Vis/NIR Chemical Imaging Technique for Predicting Sodium Humate Contents in Aquaculture Environment

This study aimed to develop a visible and near-infrared (Vis/NIR) chemical imaging (400–1000 nm) technique to provide rapid prediction of the contents of sodium humate dissolved in aquaculture environment. Gray reference image with 5% reflectance value was first used to correct the obtained raw images in order to promote the reflectance values as compared to that with 99% reflectance for further spectral analysis. Successive projection algorithm (SPA) was introduced to extract four optimal wavelengths, which were then used for the establishment of back-propagation artificial neural network (BP-ANN) models. The results revealed that the BP-ANN model based on the selected four optimal wavelengths better performed (\( {R}_C^2 \) = 0.986, \( {R}_{CV}^2 \) = 0.985, \( {R}_P^2 \) = 0.993, RMSEC = 0.329 mg/L, RMSECV = 0.433, RMSEP = 0.734 mg/L) than that based on the whole 381 wavelengths (\( {R}_C^2 \) = 0.978, \( {R}_{CV}^2 \) = 0.996, \( {R}_P^2 \) = 0.977, RMSEC = 0.388 mg/L, RMSECV = 0.625, RMSEP = 0.734 mg/L). Finally, a series of chemical images were developed to clearly display the concentration distribution of the sodium humate dissolved in water, demonstrating that Vis/NIR chemical imaging technique was feasible to quantify the contents of sodium humate in the aquatic environment and could be further used for real-time monitoring the quality of aquaculture water.