Soil-Available Potassium Content Measurement Based on Nernst Informed Neural Network With Electrical Signal

Soil-available potassium content is essential for agricultural crop management. The electrochemical measurement method exhibits advantages of high speed, low cost, and simple operation. However, it encounters challenges, such as poor signal reproducibility, limited sample accessibility, and restrictions on measurement accuracy. To address these issues, the Nernst informed neural network (NINN)-based electrochemical measurement of soil-available potassium content is proposed. The framework for the electrochemical soil-available potassium content measurement is established. The proposed method is developed to exploit the nonlinear relationship between the electrical signal and the soil-available potassium content end-to-end with a single potassium ions-selective electrode. The NINN in the proposed method is developed to improve the measurement accuracy on the dataset with a limited number of samples by integrating Nernst mechanism into the network. The potassium-content dynamic estimation in the proposed method is developed to overcome the uncertainties of the dilution factor and influence of interfering ions. The practical measurement exhibits the mean absolute percentage error in a low range and the time within 10 min. Experimental results demonstrate that the proposed method improves the accuracy compared to purely data-driven models and reduces the electrochemical measurement costs. Furthermore, the proposed method also exhibits the superiority in terms of measurement speed; operational complexity compared to the commercial products relied on spectroscopy. The proposed method can contribute to the advancement of the measurement of soil nutrients.