A system for image measurement of shape information of leaves, named "RTL: RealTime Leaves", was developed. RTL is the first system to implement the image measuring method proposed by Shono (1995) and Shono et al. (1996) for extracting 3-dimensional shape information from multiple leaf images. RTL can, via Internet, use images saved automatically and periodically on distant Web-servers, and can extract time series of growth status indices using built-in neural networks.
RTL consists of several components as follows; (1)an image data logger, (2)a texture analyzer, (3)a neural network for calculating shape information, (4)a procedure for reconstructing 3-dimensional shape information, (5)a neural network for monitoring growth status indices. Furthermore, an external supporting subsystem for the construction of neural networks maintains the effectiveness of the neural networks against changes in leaf shape with growth. This subsystem can be used to interactively and easily define and train up the neural networks.
RTL is designed to function as a reliable doctor, who cautiously monitors plants' growth status indices to find signs of undesirable growth status, for example wilting. After finding such a sign, RTL can not only show a warning message, but also show a realtime plant image. This image is exaggeratedly deformed to reflect the recent index fluctuations, allowing easy and quick understanding of the situation.
From the results of an experimental trial using remote images of tomato plants, it was shown that RTL could complete its image measurement processes within about 6 minutes, and could properly realize its intended functions.
View full abstract