{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,9,9]],"date-time":"2026-09-09T06:07:42Z","timestamp":1788934062415,"version":"build-2803163510"},"reference-count":45,"publisher":"Frontiers Media SA","license":[{"start":{"date-parts":[[2026,9,9]],"date-time":"2026-09-09T00:00:00Z","timestamp":1788912000000},"content-version":"vor","delay-in-days":0,"URL":"https:\/\/creativecommons.org\/licenses\/by\/4.0\/"}],"content-domain":{"domain":["frontiersin.org"],"crossmark-restriction":true},"short-container-title":["Front. Plant Sci."],"abstract":"<jats:sec>\n                    <jats:title>Introduction<\/jats:title>\n                    <jats:p>In artificial intelligence (AI)-operated mobile potato harvesting, separating field impurities (stones and soil clods) from potatoes determines product quality. Because both types trigger the same pneumatic removal action, merging them into one impurity class is a natural simplification; we found that it degrades detection quality in every random seed tested, relative to a 3-label formulation (potato vs. stone vs. soil clod).<\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Methods<\/jats:title>\n                    <jats:p>We trained YOLOX-Small under two label structures, with identical images, boxes, augmentation and hyper-parameters: 3-label (potato, stone, soil clod) and 2-label (potato, impurity). The models were fitted on 11,869 images of ten cultivars from seven Hokkaido farms collected up to the 2024 harvest \u2014 the training and validation splits of a 13,189-image corpus divided 7:2:1. The test set was a separate 2,960-image corpus from the 2025 season sharing no image or acquisition session with it (110,420 potato and 15,234 impurity instances; 12.1% prevalence). Each structure used five seeds. Both were scored on a common two-class target, potato versus impurity: the stone and soil-clod predictions of the 3-label model were pooled into a single impurity class before any metric was computed. Detection was assessed with potato misclassification rate (PMR), impurity detection rate (IDR), impurity precision and average precision (AP) over a 17\u00d717 threshold grid; the 512-dimensional backbone representations were compared on the same target.<\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Results<\/jats:title>\n                    <jats:p>The 3-label structure achieved higher impurity precision than the 2-label structure at the default operating point (83.20\u00a0\u00b1\u00a00.85% vs. 70.07\u00a0\u00b1\u00a01.59%; +13.1 percentage points (pp); 95% confidence interval (CI) [10.5, 15.7]) and higher threshold-free impurity AP at 50% box overlap (79.8\u00a0\u00b1\u00a01.3% vs. 65.6\u00a0\u00b1\u00a00.8%; +14.2 pp), in every seed across the operational threshold region. The 2-label representation showed a 39% smaller centroid distance, 37% lower silhouette separability and 30\u201341% lower per-channel Fisher ratios. The advantage persisted although the subclass task failed asymmetrically: soil clods collapsed into stone (recall 1.84\u00a0\u00b1\u00a00.13%) while stones were correctly subclassed (80.69\u00a0\u00b1\u00a01.12%).<\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Discussion<\/jats:title>\n                    <jats:p>Fine-grained supervision appears to shape the backbone representation independently of subclass accuracy: operational equivalence does not entail representational equivalence. The link between representation and performance is observational, and all results come from one detector architecture and region.