Research On Shadow Processing And Information Extraction Of Unmanned Aerial Vehical Images

UAV aerial survey system has gradually become the main source of image data in large-scale engineering surveying and mapping projects due to its high flexibility,good adaptability,high cost effectiveness and other advantages.However,easily affected by factors such as light,clouds,and terrain,the acquired images sometimes contain shadow that will result in image degradation such as low contrast,unclear target features,and blurred details.Theses degraded images cannot be used effectively and the subsequent projects will be consequently affected.Therefore,finding a certain shadow processing method that is able to transform low-quality images into high-quality ones and restore shaded information from shadowed areas has become a challenging scientific task.This article focuses on cloud shadows of the UVA images in actual projects and selects representative image data to discuss the following two core contents:shadow processing and information extraction of UAV images.The main research works and conclusions are as follows:1)For the spectral characteristic of shadow that the gray value in the shaded area is smaller than the gray value of non-shaded areas in this paper,a shadow detection method based on the histogram threshold method is studied.Meanwhile,according to the characteristic of image that it shows different spectral features under different color spaces,a shadow detection algorithm that combines the invariant color model with the Otsu method is proposed.Experimental results show that both shadow detection methods can distinguish between shadow and non-shadow in the image.The combined algorithm proposed in this paper can be better applied to shadow detection in that the result of shadow detection is more reliable than the histogram threshold method.2)Firstly,the theory,characteristics and applicability of shadow processing of three image processing algorithms based on non-model--algebraic multiplication,histogram equalization,and homomorphic filtering--are studied,and the three algorithms are implemented to process the shadow in the image.The experimental results show that the non-model-based image processing algorithm increases the image quality index value to a certain extent,removes the influence of the shadow,and restores the partial feature information of the shadow area.Secondly,this paper mainly focuses on the applicability of Retinex theory based on RGB color space in image shadow processing,and specifically analyzes the applicability and implementation process of single-scale algorithm,multi-scale algorithm,and improved multi-scale algorithm.Applying to the shadow processing of UAV images,the experimental results show that all three algorithms can process shadowed images,but the improved multi-scale algorithm is more effective.After processing,the quality index value and the definition of the image are higher,the detail information and the information volume of the image are larger,and the quality of the image is improved.But compared to the original image,there are some deviations of color from the original image.For the disadvantages of the improved multi-scale algorithm in the RGB color space,the superiority of the Retinex algorithm based on the HSV color space in shadow removal is studied.The experimental results show that this algorithm can,to a certain extent,solve the problem of image color deviation and at the same time improve the image quality and restore the features in the shadow region.3)For the image processing results based on non-model image processing algorithms and Retinex algorithm for image shadows,minimum distance method,maximum likelihood method,and Mahalanobis distance method are selected through experiments for the information extraction of original images and processed images.The experimental results show that the Mahalanobis distance method do well on the information extraction of shadowed original images and processed images,Visual effects can be seen by using the Mahalanobis distance method to extract relatively complete artificial surface features,bare soil,and vegetation.There are fewer spots in the image,and the effect is more obvious.In the meanwhile,through the accuracy analysis,it can be shown that the image information extraction accuracy and its OA value and Kappa value are the highest.