S23 - Session O4 - A Computer vision system to estimate flesh colouration in red kiwifruit

Authors: Gianmarco Bortolotti *, Niccolò Raule , Letizia Manzoni, Mirko Piani, Luca Corelli Grappadelli, Francesco Spinelli, Luigi Manfrini

Computer vision is becoming widely used for estimating fruit appearance and defects. In the last years, many new fruit cultivars (apples, pears, kiwifruit) with red flesh have reached the market with positive feedback from the consumers. Computer vision may allow to estimate the red colouration of the flesh in these new varieties assuring a fast, robust and standardized evaluation of this trait. Our research aimed at developing a computer-based tool for objectively estimate red flesh surface area for red kiwifruit cultivar. The proposed solution applies firstly an object detection based on YOLOv5 CNN algorithm to detect sliced fruits, then multiple levels of colour filtering are applied to isolate red pixel in the image. The system allowed the estimation of red flesh area with a 90% agreement with visual inspection by skilled operators . Further studies, including red pigment extraction and quantification as well as comparison with colorimeter results, are needed to validate the results.