Image Enhancement-Based Method for Detecting Spatial Targets in Ground-Based Observations
In the task of detecting spatial targets from ground-based observation images,the degradation or blurring of the star image will lead to unsatisfactory detection results,and the image enhancement methods based on filtering and denoising are difficult to apply to the enhancement of the star image that is degraded or blurred due to various factors,which will affect the subsequent target detection tasks.In order to solve the above problems,this paper proposes a spatial objects detection method for ground-based observation based on image enhancement,which can be divided into two parts:1)In the image enhancement part,inspired by the image dehazing deep learning model,an image enhancement model is designed to estimate the degradation parameters,and the model uses U-Net as the basic structure for multi-scale feature extraction and fusion,and embeds the pyramid pooling module and the structure multi-scale residual module in the network structure to improve the quality of image enhanced and restored.Finally,the estimated parameters are used to restore the enhanced image from the degraded image.2)In the target detection part,the FLD(Fast Line segment Detector)algorithm is used to detect the straight line trajectory left by the relative motion of the target and the detector in the star image.The experimental dataset verifies that the average PSNR(Peak Signal-to-Noise Ratio)and SSIM(Structural SIMilarity)of the images enhanced by our method are increased by 124.51%and 64.28%,respectively.Before and after image enhancement and restoration,the average accuracy of object detection is increased by 22.15%,and the average missed detection rateis decreased by 20.6%.Compared with other image enhancement methods based on filtering and denoising,this method can adapt to a variety of star image degradation or star image blurring situations,has better image enhancement effect,and also achieves higher accuracy in target detection tasks.
NETL
NSTL
万方数据
