Strawberry flower and fruit detection based on an autonomous imaging robot and deep learning

Abstract

The number of flowers, immature strawberries, and mature strawberries is good indicators for yield estimation and prediction. This study used an autonomous imaging robot for automated image acquisition on four different days between January and March 2021 (January 22, February 24, March 3, and March 10). A total of 565 images collected on January 22 were labeled for training and testing the YOLOv5. The mean average precision of the model was 0.826. Then, the trained YOLOv5 model was used to count the total number of flowers, immature fruit, and mature fruit. The average counting accuracy was 0.853, 0.811, and 0.859 for the three image datasets collected on February 24, March 3, and March 10, respectively. This study showed the great potential of using the deep learning model and the autonomous imaging robot for automated detection of strawberry flower and fruit.