Automatic recognition of rice lodging using fine-tuning technology. (a) The training data for lodging detection model, which included 100 and 83 images for lodging and nonlodging, respectively. (b) Pseudocolor images of hyperspectral data (bands 77, 50, and 18) from 39 experimental fields. The ROIs were intercepted according to planting area to establish testing data set, which included 14 and 25 lodging and nonlodging, respectively. (c) Confusion matrix of the model on the training data. (d) Confusion matrix of the model on the testing data.