Method for detecting soil total nitrogen contents based on pyrolysis and artificial olfaction

Soil nitrogen is an essential nutrient element for crop growth and development, and an important indicator of soil fertility characteristics. This study proposed a method based on pyrolysis and artificial olfaction to quickly and accurately determine the soil total nitrogen (STN) content. A muffle furnace was used to pyrolyze the soil samples, and ten different types of oxide semiconductor gas sensors were used to construct a sensor array to detect the soil samples’ pyrolysis gas. The response curves of the sensors were tested at pyrolysis temperatures of 200°C, 300°C, 400°C, and 500°C and at pyrolysis times of 1 min, 3 min, 5 min, and 10 min to obtain the optimal pyrolysis state of the soil samples. The optimal pyrolysis temperature was 400°C, and the pyrolysis time was 3 min. The response area, maximum value, average differential coefficient, variance value, maximum gradient value, average value, and 8th-second transient value of the sensor response curve were extracted to construct an artificial olfactory feature space of 121×10×7 (121 soil samples, ten sensor numbers, seven extracted eigenvalues). Back-propagation neural network algorithm (BPNN), partial least squares regression algorithm (PLSR), and partial least squares regression combined with back-propagation neural network algorithm (PLSR-BPNN) were used to establish a prediction model of artificial olfactory feature space and STN content. Moreover, coefficient of determination (R2), root mean square error (RMSE), and the ratio of performance to deviation (RPD) were used as the performance indicators of the prediction results. The test results showed that the R2 of the PLSR, BPNN, and PLSR-BPNN models were 0.89033, 0.81185, and 0.92186, and the RMSE values were 0.24297, 0.37370, and 0.21781, and the RPD were 2.9964, 1.9482, and 3.3426, respectively. The model established by the PLSR-BPNN algorithm has the highest R2 and RPD and the smallest RMSE, can achieve the accurate prediction of STN content, and therefore the model is rated as “excellent”. The detection method in this study achieves a low-cost, rapid, and accurate determination of STN content, and provides a new reference for the measurement of STN.