S18 - Session P3 - Construction of flower bud diagnosis system using AI image analysis in strawberry cultivation

Authors: Seiji Matsuo *, Yoko Takahashi, Chiharu Tokor, Masahide Isozaki

For the implementation of smart agriculture in the strawberry cultivation system, the construction of a system for diagnosing the flower bud state by image diagnosis using AI technology was examined. In the construction of the system, learning of flower bud images is important, but a large computational load, especially the load of annotation (labeling) work of cultivated images, becomes a problem. Here, we examined automation for improving these loads and the effectiveness of flower bud diagnosis for growth diagnosis by a diagnostic system using it. The specific approach is as follows. In the conventional annotation method, learning and verification of learning results are repeated, and at that time, annotation, which is an image labeling, is performed each time. In the proposed method, the image verified by the learning result and the label information recognized there can be used as they are for the next annotation. Therefore, the annotation work is greatly reduced each time the learning is repeated. We also considered the development of a fully automated annotation system using image extraction by color and background subtraction method. Furthermore, by visualizing the differentiation time of each fruit cluster using this proposed system, we constructed a program that enables diagnosis of flower growth. This is a system that diagnoses the differentiation time of fruit clusters by detecting the flowered flowers with yolo and recognizing the number and position by image analysis. The effectiveness of this diagnosis could be verified by analyzing the images of the field data using this system.