ISHS Acta Horticulturae 1037: International Symposium on New Technologies for Environment Control, Energy-Saving and Crop Production in Greenhouse and Plant Factory - Greensys 2013
BASIC STUDY ON NON-DESTRUCTIVE GROWTH MEASUREMENT OF STRAWBERRY PLANTS USING A MACHINE-VISION SYSTEM |
| Authors: | S. Yamamoto, S. Hayashi, S. Saitoh, Y. Ochiai , S. Tsubota, Y. Takeshima, K. Imou, K. Saga |
| Keywords: | growth information, movable bench, depth sensor, 3D measurement, green illumination, spectral reflectance |
| DOI: | 10.17660/ActaHortic.2014.1037.83 |
| Abstract:
The movable bench system is one method of increasing area productivity and minimizing the input energy needed for heating and lighting in a greenhouse. In the movable bench cultivation system, all plants pass through a single location, termed the access point, for watering every day. The access point should thus be the ideal place to measure precise growth information. Machine vision is a promising technique for efficiently monitoring all plants at the access point, since it is non-destructive. This study examines some basic techniques for developing a non-destructive growth measurement system for strawberry plants cultivated on the movable bench system using machine vision. First, an experimental system was developed for 3D measurement of strawberry plants cultivated on a 1-m long bench. It consisted of a depth sensor, a source of illumination and a wheeled platform that ran on rails and carried the bench. To obtain 3D growth information, the experimental system tiled line image output from the depth sensor while the bench passed below. A noise reduction method for the tiled images was also developed. Ten benches were tested and their 3D information was reconstructed; the width and height profiles were then calculated. 3D data were evaluated for accuracy of maximum height and width measurements. The average estimation error was 3.7 cm for width and 1.6 cm for height. Second, the characteristics of spectral reflectance of strawberry leaves were investigated, and an algorithm was developed for new leaf detection using an ordinary color camera and green illumination. After four potted plants were monitored for a month, new leaf detection accuracy proved to be 74.2%. | |
Download Adobe Acrobat Reader (free software to read PDF files) | |
