Optimizing photoexcitation parameters to enhance salt stress features conveyed by light-induced bioelectrogenesis

Previous studies have demonstrated that light-induced bioelectrogenesis (LIB) can be a valuable tool for predicting crop salt tolerance. However, individual differences among plants lead to significant waveform variations under salt stress, which reduces the accuracy of LIB-based assessments of salt tolerance. To address this issue, this study optimized photoexcitation parameters to enhance LIB’s salt stress features. LIB were collected from wheat seedlings exposed to different salt stress levels, light qualities, and illumination/darkness cycles. The classification accuracy of deep learning models was compared across different excitation parameters to determine the optimal parameter combination. Additionally, SHapley Additive exPlanations (SHAP) analysis was employed to quantify the contribution of individual LIB features to classification performance. The results revealed that under a fixed photosynthetic photon flux density (PPFD) of 200 μmol·m⁻²·s⁻¹, the combination of red light and a 10 min/10 min illumination/darkness cycle provided the best enhancement of LIB salt stress features, with classification accuracy reaching 92.00 % in the one-dimensional convolutional neural network (1D-CNN) representing a 10 % improvement compared to white light. SHAP analysis further revealed that the features with the most significant impact on classification performance were concentrated in four specific time intervals. Moreover, shorter darkness periods, and blue light excitation significantly reduced the mean potential difference (MPD) and peak value of LIB, which in turn decreased the model’s classification accuracy. This study not only provided new insights into the mechanisms underlying LIB generation under salt stress, but also advanced its application in breeding stress-resistant crops, thereby accelerating the development cycle of salt-tolerant varieties.