Soil Nutrition Content Detection And Instrument Development Based On Visible Near Infrared Spectrum Technology

It is necessary for precision agriculture to understand the spatial-temporal variability and real-time nutritional status of soil. And the digital agriculture also requests that the soil nutrition detection should be timely and effective. Soil organic matter (OM), total nitrogen (TN), available nitrogen (N), available phosphorus (P), available potassium (K) are the main nutrients for crop growth, soil nutrient management and soil testing important objects. Conventional detection methods of these parameters are adopted by many laboratories and soil fertilizer manage station, these methods require expensive testing equipment and complex manual and have many disadvantages such as low efficiency, few samples and high cost problem, which hold back soil fertility management and the development of precision agriculture management. With large-scale promotion of fertilization technology, it is urgent for a low-cost and reliable method to rapidly detect soil nutrients. As a rapid, convenient, nondestructive and green technique, spectroscopy analysis becomes more and more important in the area of soil nutrition detecting. Near infrared spectroscopy technique offers a quick analysis, little sample preparation requirement, and low cost. They are highly sensitive to both organic and inorganic components of the soil, making their use in the agricultural and environmental sciences particularly appropriate. The analytical abilities of visible near infrared spectrum (Vis/NIRS) depend on the repetitive and broad absorption of Vis/NIRS light by C-H, O-H and N-H bonds.Soil organic matter (OM), nitrogen (N), phosphorus (P) and potassium (K) are the main nutrients for crop growth, soil nutrient management and soil testing important objects. With large-scale promotion of fertilization technology, it is urgent for a low-cost and reliable method to rapidly detect soil nutrients.Different calibration methods were used to detect the soil nutrition based on near infrared spectrum technology. Near infrared diffuse reflectance spectroscopy data of soil samples was used for the principal component analysis (PCA) to get the first six principal components (PCs), and PLSR mold was built to get six latent variables (LVs), respectively. PCR, PLSR, BPNN-PCs, BPNN-LVs, LS-SVM-PCs and LS-SVM-LVs modeling methods were built to predict the content of soil organic matter, available nitrogen, available P and available K. These modeling methods were evaluated respectively and selected the best model. The results showed that all LS-SVM-LVs models outperformed PCR, PLSR, BPNN-PCs, BPNN-LVs and LS-SVM-PCs models. The best predictions were obtained with LS-SVM-LVs model for OM (R2=0.8734and RMSEP=2.92), N (R2=0.7801and RMSEP=16.49), P (R2=0.7801and RMSEP=4.97) and K (R2=0.7353and RMSEP=13.42). The near-infrared diffuse reflectance spectroscopy based on LS-SVM combined with PLSR can be used for the measurement of soil organic matter, available N, available P and available K.Near infrared diffuse reflectance spectroscopy was investigated for measurement accuracy of soil properties, namely, available nitrogen (N) and available potassium (K). Three types of pretreatments including standard normal variate (SNV), multiplicative scattering correction (MSC) and Savitzky-Golay smoothing+first derivative were adopted to eliminate the system noises and external disturbances. Then partial least squares (PLS) and least squares-support vector machine (LS-SVM) models analysis were implemented for calibration models, LS-SVM regression preferably solved the practical issues such as non-linearity, multi-dimension and so on. Simultaneously, the performance of least squares-support vector machine (LS-SVM) models was compared with three kinds of inputs, including PCA (PCs), latent variables (LVs), and effective wavelengths (EWs). The results indicated that all LS-SVM models outperformed PLS models. The performance of the model was evaluated by the determination coefficient (R2), RMSEP. The optimal EWs-LS-SVM models were achieved, and the determination coefficient (R2), RMSEP was0.82,17.2for N and0.72,15.0for K, respectively. The results indicated that visible and near infrared spectroscopy (Vis/NIRS)(325-1075nm) combined with LS-SVM could be utilized as a precision method for the determination of soil properties.For the reason that using the raw spectra data has many drawbacks such as big data, too many wavelength, So this research studied some selecting characteristic wavelengths way to choose characteristic wavelength or characteristic variables, these ways includegenetic algorithm (GA), successive projections algorithm (SPA),uninformative variable elimination (UVE) and effective wavelengths (EWs) and so on. In order to improve the predictive precision, and eliminate the influence of uninformative variables for model robustness, Monte carlo uninformative variables elimination (MC-UVE) methods were proposed for variable selection in available nitrogen (N) and available potassium (K) spectral modeling.Partial least squares (PLS) models analysis were implemented for calibration models.The modeling variable number was reduced to210from751for available nitrogen (N) calibration model and150for available potassium (K) calibration model. The performance of the model was evaluated by the determination coefficient (R2), RMSEP. The optimal MC-UVE-PLS models were achieved, and the determination coefficient (R2), RMSEP were0.86,17.1for N and0.78,15.4for K, respectively.The results indicated that visible and near infrared spectroscopy (Vis/NIRS)(325-1075nm) combined with MC-UVE could be utilized as a precision method for the determination of soil properties.The calibration was optimized by genetic algorithm (GA) in the wavelength range of325-1075nm. After optimizations, the sample number of calibration set decreased from751to17, then least squares-support vector machine (LS-SVM) models analysis were implemented for calibration models. Simultaneously, the performance of least squares-support vector machine (LS-SVM) models was compared with PLS models. The results indicated that LS-SVM models outperformed PLS models. The performance of the models was evaluated by the determination coefficient (R2), RMSEP. The optimal GA-LS-SVM models were achieved, and the determination coefficient (R2), RMSEP was0.81,17.8for N and0.71,15.6for K, respectively.The results indicated that visible and near infrared spectroscopy (Vis/NIRS)(325-1075nm) combined with LS-SVM based on GA could be utilized as a precision method for the determination of soil properties.Successive projections algorithm (SPA) based on NIR was investigated in this study for measurement of soil organic matter (OM) and available potassium (K). Four types of pretreatments including smoothing, SNV, MSC and SG smoothing+first derivative were adopted to eliminate the system noises and external disturbances. Then partial least squares regression (PLSR) and least squares-support vector machine (LS-SVM) models analysis were implemented for calibration models. The LS-SVM model was built by using characteristic wavelength based on successive projections algorithm (SPA). Simultaneously, the performance of LS-SVM models was compared with PLSR models. The results indicated that LS-SVM models using characteristic wavelength as inputs based on SPA outperformed PLSR models. The optimal SPA-LS-SVM models were achieved, and the determination coefficient (R2), RMSEP were0.8602,2.98for OM and0.7305,15.78for K, respectively.Spectra in the calibration set were subjected to partial least squares regression (PLSR) to establish calibration models of soil properties. Except for the Wangjia farm, individual farm models provided successful calibration result for total nitrogen (TN) with coefficient of determination (R2) of0.82-0.88and0.72-0.82and residual prediction deviation (RPD) of2.62-3.27and2.02-3.07for the calibration dataset and independent validation respectively. General calibration models gave improved prediction accuracies compared with models of farms in the Wangjia and Changdong, which was attributed to larger ranges in the variation of soil properties in general models compared with those in individual farm models. The results showed that larger standard deviations (SDs) and wider variation ranges have resulted in larger R2and RPD, meanwhile larger root mean square errors of prediction (RMSEP).Therefore, a compromise solution, which also results in small RMSEP values, soil samples should be selected for calibration to cover a wide variation range.The soil organic matter detection instrument was based on NIRS technology including USB4000optical spectrum instrument. The instrument consisted of two parts, software section and hardware section. The software section included soil organic matter detection software based on JAVE language and USB4000driven program. The hardware section included lamp source driven circuit, Y type optical fiber, win CE development board, portable power, and touch liquid crystal display circuit and instrument box. Incident light signals was transmit through optical fiber to the measured soil surface, diffuse reflection data was caused from the soil surface through reflection optical fiber transmission to the USB4000spectrometer to obtain soil reflectance value, the software system for processing, display, storage and other processing.