The gray level values of target pixels of infrared images processed by the SUSAN noise filtering algorithm and the background filtering algorithm are calculated,assuming that the target distance is given in image acqu...The gray level values of target pixels of infrared images processed by the SUSAN noise filtering algorithm and the background filtering algorithm are calculated,assuming that the target distance is given in image acquisition.The equivalent blackbody temperature is obtained by the temperature calibration model that represents the quantitative relation,at a certain integration time,between the gray level value and the equivalent blackbody temperature.The temperature calibration model has been acquired by the curve fitting with the experimental data of blackbody temperature calibration using Origin 6.0.Then the target radiance can be determined by the radiant flux equivalence model built on condition that the target radiant flux received by the infrared detector is equivalent to that of the blackbody placed in front of the infrared detector at a close distance.This measuring method is based on the temperature calibration parameters and real image information,so it can avoid subjectivity of the traditional theory modeling methods.展开更多
In order to apply Satellite Remote Sensing (RS) to mining areas, some key issues should be solved. Based on an introduction to relative studying background, related key issues are proposed and analyzed oriented to the...In order to apply Satellite Remote Sensing (RS) to mining areas, some key issues should be solved. Based on an introduction to relative studying background, related key issues are proposed and analyzed oriented to the development of RS information science and demands of mining areas. Band selection and combination optimization of Landsat TM is discussed firstly, and it proved that the combination of Band 3, Band 4 and Band 5 has the largest information amount in all three-band combination schemes by both N-dimensional entropy method and Genetic Algorithm (GA). After that the filtering of Radarsat image is discussed. Different filtering methods are experimented and compared, and adaptive methods are more efficient than others. Finally the classification of satellite RS image is studied, and some new methods including classification by improved BPNN(Back Propagation Neural Network) and classification based on GIS and knowledge are proposed.展开更多
We propose an algorithm of embedding ensemble tracking in a stochastic framework to achieve robust tracking performance under partial occlusion,illumination changes,and abrupt motion.It operates on likelihood images g...We propose an algorithm of embedding ensemble tracking in a stochastic framework to achieve robust tracking performance under partial occlusion,illumination changes,and abrupt motion.It operates on likelihood images generated by the ensemble method,and combines mean shift and particle filtering in a principled way,where a better proposal distribution is de-signed by first propagating particles via a motion model,and then running mean shift to move towards their local peaks in the likelihood image.An observation model in the particle filter incorporates global and local information within a region,and an adaptive motion model is adopted to depict the evolution of the object state.The algorithm needs fewer particles to manage the tracking task compared with the general particle filter,and recaptures the object quickly after occlusion occurs.Experiments on two image sequences demonstrate the effectiveness and robustness of the proposed algorithm.展开更多
基金Supported by Innovation Project of Chinese Academy of Sciences(Grant No.C04708Z)
文摘The gray level values of target pixels of infrared images processed by the SUSAN noise filtering algorithm and the background filtering algorithm are calculated,assuming that the target distance is given in image acquisition.The equivalent blackbody temperature is obtained by the temperature calibration model that represents the quantitative relation,at a certain integration time,between the gray level value and the equivalent blackbody temperature.The temperature calibration model has been acquired by the curve fitting with the experimental data of blackbody temperature calibration using Origin 6.0.Then the target radiance can be determined by the radiant flux equivalence model built on condition that the target radiant flux received by the infrared detector is equivalent to that of the blackbody placed in front of the infrared detector at a close distance.This measuring method is based on the temperature calibration parameters and real image information,so it can avoid subjectivity of the traditional theory modeling methods.
基金Under the auspices of the Research Foundation of Doctoral Point of China(No.RFDP20010290006).
文摘In order to apply Satellite Remote Sensing (RS) to mining areas, some key issues should be solved. Based on an introduction to relative studying background, related key issues are proposed and analyzed oriented to the development of RS information science and demands of mining areas. Band selection and combination optimization of Landsat TM is discussed firstly, and it proved that the combination of Band 3, Band 4 and Band 5 has the largest information amount in all three-band combination schemes by both N-dimensional entropy method and Genetic Algorithm (GA). After that the filtering of Radarsat image is discussed. Different filtering methods are experimented and compared, and adaptive methods are more efficient than others. Finally the classification of satellite RS image is studied, and some new methods including classification by improved BPNN(Back Propagation Neural Network) and classification based on GIS and knowledge are proposed.
基金Project(No.2006AA10Z204)supported by the National High-Tech Research and Development Program(863) of China
文摘We propose an algorithm of embedding ensemble tracking in a stochastic framework to achieve robust tracking performance under partial occlusion,illumination changes,and abrupt motion.It operates on likelihood images generated by the ensemble method,and combines mean shift and particle filtering in a principled way,where a better proposal distribution is de-signed by first propagating particles via a motion model,and then running mean shift to move towards their local peaks in the likelihood image.An observation model in the particle filter incorporates global and local information within a region,and an adaptive motion model is adopted to depict the evolution of the object state.The algorithm needs fewer particles to manage the tracking task compared with the general particle filter,and recaptures the object quickly after occlusion occurs.Experiments on two image sequences demonstrate the effectiveness and robustness of the proposed algorithm.
