To ensure a stable supply of vegetables at low prices, standards have been established for shape and size. However, a certain quantity of substandard vegetables is typically produced due to unfavorable soil and weather conditions during cultivation. These not only increase the cost of production for growers but can also become an environmental burden when they are discarded. The issue presented is how to determine the size of each vegetable and when to harvest them.
Associate Professor Wei Guo and Project Professor Masayuki Hirafuji of the Graduate School of Agricultural and Life Sciences at the University of Tokyo have developed a system to automatically estimate and predict the size of all broccoli plants in a given growing area. As the plants tend to vary in size, a deep learning model is applied to detect the location of all plants and classify areas with flowers and buds based on aerial images captured by a drone. The system automatically estimates and predicts the size of all plants. Conventional aerial drone photography was performed to verify the effectiveness of the proposed approach by estimating the size of more than 10,000 broccoli plants grown over two years in a field greenhouse. As a result, the sizes of many of the plants were estimated with an accuracy of approximately 2 to 3 cm.
The growth model was combined with weather forecast data to predict the size of flower buds up to approximately 10 days later. By combining information on the growth of broccoli and shipping prices for each size, they calculated the total shipping price each day to obtain the producer's expected income. It was found that deviating from the optimal harvest date by a single day could increase the number of grade-out vegetables by approximately 5 percent and reduce income by approximately 20 percent.
A point of note was that this study was conducted under the condition that all individual vegetables in the field were harvested at the same time. However, this system can also be applied to a variety of open-air vegetables, such as cabbages, and further development of this system is expected to support efforts to realize sustainable agriculture.