NARO has developed a method for diagnosing water stress in peach trees using Artificial Intelligence (AI) powered by deep learning and image analysis. This method enables image-based diagnosis even for large fruit trees by analyzing videos of the trees. By enabling simple estimation of the degree of water stress in trees, it becomes possible to irrigate at the optimal timing, which contributes to improving fruit quality and other cultivation outcomes. Furthermore, by expanding this technology beyond irrigation support to other aspects of cultivation management, it is expected that even those with limited experience in fruit tree cultivation will be able to make appropriate decisions based on data.
Overview
The decline in the agricultural workforce is a serious issue in Japan. To ensure a sustainable labor supply for the future, it is essential to create an environment that encourages new entrants and supports the recruitment and retention of workers. To enable even non-experts to manage fruit tree cultivation like experienced growers, the adoption of innovative smart agriculture technologies such as AI and Internet of Things (IoT) is considered highly effective. In response, we have developed a technology to estimate water stress in fruit trees by applying image analysis techniques based on deep learning, which has seen remarkable advances in recent years.
In peach trees, water stress affects fruit size and sugar content, making proper water management essential for producing high-quality fruit. Until now, irrigation has been performed empirically based on weather conditions and tree growth. However, if the water status of the tree, specifically stem water potential, is known, irrigation can be applied just before the onset of water stress (e.g., when the stem water potential reaches −1.4 MPa), enabling more precise water management. Measuring water stress in trees typically requires expensive specialized equipment and labor-intensive procedures, such as cutting leaves for measurement. However, if water stress can be diagnosed from images taken with a smartphone, both the labor and equipment costs can be significantly reduced.
To address this, we developed a technology that estimates water stress in peach trees by analyzing images captured with a smartphone using deep learning. By utilizing video footage, this method successfully enables image-based diagnosis even for large, three-dimensional fruit trees.
This technology is expected to allow even novice growers to perform optimal irrigation management based on the tree's water status, which was previously difficult to assess. As a result, it can help promote healthy tree growth and enhance fruit quality. This technology also has potential applications beyond water stress estimation, such as analyzing tree growth and fruit maturity through image-based diagnosis. Furthermore, it could be extended in the future to enable automated diagnosis using agricultural robots. By expanding the range of diagnostic targets in the future, this technology is expected to enable even those with limited experience to perform appropriate cultivation management based on data.
Publication
Yamane T, Habaragamuwa H, Sugiura R, Takahashi T, Hayama H and Mitani N (2023) Stem water potential estimation from images using a field noise-robust deep regression-based approach in peach trees, Scientific Reports. 13, 22359. https://doi.org/10.1038/s41598-023-49980-8
Related Information
Budget: JSPS KAKENHI Grant Number 21K05585
