Estimation of soil texture by fusion of near-infrared spectroscopy and image data based on convolutional neural network

•A novel CNN model was developed for predicting soil texture based on fusing NIR and image data.•Data sizes, NIR pre-processing methods, and image types were optimized.•HSV images and NIR spectra preprocessed with derivation boost CNN performance.•Multi-dimensional CNN accurately predicts clay and silt fractions.•Fused model achieves high prediction accuracy with smaller dataset.