Research On Small Target Imaging Technology And Optimization Method Based On Complex Water Surface Environmen

Image optimization is widely used in the field of water surface imaging as a way to obtain quality environmental information.In the southern region of China,a complex water surface environment is usually a water surface with interference from one of the variable weather conditions such as rain,fog or high light.By optimizing the complex water surface image,information such as geometry,spatial orientation,and color of the water surface image target can be obtained.Such information can be paired and applied to intelligent offshore equipment,environmental protection vessels and commercial vessels,thus automating the functions of surface target striking,collision avoidance,search and rescue,salvage,floating object recovery,reconnaissance and counter-reconnaissance.To this end,this paper will take the water surface under the weather conditions of fog,rain and high light as the research environment,by designing and building an unmanned boat imaging platform,using Alex Net convolutional neural network for automatic weather classification of complex water surface environment images,obtaining the weather attributes of degraded images,and then using the corresponding algorithms for de-fogging,de-rainage and de-highlighting image optimization.Finally,the impact on two advanced vision tasks,weak target recognition and background rejection,before and after optimization of complex water surface images is analyzed,which makes the localization of weak targets on the water surface clear at a glance through the recognition and localization of weak targets on the water surface and image background rejection.(1)Foggy water surface image optimization processing.According to the characteristics of imaging on the water surface environment,the appearance of foggy days will bring interference to the image acquisition,making image quality degradation,serious interference with the identification of weak targets on the water surface.To address this phenomenon,this paper analyzes the difference between foggy days on land and foggy days on water based on the atmospheric scattering model,and proposes an optimized dual quadtree method for searching the value of atmospheric light,which solves the defect of quadtree unilaterally taking the value of atmospheric light in the sky or water surface area.The coarse transmittance of the water surface image is estimated by using the feature loss optimization,and then the coarse transmittance is refined according to the guided filtering,and the refined transmittance is calculated,and the defogged image is obtained by combining the atmospheric scattering model.Then,the weak target recognition and background rejection are performed on the defogged image,which solves the problem of difficult detection and recognition of weak targets on the water surface in foggy days.(2)Rainy water surface image optimization processing.According to the characteristics of the water surface environment imaging in rainy days: the general situation of rain streaks as the foreground,will make the image blurred,resulting in the water surface weak targets are difficult to detect and recognize.For this reason,this paper introduces the attention mechanism of convolutional module(CBAM)for optimization in the algorithm based on the quasi-sparsity training of multi-decoding derivative network(QSMD)for image de-rainage.In the image rain removal process,the feasibility of quasi-sparsity rain removal for water surface rain images is first analyzed based on sparsity a priori theory,and then the data set is collected and the network model is trained.In the network model with the attention mechanism,the extraction efficiency and accuracy of the "rain layer" and "background layer" features in the channel and space are improved,which leads to better rain removal images.Through experimental comparison and analysis,the optimized rain removal method effectively improves the recognition rate and robustness of weak targets in rainy water surface environment.(3)Optimized processing of water surface highlight images.Under the sunny day highlight environment,the strong highlight will degrade the water surface image,resulting in the difficult recognition of weak targets on the water surface.Since the traditional highlight removal algorithm is not effective in the removal of water surface image highlights,this paper proposes a highlight detection based on feature fusion and an improved Criminisi highlight compensation algorithm.Firstly,the causes of highlight water surface formation are analyzed.By collecting the water surface highlight image dataset,the highlight image pixel color features as well as the image location features are fused to train the FCN(Full Convolutional Network)network model to detect the image highlight areas and to remove the highlights from the sky area.For the detected water highlight areas,the water highlight compensation is performed using the improved Criminisi-based algorithm.Then,weak target detection and background rejection are performed for the water surface image after water surface highlight compensation.Through comparative analysis,the results of this paper after highlight compensation greatly improve the recognition rate of weak targets in high-light water surface environment.