Research And Application Of Color Uniformity Algorithm For High Spatial Resolution Remote Sensing Image

The analysis and processing of time series remote sensing images is an important technical means of ground object change monitoring.Due to the influence of remote sensing image acquisition time and ambient light,there is a problem of inconsistent color of remote sensing images in different phases in the same area.In addition,most of the largescale remote sensing images obtained in reality are mosaic and splicing of multiple images collected by different types of satellites.Due to the differences of satellite shooting conditions,the problem of color discontinuity in the same remote sensing image is also very prominent.Therefore,it is an important premise to realize the accurate recognition and dynamic monitoring of remote sensing ground object targets to recognize the ground object after the color of remote sensing image is consistent(uniform color).Histogram matching is a commonly used uniform color method,but this method ignores the spatial structure information of ground objects.The uniform color effect is quite different from the real spectrum of ground objects,and the effect of ground object recognition is often not good.Another method is to carry out feature recognition for areas with different colors.The trained feature recognition model is not universal,which greatly increases the cost of manpower and time.Soft color segmentation refers to the color stratification of the image to make the color of the ground objects in each color layer the same,so that the ground objects with different colors can be separated to realize the change of color.In this paper,the soft color segmentation model is introduced into the uniform color processing of remote sensing images.The main work is as followed:(1)In order to obtain image color clustering information in advance for soft color segmentation,the automatic extraction of color is integrated into the soft color segmentation model,and the joint training model of color extraction and color decomposition of remote sensing image is established.For the color layer obtained by decomposition,the ground object matching method is given to realize the consistency of the same ground object color in different images.Using the places365 stand data set training model,the model is applied to the image uniform color processing of two public remote sensing data sets,namely the high spatial resolution aerial remote sensing data sets of Christchurch,New Zealand in 2012 and 2016(remote sensing images with color differences in different phases in the same area,with spatial resolution of 0.2m),And the high spatial resolution satellite remote sensing data set in East Asia(spliced from six satellite remote sensing images,with a spatial resolution of 0.45M).The results show that compared with histogram matching,after using the proposed method for uniform color,the accuracy and average intersection ratio equalization index of feature recognition have been improved,which verifies the effectiveness of the proposed method.(2)The proposed uniform color algorithm is applied to the dynamic monitoring of reclamation,and the uniform color processing of high-resolution satellite remote sensing images in Tianjin Binhai New Area in 2016 and 2020 is realized.On this basis,aiming at the problem that the complex ground feature of high spatial resolution remote sensing image data set makes the neural network difficult to train and the classification accuracy is not high,a high-resolution remote sensing image reclamation detection framework based on MCFCN is proposed.The U-Net structure network with multi-level constraint loss function is used to segment the remote sensing image semantically and extract the ground object information,and then the post-processing methods such as full connection conditional random field are introduced to optimize the results.The experimental results show that the overall accuracy of reclamation feature segmentation,F1 score,Kappa coefficient and m Iou are 96.73%,92.87%,90.28% and 86.82% respectively.On this basis,the dynamic change characteristics of land use in the reclamation area from 2016 to 2020 are analyzed and extracted,which provides effective technical support for the intensive use management of reclamation.