Fusion Classfication Algorithm Based On Infrared With Visible Image And Infrared Target Detection

In recent years,image recognition technology has been widely used in scene identification,target detection,medical diagnosis,ship detection and other fields.With the development of pattern recognition algorithms,the multi-feature fusion classification mechanism based on visible image has become a research hotspot.Although this method can improve the classification effect to a certain extent,but it does not change the problem that the characteristics are sensitive to environmental changes due to the limitation of the visible sensor.Against this problem which using a single sensor to obtain data lead to an inaccurate recognition results,This paper mainly studies the image feature fusion of the same image under infrared and visible.At the same time,this paper also studies the detection of weak targets obtained by infrared sensors.The main contents are as follows:Firstly,this article introduce the pattern recognition and fusion classification from the aspects of research background and application fields,as well as the domestic and abroad development status,and explain the significance of the research of fusion classification algorithm.Analyze the imaging characteristics,the advantages,disadvantages of infrared images and visible images,respectively,which provided a feasible condition for the feature fusion.Analyze the difficulties in image fusion,and find a solution to this problem.Secondly,in scene detection and target(ship)detection,the visible image is easily affected by external factors such as environment and light,and it sometimes fails to completely represent the information of the object,which cause defects of misclassification and misclassification.Aiming at the shortcomings of visible image data classification,a fusion classification strategy based on infrared images and visible images is proposed.First,using DSIFT-CLM to extract the infrared and visible images' feature.Then map these features are to a high-dimensional space through the kernel functions,And through the composite kernel to achieve the feature-level fusion.This method combines the respective advantages of infrared and visible image to make up for the lack of visible classification alone.The fusion features are more comprehensive and complete.Finally,this strategy adopts the method of support vector machine to classification.After the test comparison,the classification effect of this method is better than that of only using visible.Thirdly,aiming at infrared image has low contrast and the weak objects are often submerged in a complex background that caused the target to be confused with the background or unable to detect the target during the detection process.A detection method combining two-dimensional empirical mode decomposition based on image(IEMD)and robust principal component analysis(RPCA)was proposed for infrared weak targets.The background suppression of infrared small target images is achieved through IEMD decomposition,and extract the useful IMFs.Then use RPCA method to reconstruct the background image and target image.This method not only can detect the small target position under complex and heavy clutter background,but also can reduce the false alarm by using the IEMD decomposition.