The soil nitrate-nitrogen (NO-3-N) is essential element for crop growth. Because of the obvious advantages on cost, applicability and easy-implementation, the nitrate ion-selective electrode (ISE) was demonstrated potentials in both laboratory and in-field researches on soil available nitrogen detections. However, problems of unidealistic selectivity and potential drift usually limited the application of ISE. The extreme learning machine algorithm was used to decouple the signals of nitrate ion-selective electrode from the interference of chloride. Three data processing algorithms, including drift correction, Nernstian model and extreme learning machine were systemically analyzed. Experiments were carried out on the self-designed multi-channel nutrient detection platform. Totally 150 soil samples were selected for the system validation. The experimental results indicated that the accuracy and consistency of sensor’s scaling equations were effectively improved by drift correction algorithm. The variations of response slope and intercept potential were reduced by 3.67% and 7.25%, respectively. The neuron number in hidden layer of the extreme learning machine was 14，which were tested as optimized parameter. The extreme learning machine could effectively decouple the interference of chloride from nitrate ion-selective electrode in saline alkali soil. The maximum absolute error and root mean square error were 6.36mg/L and 4.02mg/L, respectively. In conclusion, the research results can provide references in the related studies for soil detection by ion-selective electrode.