Research On Underwater Image Restoration And Target Detection

The routine and emergency detection of underwater facilities is the key to ensuring the normal operation of the ship lock.Due to the complicated water body of the ship lock,there are suspended materials such as sediment and microorganisms,and the water turbidity is uneven,resulting in low contrast and blurred image of underwater facilities.Problems such as color distortion directly affect the judgment of the operation status of underwater facilities.The research on underwater image restoration and target detection methods is of great significance for improving the detection level of underwater facilities of ship locks.According to the similarity between underwater and foggy and nighttime environmental image imaging models,dark channel and LMCV priors are directly used for underwater image restoration.The results show the limitations of the two methods.Based on the similarity of transmittance in dark channel and LMCV prior restoration methods,a new expression of transmittance is proposed,which can be effectively used for the estimation of transmittance of images in various scenes.It is obtained by transmittance refinement and background light estimation.The restoration results show that the image has a higher evaluation index(UIQM)after restoration,and can also effectively restore nighttime and foggy images,but it cannot be used for turbid underwater image restoration.The turbid underwater environment was constructed to test the visibility and attenuation characteristics of RGB three kinds of light in a hydrophobic environment.The results show that,unlike the seawater environment,the blue light decays fastest in the turbid water environment,followed by the red light and the green light.On this basis,based on the relationship between the attenuation difference between the color channels and the transmittance,the transmittance and background light of the image are estimated.An underwater image restoration method based on channel attenuation difference is proposed.The restoration results show that compared with the above methods and the existing restoration methods,the UIQM evaluation index of the turbid underwater image after restoration is higher,and the seawater image can be effectively restored.According to the detection requirements of the fastening screws of underwater facilities,the characteristics of different underwater targets are analyzed.The underwater targets are detected by Adaboost algorithm and cascade classifier,and the repeated windows and error windows are eliminated by non-maximum suppression.The detection results of the LBP,Haar and HOG classifiers in the turbid underwater image are compared.The results show that the classifier corresponding to the HOG feature has better detection effect under turbid water.