| Crop vegetation coverage is asignificant parameter about agricultural research.On the one hand,crop density reflects crop growth stage,provides the guidance for crop irrigation and fertilization.On the other hand,it can make investigation for crop distribution and estimate overall crop yield,price forecasts and mass crop evaluation.Due to the acquisition of satellite data is restricted by the cycle and low resolution,there are some limitations in the monitoring of farmland information by satellite remote sensing.With the development of unmanned aerial vehicle(UAV)low altitude remote sensing platform,the platform has the flexibility and precision to improve the shortcomings of the satellite remote sensing.Therefore,this paper selected Inner Mongolia Wuyuan County of Bayannur city as the study areaof estimating crop vegetation coverage.The research established agricultural low-altitude UAV spectral information perception system,andselected corn and sunflower crops as the research object to collect multi-spectral image data of crop field,by processing and analyzing the image,we studied spectral feature classification method to achieve the evaluation of crop coverage.The main research contents included Based on the estimation of the degree of coverage of a single crop land and the coverage of the mixed crop land.The following content shows the main research process and conclusions:(1)The study selected field of sunflowers for UAV image data acquisition,and accomplished supervised classification ofcrop and soil based on support vector machine(SVM),minimum distance(MD),neural network(NN)and maximum likelihood(ML),then calculated crop coverage.By using confusion matrix to carry out precision test,The image effectof the SVM method was best.The overall accuracy was 98.71%,Kappa coefficient reached 0.97.Combined with the overall accuracy and the accuracy of the extraction of sunflower,the results of the SVM method is the most close to the actual coverage,The sunflower coverage was 17.86%by SVM.(2)Paperconducted classification experiment with the selected plots of sunflowers image data by unsupervised classification method including Iterative Self Organizing Data Analysis Techniques Algorithm(ISODATA)and K-means clustering method,and evaluatedthe accuracy of the results.By comparing the experimental results of supervised classification and unsupervised classification,the classification accuracy of supervised classification method was above 95%,which was better than the unsupervised classification method,As a result,the study used the supervised classification method as a means to assess the degree of coverage of different objects,and then estimated the degree of crop coverage.(3)Research selected the remote sensing data containing sunflower and maize field to establish decision tree with expert knowledge to divide image into light soil,sunflower,shaded soil,and plastic film in five categories,and evaluatedtheir classification accuracy.The total accuracy of the decision tree was 81.64%,Kappa coefficient was 0.73.The coverage of sunflower was 34.91% and the coverage of maize was 20.79%,which was obtained by the decision tree algorithm.(4)Research establishedSVM model for surface features classification of mixed cropsfields,calculated the coverage of all kinds of surface featuresand evaluated the classification accuracy.Total accuracy of SVM model reached 89.11%,kappa coefficient was 0.85.With compared,the SVM model was better than the decision tree model to estimate the crop vegetation coverage.The coverage of sunflower calculated by SVM was 22.23%,and the coverage rate of maize was 24.46%.With the random selecting plot for universal verification,the total accuracy of the SVM model was 91.46%,Kappa coefficient reached 0.89.The applicability of the SVM model is proved to be more generalized.The studiesshow SVM model is available and have higher estimation accuracy.The rapid and timely monitoring of crop cover degree is significant for planting decision,yield estimation andcrop water requirement calculation. |
PDF Full Text Request