In-field fruit size monitoring is useful for estimating fruit maturation and size distribution, making harvesting and marketing decisions, estimating profit, and controlling taste. Many algorithms have been proposed for size estimation, but estimation accuracy for occluded fruits (e.g., by branches and leaves) remains low. In this research, a method of estimating pixel area for occluded circular masks was developed. This method involves distant pixel estimation and radius of curvature estimation. This method was evaluated via application to randomly generated occluded circular masks. The results showed that this method can decrease root mean squared error (RMSE) by an average of 87.1% for generated occluded masks. This method was then applied to realistic segmentation masks generated from in-field snapshot data as a practical application. Snapshots of peach fruits were taken with an infrared camera once every 30 min at night during the growing season. The sizes of the fruits were then estimated by the following method. First, masks were detected using Mask-RCNN, which is a popular method for instance segmentation. Next, assuming a circular fruit, the developed size estimation method was applied. Estimated results for actual images were evaluated using a logistic model tuned by an annotated mask area, which showed a maximum of 9.55% mean absolute error (MAE)-based size improvement. This size estimation method can also be applied to the field monitoring of other circular plants.