Answers are grounded in this paper and cite their sources.
Early and accurate crop yield prediction is crucial for developing adaptive strategies and improving breeding efficiency. However, traditional yield measurement of winter wheat is labor-intensive that is performed after harvesting, leaving limited time for decision making. Most models for yield prediction rely on observed spectral features at the grain filling stage rather than time-series predictions of key phenological stages. In this study, we collected multispectral images of recombinant inbred lines (RILs) from Zhongmai 578/Jimai 22 cross by unmanned aerial vehicles (UAV) at two experimental sites (Dezhou and Gaoyi) during three growth stages (flag leaf, anthesis, and grain-filling). We developed a framework named LSTMRFNet for yield prediction, which integrates Long Short-Term Memory (LSTM) networks with Random Forest (RF) algorithms. Three datasets were constructed based on different feature sources and designated as Integrity, Forecast and Deletion Datasets. LSTM model was shown to effectively predict time-series wheat canopy spectral features and vegetation indices. In addition, using time-series LSTMRFNet framework, the Forecast Dataset achieved yield predictions R 2 value of 0.72 (Dezhou) and 0.67 (Gaoyi), approaching the performance of Integrity Dataset based on measured data ( R 2 =0.75 and 0.67, respectively). In summary, the LSTMRFNet model serves as an effective tool for early yield prediction and line selection in wheat breeding.
No takes yet. Share an insight, caveat, or question.
Wan et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: