Detection Of Soil Information Based On Spectral Analysis And Digital Image Processing Technology

Precision agriculture is the combinative production of both modern information technology and traditional agriculture. It can reduce the pollution and waste caused by the agriculture production, reduce costs and raise farm yields. It is the direction for future agricultural development. The collection and monitoring of soil information is an important part of precision agriculture. How to detect the soil information rapidly and accurately is a difficult problem for precision agriculture, also it is the hot research topic today.There are many problems of the Fast Check-test Technology for soil information to be solved. This paper studied the application of spectroscopy techniques and digital image processing technology in this area. The Fast Check-test of soil organic matter content and soil moisture content in soil was studied. The main contents are as following:1. This paper studied the effect of pretreatment method on detection precision of soil organic matter content detected by spectral model. Firstly, process spectral signal using different pretreatment methods. Secondly, built the thorough spectra waves model(350～2500nm) individually by PLS. Through the model's correlation coefficient of prediction (R), it can be found that smoothing algorithm could increase the detection precision of soil organic matter content detected by PLS model. Of them, the effect of Savitzky Golay Smoothing (three points) is the best for PLS model, the R and RMSEP for the prediction of soil moisture content by this model were 0.8694 and 0.2558.2. This paper studied the application of Non-Linear Model LSSVM and BPANN in the spectral detection of the soil organic matter content. PCA are applied to compress the initial spectral data with wavelengths from 350 to 2500nm, which was not pretreated, and then we built LSSVM and BPANN models individually. The results of LSSVM model indicated that the R and RMSEP for the prediction of soil organic matter content were 0.8544 and 0.2657. The results of BPANN model indicated that the R and RMSEP for the prediction of soil organic matter content were 0.8829 and 0.2618. A PLS model using the same data is also built, the R and RMSEP for the prediction of soil organic matter content by PLS model were 0.8373 and 0.2905. By comparison, the detection precision of soil organic matter was BPANN>LSSVM>PLS.3. This paper studied the application of digital image processing technology to measure soil organic matter content. We extracted the average gray value of soil images of green heft,red heft and near infrared heft, and then we built the Non-Linear Model LSSVM and Linear Model PLS individually. The R and RMSEP for the prediction of soil moisture content by PLS model were 0.7692 and 0.3316. The R and RMSEP for the prediction of soil moisture content by LSSVM model were 0.7792 and 0.3163.4. This paper studied the application of digital image processing technology to measure the water contents in "ideal soil". We prepared artificial "ideal soil" with different water contents and extracted the average gray value of soil images of three hefts. Finally, the LSSVM Model and PLS Model were built individually. It can be found that the models' prediction outcome was good. The R and RMSEP for the prediction of soil moisture content by PLS model were 0.9707 and 0.6997. The R and RMSEP for the prediction of soil moisture content by LSSVM model were 0.9767 and 0.4016.