| The digital agricultural technologies in agriculture are one of the most frontier research in the area of modern information-based agriculture, and it will be a key technique for our country to fulfill the modernization and sustainable development in agriculture in future. The digital agriculture needs fast, real-time, accurate and the positional plant growth information and condition, but the traditional chemistry methods have not been able to satisfy the request of digital agriculture development. Therefore, it is an urgent need to some scientific research about the fast determination of plant growth information and the monitor growth condition in crops. This study mainly focused on the Brassica napus L.,which is a widely planted, high economic valued and alternative energy resource plant. By using the Cropscan multi-spectrum radiometer, the canopy values of SPAD in oilseed were predicted, and the determination of pests in leaves and straws were analyzed by using Hyperspectral imaging techniques. The main conclusion was as follows:(1) The quick and nondestructive prediction of canopy SPAD values were studied by the Cropscan multispectral radiometer. The model input variables which were spectral reflectance values underl5 bands were taken in the linear Partial Least Square models (PLS). The correlation coefficient of the calibration set was 0.7323, and root mean square error (RMSEC) was 3.4528. The correlation coefficient of prediction set is lower than the calibration set at 0.6640, and the root mean square error in prediction (RMSEP) was 2.7859. In non-linear Least square-support vector machine model (LS-SVM), the correlation coefficients of the calibration set was 0.7459, and RMSEC was 3.1567, the correlation coefficient and RMSEP of prediction set were 0.6805,3.1227 respectively.(2) The Vegetation Indices (VI) based on different bands Combination were studied for predicting SPAD values in canopy. The different vegetation indices were obtained as (R1100-R690)/(R1100+R690), (R1650-R690)/(R1650+R690), R830/R690, and R830/R560. The linear model, binomial model, the logarithm and exponential model for predicting SPAD values were established based on every vegetation indices. By comparing these models’results, the binomial models have the best performances in every vegetation indices. The optimal binomial model were derived from (R1100-R69o)/(R1100+R69o) vegetation index, and the results of correlation coefficient in calibration and prediction sets were 0.7691,0.7012 respectively, the results of root mean square error were 1.98,2.03, respectively.(3) The visible/near infrared spectral features in hyperspectral imaging system were studied for detection and recognition oilseed rape pests. The whole bands were subjected by SG, MSC, SNV and Detrending pretreatment algorithms, then the LS-SVM models based on different pretreatment methods were compared. The SNV-LS-SVM model performed best with the predicted correction rate at 90.57%. In order to make the model fast response, less input, the feature extraction algorithms of the successive projections algorithm (SPA), the coefficient method (x-LW), principal components load contribution rate analysis (PCA-Loadings) were discussed for optimal model. The results show that 13 spectral feature extraction based on the PCA-loading algorithm of LS-SVM model achieved the 88.68% correction rate. Finally, three different models (Least Squares Support Vector Machines (LS-SVM), the Least-Squares Discriminant Analysis (PLS-DA), Extreme Learning Machine (ELM)) were studied to discriminate the healthy leaves and affected leaves by aphids. The optimal model was SNV-PCA-loading-LS-SVM.(4) The discrimination of aphids in the stems of oilseed rape were studied by using spectroscopy analysis. The conclusions shows that after the minimum spectral features by SPA algorithm after the De-trending preprocessing method of LS-SVM model had the best calibration correction rate of 89.66%, and the prediction correction rate was 86.21%, the other models of the correction rate were around 80%.(5) In the research of discrimination of aphids, the textural features in hyperspectral imaging system were studied by image analysis for detection and recognition oilseed rape pests. The features concluded Contrast, Homogeneity, Correlation, Energy, Entropy and Covariance and counter gap, Distance and correlation of characteristics of second order. Taking these textural features as input variables in different models. The results show that only 69.81% prediction correction rate was get by best Extreme Learning Machine model (ELM). However, after combination of spectral signatures and image features, all the models had more than prediction correction rate 85%, and PLSDA obtained the best performance of 92.45%.(6) The dynamic devouring process of rape leaves by the pieris rapae were monitored by ADC cameras. The period of 5 hours of devouring pictures were acquired and analyzed to evaluate the damage degree of pieris rapae based on pixels processing.set the threshold was set to 0.2745,then the background segmentation worked best. Through the background processing such as segmentation, binarization and other algorithms, the missing pixels of leaves ratio was calculated for assessing the degree of seriousness of the oilseed rape pests. |
PDF Full Text Request