Fully Automatic Controlled Environment Agriculture using Machine Learning based Plant Size Estimator

Abstract

Due to climate change, the global food supply crisis has become an urgent international problem. Given the circumstances, a controlled environment that artificially adjusts the climate for agriculture has attracted considerable attention as a solution to the problem. Various kinds of research have been proposed to develop the technology of controlled environments. However, the accuracy and scalability problem of these methods is a burden for the expansion to real-world scenarios. In particular, it is necessary to research and implement the computer vision-based algorithm, which is the key technique that enables the controlled environment system to be fully automatic. To solve the aforementioned problem, I propose a novel controlled environment agriculture system. The proposed system is composed of a plant life cycle regression module and a device control module. The system predicts the actual size of the plants and outputs the life cycle indicator of the plant, which is the growth rate of the plants. Based on the life cycle indicator, the device control module adjusts the essential factors, such as the amount of water and the strength of UV (Ultra Violet) light, for photosynthesis. As the proposed system is aware of the life cycle of plants, it can provide fully automatic controlled environments. I also propose and demonstrate the application machine to show how the proposed method can be applied to the real world. Through the experiments, it is shown that the proposed PLCR outperforms the existing state-of-the-art methods on the COCO dataset.

Plant phenotyping relevance

植物サイズと成長率を画像ベースで推定する手法が研究の中心であり、制御環境農業への応用も示しているため、植物フェノタイピング手法として採用する。

Code and data availability

The paper uses the generic public COCO dataset (cited prior work, not a paper-specific deposit) and provides no availability statement, URL, or deposit for its code, trained model, or plant images. No qualifying paper-specific public assets exist in the supplied blocks.

No evidence-backed public reproduction asset is currently recorded.