Japan’s agricultural industry is facing a growing shortage of labor due to the country’s aging population and declining birthrate, and it is becoming increasingly difficult to pass on cultivation management that relies on experience and intuition. Therefore, there is a need to automate agriculture using AI and object recognition technology. However, a large amount of annotated data is required to train AI, and the lack of data is a challenge for practical application. As one of the most representative object detection models, YOLO (You Only Look Once) is widely used, but its performance declines under limited training data conditions. In this study, we propose a data augmentation method using fuzzy clustering to improve the accuracy of Yuzu fruit detection. In this study, We extend the dataset by extracting cluster information of Yuzu fruit using fuzzy C-means clustering, thereby generating new images based on them. In the experiments, we evaluated the detection performance of YOLO using datasets containing 5, 10, and 104 original images, both with and without the proposed data augmentation. The results show that the models trained with the augmented data achieved higher detection compared to those on the original datasets alone. The biggest improvement was 11.7% in F1 score and 6.8% in AP50. On the other hand, the results suggest that excessive data expansion has no effect on the model itself.