Research On Yield Estimates Of Lodging Winter Wheat Based On Image And Spectral Characteristics

Winter wheat is planted in our country in large area, and is the main alimentary crop inNorth China. Lodging is one of the main factors affecting the yield of winter wheat. Therefore,accurate yield estimation of the lodging winter wheat has an important practical significanceand application value for agricultural disaster damage evaluation.To overcome the shortage of the conventional methods of lodging crop yield estimation,this paper proposes a method of estimating lodging winter wheat yield based on imageprocessing and spectral analysis. A winter wheat lodging experiment is done in LangfangIndustrial Park of the Chinese Academy of Agricultural Sciences to make different lodginglevels of winter wheat artificially. RGB images and canopy spectral reflectance data forwinter wheat of different lodging grades are collected, and the yield of winter wheat aremeasured.Based on the collected spectral curve, envelope elimination are implemented to highlightthe absorption and reflection characteristics of the spectral curve, and spectral absorptioncharacteristics are extracted; the first derivative of the spectral curve are got, and red-edgecharacteristics are extracted; normalized difference vegetation index, atmospheric resistantvegetation index and optimized soil adjusted vegetation index are calculated to choose thespectral characteristics of yield estimation. Color characteristics are extracted by color spaceconversion, and texture characteristics are extracted by gray level co-occurrence matrix tochoose image characteristics of yield estimation. Two sets of above extracted characteristicparameters are used as input of estimation models.Genetic algorithms (GA) and particle swarm optimization (PSO) are used to optimizeradial base weights respectively and establish the yield estimation model of lodging winterwheat based on radial basis function (RBF) neural network, winter wheat yield are estimated through three periods of spectrum and image lodging index. Based on the evaluation, thePSO-RBF neural network model is proposed for yield estimation of lodging winter wheat.The proposed yield estimation model of lodging winter wheat meets the needs ofnon-contact quantitative yield estimation of the lodging winter wheat. Therefore, theestimation model can forecast lodging winter wheat production better, assist in making surethat agricultural production is going on well and agricultural technology is improved, as wellas provide good guidance on the standardized and matured planting of winter wheat.