Skilled farmers generally evaluate the growth of a crop by observing the crop's appearance. However, with the increasing digitalization of agricultural practices, a numerical index representing the crop's growth status is strongly needed. For this purpose, the speaking plant approach (SPA), a sophisticated strategy for environmental control in greenhouses, has attracted a lot of attention. The first and most important step in SPA is to obtain physiological information from a plant and then judge whether the plant is healthy. The daily measurement of plant growth is therefore important in SPA, particularly for environmental control. Various imaging techniques have been studied for this purpose. A robotized chlorophyll fluorescence (CF) imaging system that evaluates daily changes in the photosynthetic function of a tomato canopy was developed in our previous study and later commercialized. In this study, for the measurement of the daily stem elongation of tomato plants, we developed a deep learning model that detects a tomato plant's shoot apex in a CF image. YOLOv3, a representative object detection algorithm, was used for the shoot apex detection model. Images captured at Ehime University were used for training and validating the model. The developed shoot apex detection model was applied to two different commercial greenhouses (Asai Nursery and Agrimind; largest greenhouses in Japan) to measure the daily stem elongation of tomato plants. The developed model detected the shoot apexes with an average F-measure of 0.97 for the greenhouse from which the training dataset was derived (Ehime University test dataset) and with average F-measures of 0.69 and 0.82 for the two commercial greenhouses (Asai Nursery and Agrimind test datasets). Then, the 19 days and 26 days consecutively measurements were conducted, and the daily stem elongation of tomato plants was measured. The obtained results indicate that the developed deep learning model is effective for measuring the daily stem elongation of tomato plants in commercial greenhouses.
Read full abstract