Research On UAV Denoising Method Based On Compression Sensing

In the electronic warfare,it is one of the trends for future battlefield reconnaissance technique that utilizing the Unmanned Aerial Vehicles(UAV)to locate objects,identify and understand image information by organically combining with conventional electromagnetic reconnaissance technique,which can significantly promote the battlefield situational awareness and analytical ability of our troops staff.And thus,it is of great value to study image denoising technique of noisy UAV images.The main work of this dissertation is presented as follows:(1)Aiming at the characteristics of UAV image and complex and diverse interference noise,after analysis and comparison,a method applying compressive sensing denoising(CSD)is presented.According to the sparse components,the CSD method is implemented by separating effective image information and noise.The validation experiment,with classic image denoising methods serving as comparison,is simulated under different type of noises.And the results show that the CSD method has better denoising effect on UAV images and higher PSNR of image components,and can fulfill the need of UAV images denoising under different working environments.(2)Aiming at the problem that fixed sparse dictionary of compressive sensing theory cannot comprehensively represent the structure characteristics of images and thus causing terrible denoising effect,a sparse denoising algorithm base on UAV noisy image dictionary learning(SDDL)is proposed.First,the algorithm uses the noise image as a training sample,selects the initialization dictionary,and sparses the noisy image.Then K-SVD method is applied on learning from the sparsely representation matrix to update the dictionary,the procedure of which is also called dictionary learning.After iteratively computation,the updating is stopped until the residual became less than the pre-set interval.And finally the new dictionary is obtained.In the dissertation,through the comparison with parameterized dictionary and the one trained on natural images,the experimental results show that the SDDL algorithm has better performances on expressing image structure characterisics and denoising.(3)Aiming at the problem that the UAV image sparsity is reduced by mixed noise,a combining denoising algorithm based on compressive sensing(CDCS)is introduced.The CDCS algorithm includes two parts: rough denoising and elaborate denoising.The complete denoising procedure was implemented by following steps: firstly dislodge the impulse noise in combining noise with filtering smoothing technique to reduce the destruction on image sparsity.Then the images that being rough denoised are represented sparsely.The measured value of image components is obtained by linear mapping.Finally the denoised image is reconstructed and output.The experimental results show that the algorithm improves the influence of mixed noise containing impulse noise on image sparseness and improves the image denoising performance.(4)Aiming at the problem that the detail feature missing caused by denoising,a layered denoising algorithm based on wavelet fusion(LDWF)is proposed.The wavelet fusion theory are applied in LDWF,compromising the images denoised respectively by the median filter method and the SDDL method.According to the characteristic of certain types images and interested target,the fusion rules are designed and presented as the low frequency fusion rule based on the energy spectrum of local information and the high frequency fusion rule based on the variance spectrum,respectively.Evaluated by the peaked signal-noise ratio,standard deviation and entropy,the experimental results demonstrates the effectiveness of the algorithm and that the LDWF can improve the visualization of denoised image details.