmerston North, New Zealand
    (Massey University, 2011) Caulton, Michael
    To maximise the storage life of packed apples, no damaged apple can be included because the ethylene produced by the damaged, rotting fruit can lead to damage in the surrounding fruit. The variability in size, shape and colour of apples has meant that the commercial inspection process has had to remain largely manual. In manual inspection, all surfaces of the apple are sequentially presented to human inspectors who remove deformed or blemished apples from the processing line. Most automated commercial inspection systems use infrared images to inspect the apples. These systems struggle to identify defects near the stem or calyx because of the reduced information in the monochrome images. In this work, the full visible light spectrum, via a full colour camera, was used to inspect apples in an attempt to isolate and identify stems, calyxes and defects. A Sobei filter and an islanding routine were used to find areas of interest in undamaged apple skin. This led to the identification of 92.7% of stems, 97.9% calyxes and 83.0% of defects. Image processing for defect identification took as little as 16ms per image; fast enough for implementation in a commercial application where speed is critical.
  • Thumbnail Image
    Item
    Analysis of the stochastic excursions of tumbling apples
    (1/09/2021) Flemmer C; Bakker H; Flemmer R
    There are strong economic pressures to improve automated inspection of apples. A considerable difficulty, acknowledged in the literature, but not adequately quantified, is the question of the extent to which the surface of apples, tumbling randomly on rollers, is covered by camera views during inspection. This work demonstrates a method to measure the roll, pitch and yaw of tumbling apples by tracking features on the skin between succeeding camera images and then to use the measured data to provide precise statistical descriptions of the tumbling process. The method was tested on an image library of four apple varietals; Eve and Granny Smith, which have mostly uniform skin colour, and Royal Gala and Braeburn which have a variegated skin colour. The images included apples that rotated stem-over-calyx (as the starting position) and apples that rotated equatorially for all varietals. The variegated varietals had many more trackable skin features (1,731–2,065 image pairs) than the mono-coloured varietals (238–859 image pairs) and stem-over-calyx rotation produced more tracking image pairs (723–2,065 image pairs) than equatorial rotation (238–2,041 image pairs), because the stem and calyx provided trackable features. Probability histograms are presented for the normalized incremental rotation in pitch, roll and yaw for each varietal and each direction of initial rotation. Skew-Gaussian distributions are fitted to the probability data to give the mean, standard deviation, skew and mean square error for the pitch, roll and yaw for each of the four varietals in each of two initial orientations (stem-over-calyx and equatorial). These stochastic characterisations can be used in future Monte Carlo simulations to provide precise determination of camera coverage during the inspection of apples tumbling on rollers. This is an important contribution to the field of automated apple inspection.