||Pear pollination is generally done by artificial pollination, which requires the collection of pollen. Pollen collection is hard work, requiring long hours at high altitudes, and in recent years, much of the pollen is imported. However, if a disease occurs in a pollen exporting country, imports are halted, resulting in a shortage of fruiting material. To solve this problem, it is necessary to mechanize the pollen collection process and strengthen the domestic supply and demand system. In this study, we propose an AI(Artificial Intelligence)-based method for estimating the optimal timing of pear pollen collection. Specifically, we use YOLO(You Only Look Once), a deep learning-based object detection algorithm, to detect pear blossoms on photographed branches by classifying them into five stages, from bud to blossom. Based on the number of flowers in each bloom stage detected and the average amount of pollen per flower, the amount of pollen collected per branch is calculated, and the optimal time to collect pollen is estimated. In this paper, we report on our evaluation of the performance of AI pollen collection estimation and our study of how to estimate the optimal time to collect pollen.