Chen, Po ShaoPo ShaoChenChang, Shan ChengShan ChengChangChang, Han BinHan BinChangHuang, Wei HaoWei HaoHuangChueh, Cheng YuCheng YuChuehLee, Cheng ChunCheng ChunLeeChien, Chia ChunChia ChunChienJOE-AIR JIANGWang, Jen ChengJen ChengWangLiu, An ChiAn ChiLiuHsieh, Ming-HsienMing-HsienHsiehPeng, Jui ChuJui ChuPengGuo, Ming ChiMing ChiGuoCHENG-YING CHOU2024-02-192024-02-192023-01-019781713885887https://www.scopus.com/inward/record.uri?eid=2-s2.0-85183582645&doi=10.13031%2faim.202300600&partnerID=40&md5=e6473e9f25be958564260979d3ec0596https://scholars.lib.ntu.edu.tw/handle/123456789/639773Taiwan, with its small land area, relies heavily on precision agriculture to maximize yield per unit area. However, this has led to a high demand for manual labor, particularly in harvesting asparagus. In recent years, there has been a decrease in the number of available farmers, making the development of automatic harvesters increasingly crucial. To address this challenge, this study employs deep learning techniques to assist in the identification and harvesting of asparagus using a harvester. One of the most challenging aspects for asparagus harvesting in Taiwan is distinguishing between the tender, edible spears and the inedible asparagus stems. To overcome this, the study uses the object detection model YOLOv5 (You Only Look Once), which is known for its fast and lightweight nature, to capture the two-dimensional coordinates of the asparagus spears on the edge computing device. By combining this information with data from an infrared depth camera, it is able to obtain the depth information of the asparagus spears, allowing for a more accurate identification. The YOLOv5 algorithm boasts an impressive accuracy of 91% in identifying asparagus spears and an inference time of 0.3 seconds. Furthermore, labeling a single object class for asparagus spears improves recognition by 40% in comparison to multiple labels for spears and mother stems. By integrating these algorithms, the automatic asparagus harvester is able to quickly and accurately determine the coordinates of harvestable asparagus spears, providing farmers with a more efficient and effective harvesting method. As a result, labor demands could be reduced and yields could be improved.asparagus | automatic harvester | deep learning | edge computing | object detection[SDGs]SDG2conference paper10.13031/aim.2023006002-s2.0-85183582645https://api.elsevier.com/content/abstract/scopus_id/85183582645