| In the development of human aerospace industry,the space-based infrared detection system and infrared early warning system constructed by researchers have excellent night adaptability.These systems are widely used in human production and life.The detection of infrared targets requires high field of view,high detection rate,high sensitivity and long detection distance.Due to the lack of color information in infrared images,the texture features are not obvious,the contrast is low,and the noise is high,infrared target detection is facing certain difficulties.The false alarm source scene in the background similar to the target radiation intensity will lead to an increase in the false alarm rate.In thesis,the false alarm source is processed as a priori information from the infrared target imaging background.Firstly,the background area that may produce high radiation is detected and obtain the information.Then,the information is used to assist infrared target detection and reduce the false alarm rate.A proposal to enhance the efficacy of an infrared target detection system is made in thesis,with more comprehensive infrared target detection technique.(1)In thesis,the radiation,texture and spectral domain features of typical infrared false alarm sources are analyzed and extracted.The visual features of the false alarm source and the target are quantitatively and qualitatively analyzed,which lays a foundation for the research of the detection algorithm.Select the appropriate features for modeling to achieve the detection of false alarm sources.This helps to improve the expression ability of false alarm source features and reduce the false alarm rate.(2)Aiming at the problem of inaccurate segmentation of the fuzzy edge of the transition between the typical false alarm source and the background,this paper proposes a false alarm source detection algorithm based on texture feature and edge feature modeling by studying the extraction principle of edge and texture feature,and realizes the more accurate detection of the false alarm source edge.Contraposing to the problem that the detection algorithm based on single feature is lack of effect on the false alarm source with specific size,this paper proposes a false alarm source detection algorithm based on orthogonal direction feature fusion based on frequency domain saliency through the idea of multi-feature fusion.The algorithm has low false alarm rate and average absolute error in different types of false alarm source detection,and ensures a high intersection ratio.(3)According to the missed detection of dim cirrus false alarm sources that are difficult to distinguish based on the feature model,this paper proposes a false alarm source detection method based on parallel iterative background dictionary learning by combining the fractal characteristics of false alarm sources and the low rank of background.The algorithm uses the principle of compressed sensing to accelerate the convergence speed and expression ability of dictionary learning,and realizes the accurate detection of complex cirrus cloud false alarm source by means of false alarm source fractal dictionary and background dictionary.Aiming at the complex causes of some false alarm sources and the lack of data samples,thesis proposes a false alarm source detection algorithm based on low-rank background dictionary constrained autoencoder.The algorithm uses the depth features of infrared images to achieve false alarm source detection of small weak samples and has excellent generalization ability.The above detection algorithm based on false alarm source and background design features has excellent performance under ROC curve,which provides a new solution to the detection of false alarm source for earth observation,and improves the response ability of space infrared satellite to earth detection system and the accuracy of small target detection. |
