Unmanned Aerial Vehicle (uav) Monitoring Key Technology Enhanced Image Independent Component Analysis

In the recent years, unmanned aerial vehicle monitoring is widely used in militaryand civilian fields to replace manual monitoring, reducing the risk and cost at work.However, unmanned aerial vehicle monitoring is always influenced by many factorssuch as noise, haze and motion blur, which will deteriorate the visual effects of theimages and affect their post processing such as target recognition and survey. For thisreason, the algorithms suppressing additive white Gaussian noise and haze are studiedand realized in this paper from the perspective of independent component analysis. Thecontent of this thesis includes the following three parts.First, to address additive white Gaussian noise, the independent component analysisshrinkage denoise technology is studied in the paper, including shrinkage function andindependent component analysis sparse coding.(1) The shrinkage function of the sparse distribution with additive white Gaussiannoise and its post processing is analyzed. The probability model selectioncriterion is discussed. Then, the connection between sparse coding andindependent component analysis is discussed from the perspective of minimummean square error.(2) The effect of preprocessing and parameter selection is discussed and the SNRcurves with different window size are shown. In our experiment, independentcomponent analysis shrinkage denoise technology is better than wiener filter andmedian filter. Compared with wiener filter, under strong noise with0.7standarddeviation condition, the exact result is2.7dB for natural images and3.4dB forartificial images.Second, to remove haze, the nonlinear independent component analysis image dehazetechnology is studied and improved, including transmission estimation, its smooth andairlight estimation.(1) Based on the derivation of transmission and its noise, the noise power oftransmission is given. Through analyzing how to derive the Markov smoothmodel of transmission, the method for solving the model is given. To estimateairlight, the current main four kinds of method are compared and their usabilityto nonlinear independent component analysis image dehaze is analyzed,improving the algorithm.(2) The effect of window size and noise to transmission’s estimation result isanalyzed. And the noise power computation result of transmission and empiricalthreshold of noise power are verified. We use synthesized picture with groundtruth to test the mean absolute error of the estimation result of transmission andobject image. They are0.072and0.089respectively. At last, based on OpenCV, multiple processor programming and GPU programming,we use C to implement the above algorithms. Through selecting proper tools, ourprogram is of portability. Through analyzing the key point, effective optimization ismade. For a512512pixel image, denoise program implemented using GPUconsume about0.38s and dehaze program implemented by GPU consume about7.97s.The denoise and dehaze programs belong to signal processing software of groundcontrol station of the unmanned aerial vehicle communication chain. The speed ofdownlink request the program can process a pixel image within one second.The program implemented through GPU can be used to image denoise in real time.However, our dehaze program can only be used to non-real-time processing.