<\/jats:p>\n                  <\/jats:sec>","DOI":"10.3389\/fpls.2026.1847038","type":"journal-article","created":{"date-parts":[[2026,9,9]],"date-time":"2026-09-09T05:59:51Z","timestamp":1788933591000},"update-policy":"https:\/\/doi.org\/10.3389\/crossmark-policy","source":"Crossref","is-referenced-by-count":0,"title":["Class merging degrades learned representations in potato-harvesting object detection despite operational equivalence"],"prefix":"10.3389","volume":"17","author":[{"given":"Joonam","family":"Kim","sequence":"first","affiliation":[{"name":"Research Center for Agricultural Robotics, NARO","place":["Tsukuba, Japan"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Rena","family":"Yoshitoshi","sequence":"additional","affiliation":[{"name":"Research Center for Agricultural Robotics, NARO","place":["Tsukuba, Japan"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Giryeon","family":"Kim","sequence":"additional","affiliation":[{"name":"Research Center for Agricultural Robotics, NARO","place":["Tsukuba, Japan"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Noriko","family":"Deguchi","sequence":"additional","affiliation":[{"name":"Hokkaido Agricultural Research Center, NARO","place":["Memuro, Japan"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Hideo","family":"Kainuma","sequence":"additional","affiliation":[{"name":"Hokkaido Agricultural Research Center, NARO","place":["Memuro, Japan"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Shinori","family":"Tsuchiya","sequence":"additional","affiliation":[{"name":"Hokkaido Agricultural Research Center, NARO","place":["Memuro, Japan"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Kenichi","family":"Tokuda","sequence":"additional","affiliation":[{"name":"Research Center for Agricultural Robotics, NARO","place":["Tsukuba, Japan"]}],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"1965","published-online":{"date-parts":[[2026,9,9]]},"reference":[{"key":"B1","article-title":"YOLOv1 to YOLOv10: a comprehensive review of YOLO variants and their application in the agricultural domain","author":"Alif","year":"2024","journal-title":"arXiv:2406.10139."},{"key":"B2","doi-asserted-by":"publisher","DOI":"10.1016\/j.compag.2024.109090","article-title":"Agricultural object detection with You Only Look Once (YOLO) algorithm: a bibliometric and systematic literature review","volume":"223","author":"Badgujar","year":"2024","journal-title":"Comput. Electron. Agric."},{"key":"B3","first-page":"3319","article-title":"Network dissection: quantifying interpretability of deep visual representations","author":"Bau","year":"2017"},{"key":"B4","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.2004.10934","article-title":"YOLOv4: optimal speed and accuracy of object detection","author":"Bochkovskiy","year":"2020","journal-title":"arXiv:2004.10934"},{"key":"B5","doi-asserted-by":"publisher","first-page":"249","DOI":"10.1016\/j.neunet.2018.07.011","article-title":"A systematic study of the class imbalance problem in convolutional neural networks","volume":"106","author":"Buda","year":"2018","journal-title":"Neural Netw."},{"key":"B6","first-page":"8730","article-title":"Fine-grained angular contrastive learning with coarse labels","author":"Bukchin","year":"2021"},{"key":"B7","first-page":"213","article-title":"End-to-end object detection with Transformers","author":"Carion","year":"2020"},{"key":"B8","doi-asserted-by":"publisher","first-page":"41","DOI":"10.1023\/A:1007379606734","article-title":"Multitask learning","volume":"28","author":"Caruana","year":"1997","journal-title":"Mach. Learn."},{"key":"B9","first-page":"895","article-title":"Understanding the impact of label granularity on CNN-based image classification","author":"Chen","year":"2018"},{"key":"B10","first-page":"9268","article-title":"Class-balanced loss based on effective number of samples","author":"Cui","year":"2019"},{"key":"B11","first-page":"48","article-title":"Large-scale object classification using label relation graphs","author":"Deng","year":"2014"},{"key":"B12","doi-asserted-by":"publisher","DOI":"10.1073\/pnas.2103091118","article-title":"Exploring deep neural networks via layer-peeled model: minority collapse in imbalanced training","volume":"118","author":"Fang","year":"2021","journal-title":"Proc. Natl. Acad. Sci. U.S.A."},{"key":"B13","doi-asserted-by":"publisher","first-page":"179","DOI":"10.1111\/j.1469-1809.1936.tb02137.x","article-title":"The use of multiple measurements in taxonomic problems","volume":"7","author":"Fisher","year":"1936","journal-title":"Ann. Eugen."