Machine Vision System for Automatic Quality Grading of Fruit

Abstract

Fruit and vegetables are normally presented to consumers in batches. The homogeneity and appearance of these have significant effect on consumer decision. For this reason, the presentation of agricultural produce is manipulated at various stages from the field to the final consumer and is generally oriented towards the cleaning of the product and sorting by homogeneous categories. The project ESPRIT 3, reference 9230 ‘Integrated system for handling, inspection and packing of fruit and vegetable (SHIVA)’ developed a robotic system for the automatic, non-destructive inspection and handling of fruit. The aim of this paper is to report on the machine vision techniques developed at the Instituto Valenciano de Investigaciones Agrarias for the on-line estimation of the quality of oranges, peaches and apples, and to evaluate the efficiency of these techniques regarding the following quality attributes: size, colour, stem location and detection of external blemishes. The segmentation procedure used, based on a Bayesian discriminant analysis, allowed fruits to be precisely distinguished from the background. Thus, determination of size was properly solved. The colours of the fruits estimated by the system were well correlated with the colorimetric index values that are currently used as standards. Good results were obtained in the location of the stem and the detection of blemishes. The classification system was tested on-line with apples obtaining a good performance when classifying the fruit in batches, and a repeatability in blemish detection and size estimation of 86 and 93% respectively. The precision and repeatability of the system, was found to be similar to those of manual grading.

Introduction

The use of machine vision for the inspection of fruits and vegetables has increased during recent years. Nowadays, several manufacturers around the world produce sorting machines capable of pre-grading fruits by size, colour and weight. Nevertheless, the market constantly requires higher quality products and consequently, additional features have been developed to enhance machine vision inspection systems (e.g. to locate stems, to determine the main and secondary colour of the skin, to detect blemishes).

Size, which is the first parameter identified with quality, has been estimated using machine vision by measuring either area (Tao et al., 1990; Varghese et al., 1991), perimeter (Sarkar & Wolfe, 1985) or diameter (Brodie et al., 1994). Colour is also an important quality factor that has been widely studied (Singh et al. (1992), Singh et al. (1993); Hahn, 2002; Dobrzanski & Rybczynski, 2002). Some fruits have one colour homogeneously distributed on the skin surface, which we call primary colour. The averaged surface colour is a good quality indicator for these fruits. However, other fruits (e.g. some varieties of peaches, apples, tomatoes) have a secondary colour that can be used as a good indicator of maturity. In this case, it is not possible to rely only on the global colour as a quality parameter.

In oranges, peaches and apples there is an interest in detecting long stems in order to avoid damage to other fruit, or because their absence could imply a quality loss. Several solutions have been proposed to determine the position of the stem, such as: the use of structured lighting to detect concavities in apples (Yang, 1993); colour segmentation techniques to differentiate the calyx and stem in citrus fruits (Ruiz et al., 1996); or the study of light reflection in apples (Penman, 2002).

Sometimes, the stem can be confused with defects or blemishes on the skin. Damage and bruise detection is a crucial factor for quality evaluation. One of the first approaches for bruise detection in apples was based on the use of interferential filters (Rehkugler & Throop, 1986). Other studies treated blemishes together with colour estimation (Miller & Delwiche, 1989; Lefebvre et al., 1994; Cerruto et al., 1996; Leemans et al. (1999), Leemans et al. (2002); Blasco & Moltó, 2002). More recent techniques combine infrared and visible information to detect blemishes (Aleixos et al., 2002) or use hyperspectral imaging (Peirs et al., 2002).

The aim of this work is to report the image analysis techniques developed in the project ESPRIT 3, reference 9230 ‘Integrated system for handling, inspection and packing of fruit and vegetable (SHIVA)’, which is described elsewhere (Moltó et al. (1997), Moltó et al. (1998)), and the results achieved in the test performed during March 1998 at the Instituto Valenciano de Investigaciones, Agrarias (IVIA). The vision system was developed for on-line measurement of several parameters related to the quality of oranges, peaches and apples, such as size, identification of secondary colour spots (required for some varieties of peaches and apples), stem location or presence of blemishes. The fruits had to be inspected in four different views in less than 1 s. In order to evaluate the efficiency of the vision system, the performance and repeatability of the automatic inspection were compared with a manual inspection made by experts.