Assessment of flower number per inflorescence in grapevine by image analysis under field conditions

Abstract

BACKGROUND Flowers, flowering and fruit set are key determinants of grapevine yield. Currently, practical methods to assess the flower number per inflorescence, necessary for fruit set estimation, are time and labour demanding. This work aims at developing a simple, cheap, fast, accurate and robust machine vision methodology to be applied to RGB images taken under field conditions, to estimate the number of flowers per inflorescence automatically. RESULTS Ninety images of individual inflorescences of Vitis vinifera L. cultivars Tempranillo, Graciano and Carignan were acquired in the vineyard with a pocket RGB camera prior to flowering, and used to develop and test the ‘flower counting’ algorithm. Strong and significant relationships, with R2 above 80% for the three cultivars were observed between actual and automated estimation of inflorescence flower numbers, with a precision exceeding 90% for all cultivars. CONCLUSION The developed algorithm proved that the analysis of digital images captured by pocket cameras under uncontrolled outdoors conditions was able to automatically provide a useful estimation of the number of flowers per inflorescence of grapevines at early stages of flowering. © 2013 Society of Chemical Industry

Image analysis‐based modelling for flower number estimation in grapevine

“…The measured results even outperformed for recall and provided similar precision results to those reported by Diago et al with the original algorithm's version, tested on three varieties and 15 images. These differences cannot be attributed to the modifications applied to the new implementation, since they were aimed at improving stability in the calculation of the required thresholds for the filtering criteria as previously described.…”

“…The algorithm for flower detection in the inflorescence images was implemented in MATLAB ( R2 010b, Mathworks, Natick, MA, USA) for processing batches of images in a fully automated way. This was an evolved version for stability improvement of the algorithm described in Diago et al …”

“…However, practically speaking, calculation of the flower number per inflorescence has been troublesome owing to strong operational limitations, as it has been traditionally assessed manually . Consequently, several methods for simplifying the flower‐counting procedure have been proposed . Poni et al used inflorescence images to manually count the number of visible flowers on photographic prints, while recent publications have relied on image analysis algorithms for automating the process.…”

“…Consequently, several methods for simplifying the flower‐counting procedure have been proposed . Poni et al used inflorescence images to manually count the number of visible flowers on photographic prints, while recent publications have relied on image analysis algorithms for automating the process. Nevertheless, these solutions only provide partial information, since an inflorescence is a three‐dimensional structure in which not all the flowers will be visible in an image due to occlusions from other flowers or inflorescence structure.…”

A new method for assessment of bunch compactness using automated image analysis

“…The shape features were calculated from the projection area and pixel accumulation curve. Recently, a new method for the assessment of the number of flowers per inflorescence by means of image analysis has also been successfully applied (Diago et al 2014a). Image analysis has been applied for the evaluation of different bunch and berry traits, including the number of berries per bunch, bunch mass, and berry size (Kicherer et al 2013, Cubero et al 2014, Diago et al 2014b).…”