Multi-features Based Hierarchical Detection Of Transmission Towers With High-resolution SAR Images

As an important part of the power system,transmission towers ensure the safety and stability of the entire transmission line during operation.Its detection has been widely concerned,and the relevant technical means are constantly being updated and upgraded.The traditional detection methods mostly adopt human survey and unmanned aircraft survey,which is time-consuming and labour-intensive.With the rapid development of Synthetic Aperture Radar(SAR)systems,massive amounts of high-resolution SAR images are available.The use of high-resolution SAR images for transmission tower inspection is becoming a mainstream technology.At a finer spatial scale,the SAR images are richer in detail of the ground target,giving transmission towers more distinct spatial and geometric characteristics.However,the interference from background clutter,coherent speckle noise and other small targets is also significantly enhanced,posing difficulties and challenges for the detection of transmission towers.To address these problems,the thesis proposes a hierarchical detection method that combines the spectral scattering characteristics,spatial distribution characteristics and geometric features of transmission towers in high-resolution SAR images.The main work and the research results obtained are as follows.(1)To investigate the scattering mechanism and scattering effect of transmission towers in SAR images.And propose to simulate the imaging characteristics of transmission towers in SAR images using ray tracing method.Obtaining the basis of the spectral scattering characteristics,spatial distribution characteristics and geometric characteristics of transmission towers provides the basis for the construction of subsequent detection models and detection methods.(2)In order to coarsely extract transmission tower pixels using amplitude feature,a new detection model(SCRBF)combining Bilateral Filtering(BF)and Signal-to-Clutter Ratio(SCR)is proposed.Firstly,a SCR detection model is constructed to enhance the contrast between transmission tower pixels and the background.Secondly,the background noise is suppressed by BF.Then,a new SCRBF detection model is constructed and the coefficient of variation method is used to calculate the model weights to balance.Finally,the best preliminary detection results are obtained by the best confidence level.(3)Among the transmission tower pixels extracted by the above method,pixels originating from the same transmission tower are spatially aggregated,and this feature is used to filter out some of the non-transmission tower pixels.However,due to the SAR imaging mechanism and the structural characteristics of transmission towers,the pixels of individual transmission towers have a certain degree of dispersion.Traditional clustering methods tend to cluster the same transmission tower pixels into multiple clusters.To this end,an improved density space clustering method(DBSCAN for Transmission tower,T-DBSCAN)for transmission tower pixels is proposed.The method firstly uses the nearest neighbor rule to calculate and mark the pixel distance to effectively avoid clustering transmission towers into multiple target clusters,and constructs the proposed transmission tower pixel clusters.Secondly,the clustering strategy of the original DBSCAN(Density-Based Spatial Clustering of Applications with Noise)algorithm is changed to cluster the pixels marked in the bottom 20% of the proposed clusters.As a result,the noise points are filtered and eliminated to improve the clustering efficiency and obtain the transmission tower candidate detection results.(4)In order to further detect transmission towers accurately,a geometric feature constraint model applicable to transmission towers is proposed.Firstly,the principal axis angle is calculated,and the Minimum Bounding Rectangles(MBR)are constructed for the candidate clusters that satisfy the principal axis angle.Then,the principal axis combined with the minimum convex package can be used to construct a more accurate MBR for the candidate transmission towers.Finally,the aspect ratio and area threshold are used for further screening,which can effectively remove the false alarm targets and achieve accurate transmission tower detection.In order to verify the accuracy and feasibility of the proposed method,one simulated SAR image and six Gaofen-3(GF-3)SAR images,both of size 433×535 pixels,are used as experimental data for hierarchical transmission tower detection experiments based on spectral scattering features,spatial distribution features and geometric features,respectively.The experimental data are compared with the two popular SAR image transmission tower detection methods.The experimental results show that compared with the CFAR combined with Extended Fractal(EF)detection method,the True Positive Rate(TPR)is the same at 100%,while the mean precision and mean F1 values are improved by 69% and 54.39% respectively,and the mean running time is saved by 47.314 seconds.Compared with the contrario detection method,the TPR,mean precision and mean F1 values increased by 33.34%,75.46% and 64.98%,respectively,and the mean running time has saved by 1.5 seconds.The results of the qualitative and quantitative analyses show that the proposed method has the capability to accurately detect transmission towers,with excellent performance in terms of detection accuracy and running speed.The feasibility,accuracy and superiority of the proposed method are fully verified.There are 55 figures,25 tables and 150 references in this paper.