Research On Key Techniques Of Camouflage Assessment For Complex Background

With the development of technology,camouflage detection technology is becoming more and more mature and has been fully automated.In terms of detection time,real-time performance is gradually realized.The rapid development of camouflage detection indicates that the research on camouflage assessment technology is an urgent task.Through the research and analysis of the characteristics of the existing camouflage technology,the objective and reliable camouflage assessment model is developed,so as to replace the manual subjective evaluation,which is the main research goal.In the camouflage assessment task,the sensitivity of military target leads to the scarcity of data set and the difficulty of research.Therefore,simple and efficient acquisition of reliable data sets has become the primary task of research.This paper analyzes the experimental process of the existing camouflage assessment research and summarizes the defects of the existing experimental methods.We found that simulating a camouflage environment in the lab ignores some of the problems that exist on the real battlefield.Ignoring these questions can lead to inaccurate research results.At the same time,in order to reduce the cost of data acquisition,this paper adopts tank miniature model and armored vehicle miniature model for data acquisition.Relatively reliable supporting experimental data sets Green Tank(GT)and Green Car(GC)are obtained.The current camouflage assessment algorithm evaluates the similarity of the target and its eight neighboring areas before the evaluation,and finally calculates the average value as the camouflage assessment result.This kind of preprocessing method which takes eight areas around the target as background area has a certain scientific nature.In addition,this method takes into account the background that is too far away from the target,which has limited influence on the camouflage effect of the target area,and reduces the computational complexity and time.However,the problem is that the segmentation result of this method is a rectangular region,which is inconsistent with the irregular region perceived by human eyes.Considering the irregularity of the feature focus and attention perception area evaluated by the human eye,this paper redivides the area around the target according to the color and the edge of the original image based on the superpixel and color clustering,and obtains the irregular area that accords with the perception result of the human eye.The existing models of camouflage assessment are based on the rectangular region,and the perceptual and attention processes of human eyes are less considered.In this paper,the traditional features are sorted out and studied,and the existing traditional features are improved based on the nonlinearity of human perception results.In this paper,the gray cooccurrence matrix is weighted according to the results of human eye evaluation during the experiment,and the features that can evaluate the texture-brightness similarity are obtained.Based on the three channels of HSV space,the recalculation is carried out to weaken the influence of brightness features and obtain the color features suitable for the camouflage assessment task.In this paper,when the evaluation results of each area are integrated,the influence of area area on human eye attention is considered,and the area ratio of use is weighted for each area to obtain the final evaluation results.In order to verify the effectiveness of the proposed method,experiments were carried out on GT and GC data sets collected in the field and compared with 14 other well-known algorithms.Experimental results show that the proposed method achieves the best results.