Research On Detection Technology Of Infrared Target Based On Convolution Neural Network

Target detection is a popular direction in machine learning.As a branch of image understanding and computer vision,it is a necessary prerequisite for a large number of visual tasks.Target detection is widely used in many fields,such as road traffic vehicles and signs,detection in security system,target recognition in military field.For different application scenarios,the core idea of target detection is to solve the accurate positioning of objects in the background.For the target in the infrared image,it is a difficult problem to detect the target and analyze its object trajectory.Because of the weak feature and the lack of color of the target itself,the traditional algorithm still has some limitations to solve this problem.In recent years,the target detection algorithm based on deep learning has developed rapidly.Because it can well fit the nonlinear problem,the target detection algorithm based on deep learning has been widely used in the field of natural light image,but it is not widely used in the field of infrared image.In this paper,based on the target detection algorithm based on the deep convolution neural network,two algorithms based on the convolution neural network are designed to improve the accuracy of the model and reduce the false alarm rate.The research content of this paper is as follows:(1)Data set of infrared target and its trajectory.For deep learning and big data processing,data sets,especially big data sets,are particularly important.In this paper,the small target detection data set of infrared image is constructed by the way of actual measurement and simulation expansion,and the capacity of the data set is expanded by the way of increasing noise and data expansion,such as image reversal,brightness adjustment,rotation and interception,so as to improve the robustness of the data set.The data set has laid a solid foundation for the subsequent research of infrared target and infrared target trajectory detection algorithm Basics.(2)The construction of enhanced infrared target network based on improved SSD.A small-scale infrared target detection method is designed to improve the SSD network.The DetNet backbone network is introduced and the cell structure is improved.At the same time,the pyramid feature fusion structure is used in the feature extraction layer of SSD network,so that the network can carry out multi-scale feature fusion target detection.The accuracy of infrared small target detection is improved,and the test results of infrared UAV data set prove the effectiveness of the method.(3)The construction of infrared target trajectory detection network based on full convolution network.Using the excellent characteristics of the full convolution network,the appli-cation of different backbone networks in the full convolution network and the ap-plication of the deconvolution network layer in the field of image segmentation are studied.A method of infrared target track detection combining the full convolution network and the deconvolution network is designed.In the case of strong noise and trajectory breakpoint due to occlusion,our model is analyzed and proved.The test results of data set show the effectiveness of the method.