},{"key":"B14","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.2107.08430","article-title":"YOLOX: exceeding YOLO series in 2021","author":"Ge","year":"2021","journal-title":"arXiv:2107.08430"},{"key":"B15","doi-asserted-by":"publisher","first-page":"64","DOI":"10.1016\/j.aiia.2024.06.004","article-title":"Computer vision in smart agriculture and precision farming: techniques and applications","volume":"13","author":"Ghazal","year":"2024","journal-title":"Artif. Intell. Agric."},{"key":"B16","doi-asserted-by":"publisher","DOI":"10.1016\/j.compag.2021.106067","article-title":"A survey of deep learning techniques for weed detection from images","volume":"184","author":"Hasan","year":"2021","journal-title":"Comput. Electron. Agric."},{"key":"B17","unstructured":"Ultralytics YOLO\n          \n          \n            \n              Jocher\n              G.\n            \n            \n              Chaurasia\n              A.\n            \n            \n              Qiu\n              J.\n            \n          \n          \n          2023"},{"key":"B18","doi-asserted-by":"publisher","DOI":"10.3389\/fpls.2023.1156734","article-title":"Machinery for potato harvesting: a state-of-the-art review","volume":"14","author":"Johnson","year":"2023","journal-title":"Front. Plant Sci."},{"key":"B19","first-page":"18661","article-title":"Supervised contrastive learning","author":"Khosla","year":"2020"},{"key":"B20","doi-asserted-by":"publisher","DOI":"10.3390\/s25154586","article-title":"Real-time object detection for edge computing-based agricultural automation: a case study comparing the YOLOX and YOLOv12 architectures and their performance in potato harvesting systems","volume":"25","author":"Kim","year":"2025","journal-title":"Sensors"},{"key":"B21","doi-asserted-by":"publisher","DOI":"10.3390\/agronomy16030383","article-title":"Mobile AI-powered impurity removal system for decentralized potato harvesting","volume":"16","author":"Kim","year":"2026","journal-title":"Agronomy"},{"key":"B22","doi-asserted-by":"publisher","DOI":"10.3390\/agronomy11101980","article-title":"Development of an optimal algorithm for detecting damaged and diseased potato tubers moving along a conveyor belt using computer vision systems","volume":"11","author":"Korchagin","year":"2021","journal-title":"Agronomy"},{"key":"B23","doi-asserted-by":"publisher","DOI":"10.3389\/neuro.06.004.2008","article-title":"Representational similarity analysis \u2014 connecting the branches of systems neuroscience","volume":"2","author":"Kriegeskorte","year":"2008","journal-title":"Front. Syst. Neurosci."},{"key":"B24","first-page":"2117","article-title":"Feature pyramid networks for object detection","author":"Lin","year":"2017"},{"key":"B25","doi-asserted-by":"publisher","DOI":"10.3390\/agriculture15070675","article-title":"Recent research progress on key technologies and equipment for mechanized potato harvesting","volume":"15","author":"Liu","year":"2025","journal-title":"Agriculture"},{"key":"B26","doi-asserted-by":"publisher","DOI":"10.1002\/fsn3.71975","article-title":"Counterfeit Citri Reticulatae Pericarpium identification: multi-class detection using Vis\/NIR spectroscopy and residual neural networks","volume":"14","author":"Ma","year":"2026","journal-title":"Food Sci. Nutr."},{"key":"B27","doi-asserted-by":"publisher","DOI":"10.3390\/s23218999","article-title":"Enhancing UAV visual landing recognition with YOLO's object detection by onboard edge computing","volume":"23","author":"Ma","year":"2023","journal-title":"Sensors"},{"key":"B28","unstructured":"YOLOX (version 0.3.0) [Computer software\n          \n          \n          \n          2022"},{"key":"B29","article-title":"What is being transferred in transfer learning","author":"Neyshabur","year":"2020"},{"key":"B30","doi-asserted-by":"publisher","first-page":"24652","DOI":"10.1073\/pnas.2015509117","article-title":"Prevalence of neural collapse during the terminal phase of deep learning training","volume":"117","author":"Papyan","year":"2020","journal-title":"Proc. Natl. Acad. Sci. U.S.A."