Rapid detection of corn moisture content based on improved ICEEMDAN algorithm combined with TCN-BiGRU model

Abstract

Rapid detection of corn moisture content(MC) during maturity is of great significance for field cultivation, mechanical harvesting, storage, and transportation management. However, cumbersome operation, time-consuming and labor-intensive operation were the bottleneck in the traditional drying process and dielectric parameter method. Thus, to overcome the above problems, a rapid detection method for corn MC based on improved complete ensemble empirical mode decomposition with adaptive noise (ICEEMDAN) combined with temporal convolutional network-bidirectional gated recurrent unit (TCN-BiGRU) model. First, based on the 405 groups of NIR spectral data of corn seeds, the crested Porcupine Optimizer (CPO) algorithm was used to optimize ICEEMDAN to reduce the noise of the original spectral data. Then the Chaotic-Cuckoo Search (CCS) algorithm was applied to extract 203 characteristic wavenumbers from the original spectrum, which were input into the constructed TCN-BiGRU network model to realize corn MC detection. Finally, the CPO-ICEEMDAN-CCS-TCN-BiGRU corn MC classification detection model was constructed. The result showed that the model accuracy was 97.54 %, which was 9.22 %, 5.58 %, 2.34 %, 4.74 %, and 5.94 % higher than those of convolutional neural networks (CNN), long short-term memory networks (LSTM), temporal convolutional network (TCN), partial least squares (PLS), and support vector machine (SVM) models, respectively. The research results can provide a reliable basis for improving corn yield, quality and economic benefits.