Linear Target Detection Based On Feature Clustering And Shape Analysis

In recent years,the target detection technology has developed rapidly.In the cases of backgroud noise or poor image quality,how to put forward an effective algorithm to detect the target accurately is the difficulty and keypoint in the field.Focused on the specific problem and fully integrated the prior knowledge of linear objects,in this thesis,based on feature clustering and shape analysis,we have a study on the linear target detection to overcome the interfering factors and improve the accuracy of detection algorithm.For the main research results,a brief overview is given as follows:(1)A method of airport target of image detection based on edge feature is proposed.The method is mainly divided into the following steps:(a)ROA edge detection;(b)feature extraction and cluster selection;(c)edge points fitting and connecting;(d)parallel line detection.In this thesis,the algorithm is used to describe the edges accurately by using multidimensional features.By clustering and screening,we avoid the situations that require different thresholds for different images.According to strong noise in SAR images,in this thesis,we propose an algorithm to fit the candidate edge points together based on the least square method.The experimental results show that the algorithm is robust and accurate,also it is suitable for real-time detection.(2)This thesis presents a new method for crack detection based on contrast-limited adaptive histogram equalization and shape feature.The steps of this method are as follows: Contrast enhancement is performed by using contrast-limited adaptive histogram equalization;By using bilateral filtering technology,the image is denoised;The foreground image is obtained by local adaptive threshold segmentation;Shape feature analysis of connected region is based on region and contour,and then crack detection is finished by screening.The algorithm proposed in this thesis is fully combined with the gray and geometric priors of cracks.The experimental results show that the algorithm has high detection rate and accurate positioning to meet the target requirements.