Wheat aging plays an important role in assessing storage wheat quality and its subsequent processing purposes. The conventional detection methods for wheat aging are mainly involved in chemical techniques, which are time-consuming as well as waste part of wheat samples for each detection. Although some physical detection methods have obtained gratifying results, it is extremely hard to expand their application fields but to stay in the theory stage. For this reason, a novel nondestructive detection model for wheat aging based on the delayed luminescence (DL) has been proposed in this paper. Specifically, after collecting enough sample data, we first took advantage of certain hyperbolic function to fit DL signal, and then used four parameters of the hyperbolic function to feature the decay trend of the DL signal. Secondly, in order to better feature the DL signal, we extracted other six features together with above four features to form the input feature vector. Finally, as the bidirectional long short-term memory (Bi-LSTM) network lacked error-correcting performance, the Bi-LSTM network based on Walsh coding (Walsh-Bi-LSTM) mechanism was proposed to establish the detection model, which made the detection model have the error-correcting performance by reasonably splitting the multi-classification target task. Shown by experimental results, the newly proposed wheat aging detection model is able to achieve 94.00% accuracy in the testing dataset, which can be used as a green and nondestructive method to timely reflect wheat aging states.
© 2024. The Author(s).