A novel quantitative detection method for soil organic matter content based on visible to near-infrared spectroscopy

Continued mining operations have resulted in substantial soil degradation, necessitating the effective restoration of ecological functions in soils. Accurate and rapid measurement of soil organic matter (SOM) is essential for boosting soil fertility, supporting ecological restoration, and facilitating effective environmental management. Combining visible to near-infrared spectroscopy with machine learning algorithms is a promising technique for quantitative analysis of SOM. Firstly, the paper utilized a spectral pre-processing method that integrates fractional order differentiation transformation (FOD) and optimal band combination (OBC) algorithm. OBC algorithm was used to construct six three-band spectral indices to obtain optimal spectral combination parameters. Then, the HOVD-TELM model was constructed based on the hybrid model of two-hidden-layer extreme learning machine and Harris hawk optimizer. The opposition-based learning, vertical crossover operator and disruption operator were added to prevent the model from converging prematurely. The experimental results showed that: (1) compared with the pre-processing methods such as integer order differentiation and two-band spectral index, the FOD and OBC methods used in this paper obtained more ideal spectral pre-processing effects. (2) compared with models such as Partial least square regression and Extreme gradient boosting, the HOVD-TELM model proposed in this paper obtained the optimal prediction performance, with the minimum RMSE of 6.7874 g·kg⁻¹ and the maximum R² of 0.9209, indicating good prediction reliability. In summary, this paper proposed a fast and accurate method for detecting soil organic matter content and improves the estimation accuracy of SOM content.