| In recent years,the environmental pollution,the quality of japonica rice,and the occurrence of diseases and insect pests and etc.in Liaoning area have been aggravated in different degrees with the continuous increase of fertilizer application.Meanwhile,problems also emerged in the field production and management of japonica rice,such as blindly pursuing fertilizer,lack of management basis,increasing production cost,reducing economic and ecological benefits,and so on.Along with the rapid development of new technologies,such as information technology and low-altitude remote sensing and etc.,the diagnosis of nitrogen nutrition status in japonica rice during its fertility period is of great scientific and practical significance for precision management and intelligent decision-making of japonica rice production.At present,there are still some problems in the traditional nitrogen nutrition diagnosis methods for japonica rice in space coverage area,detection accuracy and etc.,which is difficult to meet the actual production requirements of japonica rice.Technologies,such as low-altitude high-light spectral imaging of unmanned aerial vehicles(UAV)and etc.have been gradually mature,thus providing new methods and technical support for the accurate diagnosis of nitrogen nutrition in the field production of japonica rice in Liaoning province.The main research content and conclusions of this paper are as follows:(1)In this paper,with emphasis on the nitrogen diagnosis of japonica rice in Liaoning province,namely,the research object,the cultivation experiments were respectively conducted on the japonica rice in Northeast China with different N-fertilizer treatment in LiaoZhong japonica rice cultivation test station of Shenyang Agricultural University,and Qingshuitai Liutiaohe Township test field of Shenbei New District in 2015-2017;meanwhile,the UAV(unmanned aerial vehicle)low-altitude hyperspectral remote sensing test and plant protection UAV variable fertilizer application test was conducted in the growth process of japonica rice,involving the japonica rice variety of“Shen Nong 9816”.The canopy hyperspectral information was collected during the critical growth period of japonica rice,and the hyperspectral information of japonica rice was extracted by classifying and extracting the collected rice hyperspectral remote sensing images by means of particle swarm optimization algorithm,integrated machine learning,improved optimal index and other methods.According to the classification and extraction results of hyperspectral remote sensing images,the effect of pure hyperspectral information classification of japonica rice extracted by the feature selection and classification method based on the improved optimal index had the best performance,with the accuracy of 91.82%.The results showed that the method of feature selection and classification based on improved optimal index was suitable for accurately extracting the hyperspectral information of japonica rice from the field environment with lots of disturbance,and could be used for subsequent nutritional diagnosis modeling.(2)The hyperspectral reflectance curve and corresponding physical and chemical characteristics of japonica rice in Northeast China were analyzed,and the hyperspectral information in the range of 400nm-1000nm collected by the UAV was reduced by a variety of dimensional reduction methods,such as hyperspectral feature extraction,feature band screening and etc.On the basis of dimensionality reduction,multiple modeling methods,such as statistical analysis,vegetation index,machine learning,and optical radiation transmission mechanism,etc.were adopted to establish the inversion model of chlorophyll content and nitrogen content of japonica rice respectively.(3)The chlorophyll content inversion model of japonica rice was established on the basis of the extreme learning machine algorithm based on micro-particle swarm optimization,and the model input variables were 410nm,481nm,533nm,702nm and 798nm,which was superior to the chlorophyll content inversion model established by other inversion algorithms in the aspect of inversion precision.The optimum population size of PSO-ELM model for inverting chlorophyll content in japonica rice was determined by experimental study,and the optimal inertial weight w was in 0.9~0.3 linear decreasing,the learning factors C1=2.8,C2=1.1,and the speed-position correlation coefficient m was 0.6.In the PSO-ELM inversion model for japonica rice chlorophyll content established based on the optimized parameters,the model determination coefficient R~2 was 0.887,and RMSE was 0.783.(4)After carrying out dimensionality reduction for hyperspectral data of japonica rice,and extracting the sensitive bands of nitrogen content,this paper established the inversion model of water-japonica rice based on regression analysis,Gaussian process regression,classification regression tree,partial least square multiplication and other algorithms.The correlation between nitrogen content and canopy spectral information of japonica rice based on the use of machine learning algorithm was extremely significant.The R~2 of Gaussian process regression method was as high as 0.77,followed by partial least square method,and the numeral value of classification tree regression was as low as 0.55.After comprehensive analysis,the effective inversion could be carried out to the nitrogen content of japonica rice by using the nitrogen content inversion model based on the combination of hyperspectral remote sensing information and machine learning.(5)This paper established a fertilizer application model in japonica rice field based on hyperspectral nitrogen content inversion through improved nitrogen optimization algorithm,and applied precision spraying with UAV(unmanned aerial vehicle).The nitrogen proportion of the fertilizer selected for field fertilizer application was 0.447,and the regulation coefficient was assigned a value of 0.83 through experimental measurement.In addition,the plant protection UAV was used to achieve precise spraying of fertilizer in the japonica rice field.The results of this study can be used for rapid diagnosis of nitrogen nutrition status of japonica rice in Liaoning area,as well as precise spraying operation in the field according to the diagnosis results,thus providing certain technical basis and data support for the realization of decrease in fertilizer application and increase in efficiency of japonica rice production. |
PDF Full Text Request