Study Of Methodology In The Salt Pan Information Extraction Based On High-Resolution Images

Nowadays,with the increase of spatial resolution,high-resolution images contain growing number of information.Since the beginning of high-score project acquired images have been successively applied into national production and research fields,i.e.land use,mine monitoring,and disaster monitoring.However,due to overlarge data size and irregular noise information,the processing of the image in practice encountered with many problems.The object-based image analysis(OBIA)classification method is the mainstream of high-resolution image classification and shows a good classification effect.In this study,we take the "big geology" survey as background and adopt the object-oriented classification method to extract the land area information of a certain Salt pan in Dongying City.Working processes and results are displayed as following described.The thesis is mainly divided into two parts the image segmentation and the image classification.First of all,the image is smoothed,sharpened,and certain pre-processing operations,wherein the morphological hybrid opening and closing operations are introduced to perform smoothing processing.During the image segmentation,the actual geophysical features of Salt pan in the District considered.The phase gradient was used as the modified gradient operator for watershed segmentation,and the Canny operator with the same denoising function was also used for comparative analysis.In the watershed segmentation,the morphological threshold operator and the corrosion operation are introduced to perform watershed image segmentation based on both foreground and background,and a good image segmentation result is obtained.As high-resolution images would contain abundant features such as spectrum,texture,and geometry,it is necessary to calculate each based on the results of image segmentation before image classification.After analyzing the differences between their eigenvalues,we select the sample data for each item.In this paper,KNN and Random Forest(RF)classifiers are used to classify the data.Based on the real survey data,the real labels are classified and the confusion matrix was used to evaluate the accuracy.The article can be concluded as following:1.Experiments show that the phase-consistent gradient operator produces better results than Canny for image segmentation in the segmentation process in Yantian District in Shandong Province.2.The use of mathematical morphology filtering and marker watershed algorithm shows a considerably significant optimization effect on the process of image segmentation,by largely reducing the number of misclassification and misclassification,especially for negligible small-sized objects,forces to make the segmentation result meet the requirements of the classification.3.In this paper,KNN and RF classifiers are used to classify Salt pan district,and both have achieved good results.Specifically,we consider that the random forest classifier displays a better result,Additionally,the accuracy of the classification is improved by using the geometric and texture features for the classification of Salt pan.The classification accuracy is up to Salt pan UA=0.917,OA=0.890,and KAPPA coefficient of 0.834.