Methods Of Soil Organic Matter Content Determination Using Artificial Olfactory Technology

Soil organic matter is an important part of soil,which is an important indicator of soil fertility and nutrients,and an important information for precision agriculture.It can not only provide essential nutrients for crop growth,but also can improve the physical structure properties of soil.The decrease of soil organic matter content means that the soil quality becomes worse.Therefore,it is of great significance to measure the content of soil organic matter and grasp its dynamic change to improve soil structure and guide agricultural production.The traditional methods of soil organic matter detection are mostly chemical measurement,which are complicated,time-consuming,costly and destructive.Spectral analysis has attracted much attention in the study of soil properties for its fast,non-destructive and accurate characteristics.However,the spectral analysis of soil organic matter is easily affected by soil particle size,iron oxide and other factors,and the high-resolution spectral detector is expensive.Accurate,rapid and economical?measurement of soil organic matter content has become one of the current research focuses.In order to solve the above problems,in this paper,the artificial olfactory technology is applied to detect the content of soil organic matter,collect the response information of soil volatile organic compounds(VOCs),and explore the feasible technical methods,so as to realize the accurate,rapid and economic measurement of soil organic matter.The main research work of this paper is as follows:(1)Construction of artificial olfactory detection device for soil organic matterAccording to the bionic olfactory mechanism,this paper designed a soil organic artificial olfactory detection device.In this design,it used ten MOS gas sensors sensitive to VOCs to construct the detection array,designed the artificial olfactory signal processing circuit,and developed the upper computer detection software by LabVIEW,which realized the functions of serial port communication,real-time acquisition,display,storage and other functions of the olfactory data.The results of two different high / low concentrations test experiment of ammonia,methane and vinyl chloride standard gases showed that the device had good response performance and different response results for different kinds and concentrations of gases,which meet the basic detection requirements of pattern recognition.The results of comparing and analyzing the rinsing effect of air and helium showed that the response time of the device to air was 7.3 seconds,and the response time to helium was 4.8 seconds.The rinsing effect of helium was better than that of air.(2)Analysis of typical olfactory response curve of soilBased on detecting the soil gas with the artificial olfactory device,this paper obtained the typical olfactory response curve of soil.The results of it showed that the olfactory response curve of soil generally showed a trend of rapidly growth first and then slowly;the response of sensors controlled by different temperatures to soil gas was different,and the time to reach steady state was also different,but the time to reach steady state was generally too long,and the response of some sensors fails to reach steady state even exceeding 85 minutes.In order to improve the detection efficiency and shorten the detection time,the artificial olfactory sampling time of soil gas samples was set as 5 minutes.Three typical soil gas samples with high,medium and low content of soil organic matter were measured by artificial olfactory device.The results of it showed that the artificial olfactory device had good response characteristics to soil gas,and the olfactory response data within 5 minutes could realize the difference detection of different soil gas samples.(3)Construction and optimization of soil olfactory characteristic spaceThe paper analyzed the smoothing effect of four different filtering methods on olfactory response curve,including mean filtering,median filtering,Butterworth low-pass filtering and Savitzky-Golay convolution filtering,and concluded that the median filtering method had the best effect.Then it extracted the maximum value(Vmax),mean differential coefficient value(MDCV),response area value(RAV)and median time transient value(Vt)of the filtered response curve as the feature parameters to construct the preliminary soil olfactory feature space(PSOFS).By comparing the recognition effects of four different methods,such as Monte Carlo Cross Validation(MCCV),Leave One Out Cross Validation(LOOCV),k-means LOOCV and Mahalanobis distance,on abnormal samples in PSOFS,it was concluded that MCCV method had the best effect.By comparing the application effects of principal component analysis(PCA)and genetic algorithm-based BP neural network optimization(GA-BP)in PSOFS,it was concluded that GA-BP method had better optimization effect of olfactory feature dimension.After treatment with MCCV and GA-BP,PSOSF was optimized into a new soil olfactory feature space(NSOFS).The visual analysis results of NSOFS showed that the olfactory response of all sensors contributes to the modeling,and there was no redundancy.(4)Research of pattern recognition algorithm based on soil olfactory feature spaceThe prediction performance of BP neural network(BPNN),support vector machine regression(SVR)and partial least squares regression(PLSR)on PSOFS was compared and evaluated in this paper.The results of it showed that there was a certain correlation between soil olfactory characteristic space and soil organic matter content.BPNN,SVR and PLSR models were constructed based on NSOFS.The results showed that the prediction performance of the models were better than that of PSOFS models,and PLSR model could achieve basic quantitative analysis.And its prediction performance indexes of R2 V,RMSEV and RPDV were 0.86,2.87 and 2.60 respectively.(5)Comparison of artificial olfactory detection and spectral detection of soil organic matterThe spectral response of soil was obtained based on the mid-infrared ATR spectrum technology.In this paper,it selected the band of 400~1600cm-1 as the modeling data,and constructed the prediction model between soil ATR spectrum and soil organic matter content.The best modeling results of ATR spectrum were R~2_V= 0.94,RMSE_V= 1.48 and RPD_V= 3.85.The results of the comparison between the artificial olfactory detection and ATR spectral detection of soil organic matter showed that the prediction performance of BPNN,SVR and PLSR models based on the artificial olfactory detection was generally lower than that of the middle infrared ART spectral detection,and the models of the former were failed to achieve the goal of accurate quantitative analysis.(6)Optimization of artificial olfactory detection method of soil organic matterThe influence of the temperature modulation mode of equal temperature interval and equal voltage interval on the modeling prediction results is compared and analyzed.The result concluded that the equal voltage interval modulation method is better than the equal temperature interval modulation method.A method of ergodic comparison of modeling prediction performance based on multiple data study areas was proposed,and the optimal sampling time was determined to be 160 seconds.Two different combination prediction models of SVM-GMDH and PLSR-BPNN were proposed,and the results showed that the prediction effect of SVM-GMDH was not ideal,and the PLSR-BPNN model had higher prediction performance than the single model of BPNN,SVR and PLSR,which could be used as the optimal estimation model for soil organic artificial olfactory measurement.Although the artificial olfactory detection method of soil organic matter studied in this paper was not as accurate as art spectrum detection method,it could achieve accurate quantitative analysis and provide a reference method for rapid,economic and non-destructive detection of soil organic matter.