Huang, YunfengYunfengHuangElewa, KhaledKhaledElewaWang, Tien ChengTien ChengWangChen, Tsu WeiTsu WeiChenFANG-JING WU2023-10-252023-10-252022-01-019781665491532https://scholars.lib.ntu.edu.tw/handle/123456789/636533Smart farming technologies have the potential to facilitate low-cost crop monitoring for increasing crop yields. This work proposes a human-machine augmented learning framework for smart farming, called Half-Farmer, which effectively recognizes the germination stages of wheat seeds. Human experts (i.e., agriculturalists) and a recognition model based on convolutional neural networks (CNN) collaboratively label germination stages of seeds on images according to the unique characteristics of different germination stages. Initially, a CNN-based recognition model augments weak recognition results on images for human experts to cross-validate the germination stages. With the cross-validated labels, the initial recognition model is gradually updated with a greater understanding of seed germination characteristics. The iterative human-machine collaborative process facilitates model evolution in Half-Farmer. Therefore, human experts further jointly make decisions together with the evolved recognition model to resolve the ambiguity in germination stages arising from either human subjective opinions or incorrect recognition by early weaker models. Extensive experimental results indicate that the Half-Farmer achieves an overall Fl-score of 79.5%, an overall recall of 78.4%, and an overall precision of 80.2 %.convolutional neural networks | crop monitoring | human-machine collaboration | seed germination recognition | smart farmingconference paper10.1109/WF-IoT54382.2022.101521062-s2.0-85164166279https://api.elsevier.com/content/abstract/scopus_id/85164166279