Research on maize leaves surface action potential recognition method based on ResNet-18SE

Plant action potentials are rapid changes in cellular electrical potentials, serving as critical indicators of plant physiological activity. This study introduces a novel maize leaf surface action potential recognition method by leveraging the ResNet-18SE model combined with Short-Time Fourier Transform (STFT) to convert one-dimensional electrophysiological signals into two-dimensional time-frequency domain images. The dataset was enhanced using advanced techniques, including random cropping, transformations, and generative adversarial networks (GANs). The ResNet-18SE model, integrated with Squeeze-and-Excitation (SE) blocks, achieved superior classification accuracy (95.60%), precision (96.80%), and recall (98.90%), surpassing VGG16, ResNeXt, SqueezeNet, and AlexNet. These results highlight the robustness and efficiency of combining deep learning with frequency-domain analysis in plant electrophysiological signal processing, providing a scalable framework for real-time plant health monitoring and advancing the study of plant physiology. • A classification method based on frequency domain features of plant electrical signals improves action potential accuracy. • Combining the ResNet-18 model with SE modules achieves strong classification performance on maize leaves. • Classification accuracy reaches 95.6% with an F1 score of 97.9% using maize leaves data. • The method enables real-time monitoring of maize environmental stress for precision agriculture.