},{"key":"B31","article-title":"Faster R-CNN: towards real-time object detection with region proposal networks","author":"Ren","year":"2015"},{"key":"B32","doi-asserted-by":"publisher","DOI":"10.3390\/electronics14030638","article-title":"A performance analysis of You Only Look Once models for deployment on constrained computational edge devices in drone applications","volume":"14","author":"Rey","year":"2025","journal-title":"Electronics"},{"key":"B33","doi-asserted-by":"publisher","first-page":"53","DOI":"10.1016\/0377-0427(87)90125-7","article-title":"Silhouettes: a graphical aid to the interpretation and validation of cluster analysis","volume":"20","author":"Rousseeuw","year":"1987","journal-title":"J. Comput. Appl. Math."},{"key":"B34","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.1706.05098","article-title":"An overview of multi-task learning in deep neural networks","author":"Ruder","year":"2017","journal-title":"arXiv:1706.05098"},{"key":"B35","first-page":"618","article-title":"Grad-CAM: visual explanations from deep networks via gradient-based localization","author":"Selvaraju","year":"2017"},{"key":"B36","doi-asserted-by":"publisher","first-page":"60","DOI":"10.1186\/s40537-019-0197-0","article-title":"A survey on image data augmentation for deep learning","volume":"6","author":"Shorten","year":"2019","journal-title":"J. Big Data"},{"key":"B37","doi-asserted-by":"publisher","first-page":"31","DOI":"10.1007\/s10618-010-0175-9","article-title":"A survey of hierarchical classification across different application domains","volume":"22","author":"Silla","year":"2011","journal-title":"Data Min. Knowl. Discov."},{"key":"B38","first-page":"2579","article-title":"Visualizing data using t-SNE","volume":"9","author":"van der Maaten","year":"2008","journal-title":"J. Mach. Learn. Res."},{"key":"B39","doi-asserted-by":"publisher","DOI":"10.3390\/agriculture14081225","article-title":"The application of deep learning in the whole potato production chain: a comprehensive review","volume":"14","author":"Wang","year":"2024","journal-title":"Agriculture"},{"key":"B40","first-page":"2740","article-title":"HD-CNN: hierarchical deep convolutional neural networks for large scale visual recognition","author":"Yan","year":"2015"},{"key":"B41","first-page":"39453","article-title":"Are neurons actually collapsed? On the fine-grained structure in neural representations","author":"Yang","year":"2023"},{"key":"B42","article-title":"A Fourier perspective on model robustness in computer vision","author":"Yin","year":"2019"},{"key":"B43","doi-asserted-by":"publisher","DOI":"10.1016\/j.jfca.2026.109333","article-title":"Geographical origin discrimination of Fructus Aurantii by visible near-infrared imaging with machine learning and explainable analysis","volume":"156","author":"Zhang","year":"2026","journal-title":"J. Food Compos. Anal."},{"key":"B44","article-title":"Object detectors emerge in deep scene CNNs","author":"Zhou","year":"2015"},{"key":"B45","first-page":"29820","article-title":"A geometric analysis of neural collapse with unconstrained features","author":"Zhu","year":"2021"}],"container-title":["Frontiers in Plant Science"],"original-title":[],"link":[{"URL":"https:\/\/www.frontiersin.org\/articles\/10.3389\/fpls.2026.1847038\/full","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,9,9]],"date-time":"2026-09-09T05:59:54Z","timestamp":1788933594000},"score":1,"resource":{"primary":{"URL":"https:\/\/www.frontiersin.org\/articles\/10.3389\/fpls.2026.1847038\/full"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,9,9]]},"references-count":45,"alternative-id":["10.3389\/fpls.2026.1847038"],"URL":"https:\/\/doi.org\/10.3389\/fpls.2026.1847038","relation":{},"ISSN":["1664-462X"],"issn-type":[{"value":"1664-462X","type":"electronic"}],"subject":[],"published":{"date-parts":[[2026,9,9]]},"article-number":"1847038"}}