An improved slime mould algorithm applied to hyperspectral feature selection in soil fertility indicators

High-precision soil quality monitoring is vital for precision agriculture. Recently, swarm intelligence (SI) algorithms have become essential for identifying hyperspectral features. Aiming at the challenges of high dimensionality in soil fertility indicator retrieval, this paper integrates ZY1-02E satellite and HyMap airborne data to conduct feature selection and fertility estimation using an improved slime mould algorithm (ISMA). ISMA integrates three synergistic stages: (1) spectrally-informed initialization and pre-screening for redundancy reduction, (2) a multi-level location update strategy for balancing exploration and exploitation, and (3) a complementary stagnation control strategy for late stage search refinement. A customized fitness function is designed to optimize the estimation of soil fertility indicators. The effectiveness of the proposed ISMA was first validated on ZY1-02E satellite imagery. Experimental results demonstrated that the ISMA-XGBoost model yielded the best optimization capability, achieving the test R 2 of 0.752 for soil organic matter (SOM) on ZY1-02E data, and its multi-indicator applicability was confirmed with R 2 values of 0.675 (total phosphorus), 0.670 (total potassium), and 0.678 (total nitrogen). The model’s reliability was validated on independent HyMap data (R 2 = 0.757), a significant improvement over conventional algorithm (R 2 = 0.730). These findings provide critical technical support for regional precision agriculture.