Research On Maritime Dim Target Detection Methodology Based On Infrared Images

Target detection based on infrared image is a key technology of maritime target searching system and has been widely used in maritime target monitoring and searching in recent years.Under marine environmental background,infrared imaging quality may be affected by the thermal radiation of sea surface,the sky radiation reflected from the sea surface and the atmospheric radiation from the sea surface to the infrared imager.Due to remote imaging distance,infrared maritime targets are usually small and lack of distinctive shape and texture features.In addition,the complex environment may be caused by strong sea waves,fog and strong sun glitters,which usually leads to so low signal-to-clutter ratio for infrared image that the real targets are submerged in complicated background.All these factors increase the difficulty of infrared dim maritime target detection.On the basis of the analysising the characteristics of infrared image captured under complicated marine conditions,we deeply study the infrared dim maritime target detection technology with strong adaptability and high detection rate,which is of great significance to effectively improve the success for maritime distress target surch.The research content mainly includes the following four aspects:(1)The characteristics of dim targets in infrared images captured under common marine unvironment is not obvious,and it is difficult to effectively improve the target saliency only by any single feature.To solve this problem,the local heterogeneity of targets,the non-local self-correlation of the background and the sparsity of targets are analyzed in this thesis,and the local heterogeneity calculation method based on cross-window standard deviation is studied.Meanwhile,low-rank representation is used to predict and remove the infrared maritime image background.Then the saliency of targets is further improved by integrating the two candidate target extraction and saliency enhancement results.By synthesizing the abovementhioned research,integrated target saliency measure is proposed,which combines local features with non-local features.(2)For strong ocean wave environment,the radiation intensity of strong wave in infrared image may be similar to that of small target due to remote imaging distance,which will lead to low signal-to-clutter ratio and make it difficult to distinguish the target from strong wave.However,there is distinct dffernence between targets and wave in the gradient distribution and directional chatacteristic changes.Therefore,in this thesis,the gradient vector field characterization method based on infrared image is proposed.By the comprehensive analysis of many kinds of morphological properties,like the pixel gradient direction angle distribution,the gradient modulus distribution,gradient modulus horizontal local dissimilarity,and so on,the gradient vector distribution measure and gradient modulus horizontal local difference measure are further studied.The integration of the two models can effectively improve the saliency of small target in strong ocean wave background so as to enhance the accuracy of small target detection.(3)In order to further improve the detection accuracy for infrared dim targets in strong ocean wave environment,the infrared dim maritime target detection method based on weighted multi-directional gradient is further studied.The extraction of candidate targets,multi-directional feature description and saliency reconstruction are mainly researched.The average cumulative multidirectional gradient measure and directional difference measure are further studied,and their weighted fusion achieves target enhancement and clutter suppression for infrared maritime images with strong ocean wave background.This method realizes better robustness and higher detection accuracy.(4)Since the target has trajectory consistency in continuous frames,in order to fully mine the spatial information and temporal correlation of the target,the strategy of single frame detection followed by multi-frame decision is adopted in this thesis,and an anti-jitter spatio-temporal trajectory consistency target detection method is further studied.There may be bright or dark targets in infrared maritime images,which is influenced by the target characteristics and environmental temperature.Aiming at this problem,adaptive local gradient variation descriptor is adopted in this thesis,which uses adaptive local energy factor and local gradient module variation to enhance the saliency of bright and dark targets so as to achieve suspected target extraction.In multi-frame decision stage,in order to eliminate the interference of imager motion and vibration for the target encoding,interframe displacement compensation is introduced to rectify the location of targets in spatio-temporal domain.And spatio-temporal trajectory descriptor is designed to achieve suspected encoding and target trajectory consistency measure,which further confirms true targets,eliminates the false alarm and estimates the missed targets.In this thesis,infrared maritime images with a variety of complex scenarios are captured in different weather conditions by maritime target searching system developed by our lab.And the representative infrared maritime image sequences are utilized to verify the performance of the method proposed in this thesis.Extensive experimental results demonstrate that the proposed maritime dim target detection method based on infrared image performs a strong adaptability and high detection rate,which has high engineering application value and provide the key technical support for maritime distress target searching system.