Chen, Ssu-ChiSsu-ChiChenJian, Zi-HengZi-HengJianLin, Ya-PingYa-PingLinChen, Shih-FangShih-FangChen2026-03-312026-03-312025https://www.scopus.com/record/display.uri?eid=2-s2.0-105015555850&origin=resultslisthttps://scholars.lib.ntu.edu.tw/handle/123456789/7369062025 American Society of Agricultural and Biological Engineers Annual International Meeting, ASABE 2025, Toronto, Ontario, Canada July 13-16, 2025Phenotypic evaluation is critical for understanding plant growth characteristics and genetic potential but faces significant human resource limitations in large-scale applications. This study developed a Temporal Observation and Monitoring system for Automated Tomato phenOmics (TOMATO) that integrates with a self-guided field robot to automate phenotypic assessment. The system features a Plant Positioning System that extracts individual plant images from field videos with 93.3% success rate, an object detection framework that classifies flowers and fruits at three developmental stages (Green, Turning, and Harvest), and an integrated database that enables phenotypic evaluation. Comparative analysis of two detection architectures revealed that DETR with Improved deNoising anchOr boxes (DINO) outperformed YOLOv11, achieving 85.8% precision and 86.6% recall in challenging outdoor environments. The system successfully determined three key phenotypic parameters: Flowering date, Fruiting date, and Fruit color break date, with complete correspondence to established ground truth data. This automated approach significantly enhances efficiency and objectivity in tomato breeding programs by reducing the phenotyping bottleneck between genomic capabilities and phenotypic assessment.falseImproved deNoising anchOr boxes (DINO)object detectionphenotypic evaluationPlant Positioning Systemtomato breeding programsconference paper10.13031/aim.2025011472-s2.0-105015555850