Research on Classification of Water Stress State of Plant Electrical Signals Based on PSO-SVM

Plant electrical signals are physiological signals within the plant body that respond to both external and internal stimuli. Using plant electrical signals as an effective indicator for evaluating the plant growth status is a new theory and method to study the relationship between environmental factors affecting plant growth and plant growth responses. This novel proposed approach is advantageous in terms of response sensitivity and accuracy. Therefore, automation and intelligence of agricultural plant cultivation can be realized and implemented by monitoring the changes in the patterns of plant electrical signals. This paper investigated rapeseed plants in three groups and evaluated the effect of soil water content as the controlled environmental variable. The plant electrical signals under different soil water contents were collected for wavelet packet noise reduction processing. The plant electrical signals were analyzed from three aspects: time domain, frequency domain, and wavelet packet decomposition. The mean, root mean square, standard deviation in the time domain, the power spectral entropy (PSE) of the centroid frequency (SCF) in the frequency domain, and the electrical signal energy in the wavelet packet decomposition were used as the eigenvectors required for classification. The external plant morphological data and the rapeseed growth under different soil water contents were collected to establish plant water stress evaluation indexes. By this, the optimal water demand gradient for rapeseed growth was obtained. The plant water stress evaluation index was used to verify the classification effect of electrical signals, combined with the plants’ electrical signal changes under different water stress conditions to comprehensively evaluate the growth status of plants under different soil water contents. Finally, a support vector machine (SVM) and a particle swarm optimized support vector machine (PSO-SVM) were used to classify the water stress status of plants and establish prediction model the relationship between plant growth status and plant electrical signals under different soil water contents. The results showed that the plant water stress state classification model accuracy based on SVM was 90.83%, and the mean square error MSE was 0.175, while the accuracy of the plant water stress state classification model based on PSO-SVM was 94.3167%, and the mean square error was 0.1646. The classification experiments results show that the water stress of plants and the classification of plant growth status can be realized utilizing electrical signal analysis, with the support vector machine classification model after particle swarm optimization being more accurate. This method lays the foundation for the realization of automation in agricultural plant cultivation and monitoring through plant electrical signals.