A hybrid CNN-LSTM model for diagnosing rice nutrient levels at the rice panicle initiation stage

• DOI: 10.1016/j.jia.2023.05.032
• Publication Date: 2023-05-20T01:14:42Z
• Authors: Fubing Liao; Xiangqian Feng; Ziqiu Li; Danying Wang; Chunmei Xu; Guang Chu; Hengyu Ma; Qing Yao; Song Chen
• Keywords: Panicle; Rice plant

Nitrogen (N) and potassium (K) are two key mineral nutrient elements involved in rice growth. Accurate diagnosis of N K status is very important for the rational application fertilizers at a specific growth stage. Therefore, we propose hybrid model diagnosing levels early panicle initiation stage (EPIS), which combines convolutional neural network (CNN) with an attention mechanism long short-term memory (LSTM). The was validated on large set sequential images collected by unmanned aerial vehicle (UAV) from canopies different stages during two-year experiment. Compared VGG16, AlexNet, GoogleNet, DenseNet, inceptionV3, ResNet101 combined LSTM obtained highest average accuracy 83.81% dataset Huanghuazhan (HHZ, indica cultivar). When tested datasets HHZ Xiushui 134 (XS134, japonica variety) 2021, ResNet101-LSTM enhanced Squeeze-and-Excitation (SE) block achieved accuracies 85.38% 88.38%, respectively. Through cross-dataset method, XS134 2022 were 81.25% 82.50%, respectively, showing good generalization. Our proposed works dynamic information can efficiently diagnose EPIS, helpful making practical decisions regarding fertilization treatments