Study On Remote Sensing Image Preprocessing Method And Landslide Feature Identification Of UAV In Northeast Yunnan Mountain Area

The area of northeastern Yunnan province is prone to landslides.There have been many large-scale catastrophic landslides.A landslide in Touzhai Village,Panhe Town,Zhaotong City on September 23,1991 killed 216 people,injured 7 people,and caused a direct economic loss of about 12 million yuan.At present,it is one of the important means to reduce landslides disaster by using remote sensing image to monitor its activity,occurrence and development process.Due to the advantages of fast,flexible,low-cost UAV remote sensing technology in acquiring landslide disaster remote sensing data,and the temporal and spatial resolution of the acquired remote sensing image being higher than that of satellite remote sensing data,UAV remote sensing technology has been widely used in landslide monitoring and disaster loss assessment.However,in criss-cross canyons and mountainous areas with huge terrain elevation difference,the UAV images acquisition is usually severely affected by terrain,local circulation and airflow,which causes that the acquired UAV images often have problems such as blurring,under-exposure,large shadow area and strong distortion.At the same time,there are matching errors and mismatches in the process of UAV image mosaic.These low quality UAV remote sensing images cannot satisfy the identification and judgment of landslide features,activity and danger.Therefore,research on how to preprocess the UAV images through image processing methods to obtain high-quality UAV images that meet the needs of the work in the deep canyon mountains has become the key to monitor and research the geological disasters such as landslides.Taking the drone landslide image in the mountainous area of northeast Yunnan as the research object,this paper carried out the methods of automatic fuzzy image recognition,motion fuzzy image recovery,image enhancement and coarse error elimination in mountainous areas.These methods are used to process UAV images to obtain high-quality images that can meet the requirements of landslide feature recognition.The main conclusions of this paper are as follows:(1)The edge feature of UAV image is an important parameter to reflect the sharpness of the image.A comparative analysis of the four classic edge feature extraction operators is currently performed.The experimental results show that Gauss Laplace operator can extract the edge of UAV image feature well.In this paper,the Gauss-Laplace algorithm is proposed to extract the UAV image edge features accurately,and the fuzzy image recognition is carried out by combining the image features gray variance.Combined with the experimental analysis,the fuzzy images can be automatically identified quickly and accurately from obtained experimental 517 UAV images,with the accuracy of 100%.This algorithm has a good effect in automatic recognition of blurred images.(2)UAV images in mountainous areas are prone to motion blur due to local airflow.The cause of motion blur in UAV images is analyzed.The key factor in recovering blurred images is to obtain accurate blur kernels.This paper proposes an algorithm that uses prior knowledge to estimate the fuzzy kernel.It calculates the residuals between the fuzzy image and the sample image,fits a polynomial,calculates an accurate fuzzy kernel,and combines the Wiener filter algorithm for image restoration.The obtained drone blurred image is verified,and the estimation accuracy of the horizontal blur kernel of the motion blurred image is 100%.This method can recover the motion blurred image well and improve the image quality.(3)Affected by the mountain terrain factors,the images obtained by UAV have the phenomenon of insufficient exposure and shadow on the images.The image pixel brightness in this area is compressed,and the lacking of information affects the landslide information identification.Based on a comparative analysis of four current remote sensing image enhancement algorithms,a combination algorithm for removing shadows from UAV remote sensing images is proposed.This algorithm combines the Retinex algorithm with a two-dimensional gamma function to remove UAV images.The shadow area can also correct the unevenness of light and dark generated after image enhancement.The combination algorithm is better than the traditional UAV image shadow removal algorithm.(4)In the process of UAV image mosaic,the accuracy of feature points directly affects the final Mosaic quality,and obtaining robust feature points is the key step of image Mosaic.Based on the improvement of the traditional RANSAC algorithm,a new algorithm combining with graph theory for feature point coarse error elimination is proposed in this paper.The new algorithm can obtain high precision homography matrix.The method of combining RANSAC algorithm and graph theory is used to construct the UAV image matching feature point coarse error elimination system,which can eliminate coarse error of the feature points extracted from UAV mountainous area image and obtain the feature matching points with high robustness.(5)The pre-processing UAV remote sensing image and the original UAV image were used to identify the features of the landslide in Longtoushan Town,Ludian County and Fatu Village,Qiaojia County.The results show that the image recognition effect of pre-processing is much better than the original image.