Research On Recognition Of Wide Range Of Power Facilities Based On Sar Imagery

The northwestern part of China has built a large number of wind power bases due to its abundant wind resources,but the region is also prone to natural disasters as well.After a disaster occurs,rapid collection of effective information on wind power bases is essential to carry out emergency disaster relief operations.The characteristics of radar remote sensing make it widely used in emergency response research.For the severely damaged wind power bases,it takes a lot of manpower and material resources to obtain the location and status of power facilities by traditional methods,which will affect the power supply and power grid stability in the disaster-stricken area,so it is important to quickly obtain the location of power facilities in wind power bases.This paper takes the wind power generation base on the outskirts of Yumen City,Gansu Province as the research area.Target objects such as wind generators and power transmission towers and riverbeds,rocks are distributed in this area.Their scattering intensities in single-polarization SAR images are very similar,making it difficult to distinguish them.In this paper,we innovatively merge the elements of spatial information in geospatial,spatial distribution features and geospatial reasoning methods into the SAR image space in the target recognition process of SAR images,and assigned attributes such as spatial location,shape,angle and spatial relationship among target objects in the SAR image space.Then use the spatial reasoning method based on spatial distribution features,the spatial reasoning method under the constraint of morphological characteristics parameters,and the watershed segmentation algorithm based on spatial distribution and morphological characteristics control markers to identify and segment wind power facilities.(1)This paper first proposes a spatial reasoning method based on spatial distribution features.The fast two-parameter constant false alarm rate algorithm is used to pre-recognize the power transmission towers in the experiment.Then,combined with the spatial distribution characteristics of transmission towers,the spatial reasoning method is used to identify the orderly distributed transmission towers,which significantly solves the problem that non-target objects and transmission towers are mixed due to similar scattering intensity in single-polarization SAR images.The experimental accuracy reaches 98.52%,while the accuracy of transmission tower identification using convolutional neural network was 90%-91%,the accuracy of the traditional constant false alarm rate method was only 84.3%.The reliable accuracy can provide a new identification idea for such orderly distributed ground objects in singlepolarization SAR images.(2)To make full use of the morphological characteristics information of wind turbines,this paper proposes a spatial reasoning method based on morphological characteristics under parameter constraints by giving attributes such as position,shape and angle assigned to wind generators in the SAR image space.A fast two-parameter constant false alarm rate algorithm with FAST corner detection algorithm is used for pre-recognition,and then the morphological characteristics parameters of the wind turbine are obtained from its skeleton,and the parameters are mainly used in combination with its spatial distribution features to identify the wind turbine.The experimental results show that the recognition accuracy of this method reaches 98.31%,which is 35.8% higher than that of using the traditional method.(3)In order to further improve the recognition accuracy of wind turbines,this paper also proposes a watershed segmentation algorithm based on spatial distribution and morphological feature control markers.Marking is performed to improve the segmentation accuracy of wind generators.In the experiment,the image was firstly segmented by threshold value and the number of pixels in the connected area,and then the segmented mixed ground objects were segmented by watershed,and the segmentation results were marked and verified with the help of morphological feature parameters and spatial distribution feature parameters.The experimental results show that the method has a great improvement in pre-recognition accuracy of the images compared to the pre-recognition of the two previous methods,and the algorithm is easy to use and can be used for practical promotion.