| Compared to other means,thermal infrared remote sensing possesses unique advantages.The wave band for the formation of thermal infrared images within the research of this dissertation is 8-11?m.Images within this wave band can reflect the type and operation status of typical targets,and therefore play special roles particularly in the reconnaissance of military targets,the surveillance over hot-spot areas,the revealing of camouflaged targets,etc.Compared to visible images,space-borne thermal infrared images are fuzzy in terms of details with low contrast and low signal-to-noise ratio,making the recognition of targets in such images more difficult.However,it is still feasible and quite necessary to recognize such typical targets as ports,airports,large ships,and oil tanks based on these images.Since ports(including oil tanks)and airports are fixed targets while ships are moving targets,and since these targets maintain certain representativeness in terms of such image features as shape and texture,this dissertation selects these typical targets.Since there are many problems left to be resolved as for the recognition of thermal infrared images in terms of strip noise removal,radiometric calibration and temperature retrieval,target modeling and image segmentation,feature extraction and selection,and target detection and recognition,etc.,based on images that have gone through radiometric correction processing,this dissertation analyzes the recognition features and features of the thermal infrared images of typical targets,conducts feature comparison based on the model for feature analysis in support of manual interpretation,extracts the typical features of target recognition in thermal infrared images,and researches ways to detect and recognize targets in thermal infrared images.The main contents and innovations in this research are as follows:1.The method of feature analysis and comparison is utilized to analyze the status and information of targets in thermal infrared images.The recognition features of the thermal infrared images of oil tanks,ships and runways are analyzed,and the status of oil tanks as well as the information of airport facilities are judged from the perspective of auxiliary recognition through the method of thermal infrared feature analysis and comparison.This method is effective when the environment does not exert much influence.2.The method based on evaluators is used to research the method of extracting and selecting thermal infrared image features,and the optimal combination of the features of such targets as ports,oil tanks and ships is thereby worked out.On the question of the high efficiency and universality of image segmentation before the extraction of geometric features from the images of ship samples,the Otsu's method is selected from the four typical threshold segmentation methods.In regard to such difficulties of the recognition of the thermal infrared images of ports as the complexity of the background,the large number of targets,and the inconspicuousness of marginal information,a method to extract and select features based on the method utilizing evaluators is proposed.With such features as texture and geometric features that are effective in recognizing thermal infrared images having been compared,and with the optimal combination of features having been selected through the use of evaluators,the optimal combination of the classification features of ports in thermal infrared images is worked out.Based on the aforementioned selected combination of the features of ports,in regard to the recognition features of oil tanks and ships,a method to extract and select features based on support vector machine classification is proposed,and the combination of the thermal infrared features of oil tanks and ships with the highest accuracy of classification and the corresponding model of classification are produced.The classification model of training samples is provided for subsequent research on sample learning methods for target detection.3.The stratified multi-threshold method on account of dimensions and the method of considering shape and low contrast are proposed from the perspective of learning without sample,and thus realizes the effective detection of ships and runways.In regard to the difficulty in the thermal infrared detection of ships when the target and the background pose relatively low contrast,a stratified multi-threshold detection method based on the selection of dimensions is proposed.It is possible to filter out similar regional targets with different template dimensions according to the dimension of the template and thus realize the accurate detection of ships.In regard to the inaccuracy of the thermal infrared detection of ships when the brightness temperature of the target is rather high,under the conditions that the target detection based on wavelet packets and higherorder statistics disregards the geometric features of the target,a method to detect targets that takes the features of the shape of the target into consideration is proposed.When there is much noise in the background and high contrast between the target and the background with fuzzy texture,the detection presents rather good effects.In regard to the incompleteness of the extraction of runways in thermal infrared images when there is low contrast between the target and the background,the intuitionistic fuzzy C-mean-value binding-domain growth method as the auxiliary detection method is proposed,which compared with the directly-used-region-growth method,can facilitate manual work in a rather accurate way to extract runways.4.From the perspective of sample learning methods,the methods of deformable template,block mass features detection,stratified multi-threshold detection and support vector machine classifiction are proposed,and thus,the detection problem of oil tanks and ships in the condition of low background contrast is solved.Regarding the difficulties of detection when oil tanks are small in size and pose low contrast to their background,combined with the concluded optimal combination of features,a detection method based on support vector machine classification is proposed,which is effective in detecting oil tanks with certain strip noise and inconspicuous marginal information.Regarding the problem as for the relatively high false alarm rate that occurs while detecting oil tanks through support vector machine classification,taking into consideration the round feature of oil tanks,a detection method that combines support vector machine and deformable template is proposed,which,is effective in the detection of quasi-circular oil tanks with certain strip noise and inconspicuous marginal information.Besides,regarding the problem that a large number of false alarm targets will be presented through the aforementioned detection method,a method detecting spot and block mass features is proposed.This method is combined with support vector machine sample learning classification and the method of deformable templates.It boasts higher accuracy of detection,and compared to the method that simply conducts sample classification,it lowers the false alarm rate and improves the detection efficiency.Regarding such problems as that the detection effects will be poor when there is little difference between off-shore ships as the targets and the background in terms of brightness temperature,and that the detection based on stratified multi-threshold method sometimes cannot be full with the probability that a large number of false alarm targets will exist,combined with the selected combination of the features of ships,a detection method that combines the stratified muti-threshold method with support vector machine sample learning classification is proposed,which,based on multi-dimensional multi-threshold detection,conducts detection through SVM method.This method can recognize ships in a more comprehensive way.Compared to the simple use of the method of sample classification,it lowers the false alarm rate and improves the detection efficiency. |
PDF Full Text Request