Detection of Soil Nutrients Using an Olfactory System Based on Rf-Pcc Sensor Array Optimization Algorithms

Sensor arrays play a key role in the detection of soil gas. Nevertheless, suboptimal sensor combinations may culminate in a myriad of complications, such as overlapping selectivities, superfluous features, and deterioration of performance. To mitigate those challenges, we innovatively developed a sensor array optimization (SAO) method based on the dynamic feature importance (DFI) of the Random Forest-Pearson Correlation Coefficient (RF-PCC). The optimal sensor combinations for the four soil nutrients were identified using an “overall feature” strategy after initial data sets were derived from utilizing pyrolysis gas response curves of soil samples acquired from ten distinct gas sensors. Afterward, the performance of the chosen sensors was rigorously assessed using machine learning, the least absolute shrinkage and selection operator (LASSO), which the coefficient of determination (R2) improvement range from 2.16% to 2.46% over that of the original array, the relationship between soil pyrolysis gas and nutrients content was found while demonstrating that the selected sensors have a sufficient level of detective capability. Experimental results suggested that RF-PCC can effectively reduce the correlation of features, that the selected sensors also displayed comparable performance in the regression task even with fewer sensors. The findings presented here advance our knowledge of a new approach to sensor array optimization for soil gas detection