Amélioration du tri de semences de chicorée industrielle (Cichorium intybus L.) sur base d’attributs mesurés par vision artificielle

The commercial production of industrial chicory seeds (cypselae) includes the postharvest elimination of nonviable seeds by non-destructive tools. For this purpose, two machine vision methods were set up for the detection of nonviable seeds: color vision and fluorescence imaging. In addition, the appropriate criteria to estimate the efficiency of seed sorting were defined using a sorting optimization curve which describe the trade-off between the quality improvement of a seed lot and the loss of material. The analysis of color images of 1,500 seeds of the ‘Nausica’ variety allows only the recognition of desiccated and undeveloped seeds. This is caused by the large natural variability of seed color, shape and texture. Fluorescence imaging was set up in order to analyze the repartition of chlorophyll fluorescence (CF), a marker of seed maturity, on different areas of the seed (pericarp and pappus). It comprised a monochromatic light source, a highpass filter and a monochromatic CCD camera sensitive to red and infrared. With this device, blue light reflected by the seeds was blocked whilst red fluorescence was measured by the camera. A segmentation algorithm was designed to estimate separately the fluorescence intensities of the pappus, a crown of scales, and the main body of the pericarp. Experiments were carried out on five clones of cross-pollinated chicory plants used for seed production and four series of 30 plants from 2 common varieties grown in different weather conditions. The flower heads were labelled at flowering and harvested at different times during the maturation process expressed in "days after flowering" (DAF). Germination tests were performed according to the recommendations of the International Seed Testing Association to measure the germination percentage (GP) and the germination rate (GR), an indicator of seed vigour. The dry weight (DW), water content (WC), maturation duration, germination percentage (GP), and germination rate (GR) were measured to estimate seed quality. For all experiments, the chlorophyll fluorescence diminished mainly between 16 and 33 DAF during the filling phase of seed development. The correlations between CF and maturity parameters (DW, WC) were comprised between 0.25 and 0.75 in absolute value for seeds that matured 16 to 44 days on the stalk. Sorting simulations were made on the basis of CF alone or in combination with the achene area density (AD) (mg/mm²). The CF was efficient to select seeds produced outdoor or in the greenhouse with the aim of increasing GP or GR but AD was an even better sorting feature. In the greenhouse, where developing seeds were submitted to temperatures higher than 25 °C, the use of CF in combination with AD improved the sorting efficiency.