| The existence of mixed pixels reduces the accuracy of classification of remote sensing images.The linear spectral mixture model is the most widely used spectral mixing model,its applicable observation scale and the selection criteria for the number and the spectral characteristics of the endmembers are scientific problems to be resolved.In this study,based on theoretical derivation and computer numerical simulation methods,the effective radiant flux contribution of the micro-facet to the mixed spectrum in the observation filed is analyzed,and the integral linear spectral mixture model with the zenith angle of the micro-facet as the key parameter is established.Then based on this model,the applicable observation scale of the linear spectral mixture model is derived by analyzing the relative error between integral linear spectral mixture model and linear spectral mixture model.On the basis of this condition,the mixed spectrum observation experiment of soil and green plant was designed and carried out.At last,taking the measured spectrum as the exact value,the fitting accuracy of the two-endmember model(soil-green plant)and the three-endmember model(soil-green plant-shadow of green plants)is compared and analyzed.The main conclusions are as follows:1.The relative error of the linear spectral mixture model and the integral linear spectral mixture model depends on the semi-instantaneous field of view of the remote sensing detector.In the case where the reflectance of the ground object is uniform,the relative error has a deterministic relationship with the tangent function of the semi-instantaneous field of view.Other than that,when numerical simulating the random reflectance,the relative error has a significant linear relationship with the tangent function of the semi-instantaneous field of view,and the semi-instantaneous field of view increases as the semi-instantaneous field of view increases.Outdoor and indoor spectral experiments prove that,under the condition that instantaneous field of view of the spectrometer is 12.5°,the relative error is less than 3%.And the integral linear spectral mixture model has better fitting accuracy than the linear spectral mixture model.At the same time,the relationship between the two models is not affected by changes in the topological position of the ground object.Due to the most spectrometers and remote sensing satellites have very small instantaneous field of view,taking into account the simplicity of the calculation,the linear spectral mixture model can be used for spectral unmixing.And the fitting error is determined by the semi-instantaneous field of view.In addition,the integral linear spectral mixture model must be used.2.Due to the three-endmember model introducing the shadow of green plants has higher fitting accuracy because it considers the influence of the number of endmembers and the spectral characteristics of endmembers on the mixed spectrum.And the fitting error between the two-endmember and three-endmember models has a significant positive linear correlation with the area ratio of shadow;in addition,the light background is more sensitive to the change of area ratio of shadow.Therefore,in the process of decomposing the mixed spectrum using the linear spectral mixture model,the number and spectral characteristics of the endmembers are the key parameters to ensure the accuracy of decomposition.It is necessary to select the number and the spectral characteristics of endmembers under the corresponding observation scale conditions according to the purpose of the mixed pixel decomposition in practical applications. |
