The Simulation And Optimization Of Adjacent Effect In Optical Remote Sensing

In the optical remote sensing,there is atmospheric interference between the imaging system and the imaging target,which seriously affects the quality of the optical radiation transmission and the quality of the remote sensing image.The adjacency effect of optical remote sensing refers to the contribution of non-target pixels to the target pixel radiance in remote sensing,also known as the cross radiation effect of the atmosphere.The adjacency effect generally increases the radiation value of the dark pixels,and reduces the radiation value of the bright pixels,which makes the edge of the remote sensing image blurred,the contrast and quality of image decreased.The results show that the influence of the adjacency effect must be taken into account when the spatial resolution of the sensor is higher than 1 km.Therefore,with the continuous improvement of the spatial resolution of optical remote sensing and and the quantitative degree of the application remote sensing information,the research on the mechanism and simulation method of adjacency effect has become a key technical problem to be solved urgently.Firstly we carry out the simulation of adjacency effect based on Monte Carlo method in this paper.Based on the radiation transmission model,the photon transport environment is constructed.The reverse Monte Carlo method is used to simulate the whole process of photon transmit in the atmosphere from the remote sensor to the surface.And the photon statistical distribution is obtained,which called the atmosphere point spread function(PSF).The radiation transmission model is used to calculate the surface radiation field,and the cross radiation is calculated by the convolution of the atmospheric PSF and the surface radiation field.The influence of the atmospheric transmission upward is added to obtain the radiance value of the remote sensor entrance.It analysis the two observation conditions of vertical and oblique conditions and compute the factors of adjacency effect.The factors considered includes the surface reflectivity distribution,the imaging spatial resolution,the atmospheric visibility,the solar zenith angle,the observation zenith angle and azimuth angle.The results show that the surface reflectivity distribution,atmospheric visibility and observation zenith angle have the most significant effect on the adjacency effect.With the increase of the background reflectivity,the atmospheric visibility decreases,and the observation direction is close to the zenith direction.In this condition,adjacency effect becomes readily apparent.On the basis of theoretical and simulation research,the proximity effect simulation image is obtained on a digital scene.Then,considering the uncertainty of Monte Carlo simulation method,the algorithm optimization based on BP neural network is carried out.Based on the Monte Carlo simulation results of atmospheric PSF,a two-layer feedforward neural network with enough hidden neurons using sigmoid function and linear output neurons is designed.And Levenberg-Marquardt backpropagation algorithm is used to learn the relationship between the atmospheric PSF and atmospheric parameters,observation geometry,wavelength and other factors.The research focus on the structure of the neural network,especially the design of the output layer.The neural network with high dimension output is divided into multiple small networks with only one dimension output,running parallel on multiple workstations or PCs,which solves the difficult problem of high-dimensional output of neural network effectively,and improves the computational efficiency significantly.After training the neural network,it can obtain the corresponding atmospheric PSF quickly for any input conditions within the training sample.And for the input conditions that are different from the training samples,a reasonable estimate of the atmospheric PSF can also be obtained.The results of several experiments shows that the accuracy and efficiency of this optimization method are satisfactory.The research results of this paper enriched the entire simulation model of optical remote sensing imaging,and laid the theoretical and methodological foundation for the design optimization of imaging system.At the same time,it help to carry out research on the adjacency effect correction algorithm of remote sensing image,and improve the quality of remote sensing image effectively.