A pilot study of maize SPAD stratification from leaf surface potentials under controlled conditions

Chlorophyll content serves as a critical indicator for assessing photosynthetic efficiency and health status in maize plants. Previous studies have suggested a correlation between variations in plant electrical signals and chlorophyll content. To validate the feasibility of grading maize leaf chlorophyll content based on surface potential and to explore the potential of monitoring maize growth status via these signals, this study presents a methodological validation of a detection framework under controlled laboratory conditions. Based on a strictly controlled laboratory experiment designed to maximize signal fidelity, a detection method utilizing a Time-Frequency Feature Fusion-Based Back Propagation (BP) neural network was developed. Given the sample size constraints, the BP architecture was specifically selected to balance computational efficiency with the need to mitigate overfitting risks. The method integrates time-domain and frequency-domain features derived from leaf surface potential signals into a dual-branch BP neural network optimized with a focal loss function. This framework established a classification model for grading maize chlorophyll content, achieving a recognition accuracy of 77.60%. Comparative analyses demonstrated that this model outperformed Support Vector Machine (SVM), conventional BP, dual-branch BP, and class-weighted cross-entropy models, with accuracy improvements of 37.74, 8.02, 3.78, and 3.54 percentage points, respectively. This research provides a controlled-condition methodological basis for the nondestructive assessment of chlorophyll-related status in individual maize plants. It complements traditional optical detection methods, offers preliminary evidence supporting the use of electrophysiological signals for plant physiological state stratification, and provides a basis for subsequent studies in related biosensing applications.