Hyperspectral analysis of the content of the alkali-hydrolysed nitrogen in the soil of a millet field

Abstract

Hyperspectral imaging technology has emerged as a prominent research area for quantitatively estimating soil nutrient content owing to its non-destructive, rapid, and convenient features. Our work collected the data from soil samples using the hyperspectrometer. Then, the data were processed. The competitive adaptive reweighted sampling (CARS) algorithm reduced the original 148 bands to 13, which accounted for 8.8% of the total bands. These selected bands possess a certain level of interpretability. Based on the modelling results, it can be concluded that the prediction model constructed by the least squares support vector machine (LSSVM) exhibited the highest accuracy. The coefficient determination, root mean square error, and ratio performance deviation were 0.8295, 2.95, and 2.42, respectively. These findings can provide theoretical support for the application of hyperspectral technology in detecting the content of the AHN in soil. Moreover, they can also serve as a reference for the rapid detection of other soil components.

Keywords

• modern precision agriculture
• soil fertility
• characteristic bands selection
• model building
• content prediction