Abstract
Determining the flowering stage accurately is crucial for successful artificial fertilization. However, visually identifying the stage requires skill and is subjective. To address this, we developed a simple method for determining the developmental stage of female strobili of Pinus thunbergii using deep learning. A classification model was created, and an associated web application was developed, eliminating the dependency on human observation. The process involved several skilled investigators classifying various images of P. thunbergii female strobili into stages I, II, and III. From a total of 3,074 images with unanimous evaluations, we used MobileNetV2's transition learning to construct and evaluate the model. Although the model had a high accuracy rate of 0.974 and an F-score of 0.949, i.e., a balanced evaluation of precision and recall, it failed to predict some images correctly. Specifically, it struggled with images containing small female strobili in relation to the entire screen, as well as images that included unrelated objects. Additionally, it had difficulty with female strobili that were yellowish-green in color. However, despite these limitations, we propose that this tool can be useful for evaluating traits in the field.
