Study Of Typical Targets Recognition Key Technology And System For Optical Remote Sensing Image

Remote sensing images used for military reconnaissance not only play a big role in operational command assistant decision, and can guide weapon systems precision attack. At present, target interpretation and intelligence activities of our army are in the transition from visual, qualitative and empirical interpretation to human-computer interaction and automatic interpretation. The poor systematic research on characteristicses of military targets largely influences discovery, recognition and classification of them from high-resolution remote sensing images with large size and complex background. It reduces the utilization and timeliness of these images. Traditional methods are hard to meet our army actual requirement on reconnaissance intelligence. According to the analysis of existed ATR techniques, the features and recognition methods of typical targets such as airports and bridges in high resolution optical satellite images were deeply investigated, and a fast recognition system had been set up.Focusing on the extraction of straight line features from large targets, this paper designed an improved Hough transform approach integration of resembleing lines extraction. This approach was applied to airport recognition. Extraction of resembleing lines was pre-processing of Hough transform, which preserved edge property and detected feature points required for transform. For each extracted resembleing line, Hough transform was used respectively. Votes were cast using improved mapping scheme, which was a few feature points in image space corresponded to one point in parameter space. The straight line determined by the most points in resembleing line was seeked. New voting process aimed to avoid loss of spatial information as well as to reduce the computational complexity. The runway was recognized by detection and judgement of parallel lines.Aimed at fast detection of bridges over waters, an algorithm based on waters extraction with block was developed. In order to rasie efficiency, image was split into blocks. By using general local Fourier transform (GLFT), every block was applied to segment waters. GLFT texture gave a fine performance in computation, discrimination and segmentation. Block calculation made the waters contour regular and improved the efficiency of the whole algorithm. Then, false waters were excluded by judging the contour shape with chain code analysis. According to gradient information, suspected bridge areas were detected. Lastly, three stable knowledges were used to estimate parameters to verify bridge further. During the process of bridge recognition, priori knowledges had been used all the time to speed up detecting and recognizing.On analyzing structure characteristic of inside harbor region, a method of harbor target recognition was proposed. See area was detected preliminarily through gradient analysis and gray similarity clustering to judge the existence of harbor. After rapid segmentation, see and land were divided and coastline was extracted in rough. Starting from sea area, image was scanned along the eight directions. According to the closure degree by land area, sea area was checked whether it was a suspected harbor region. The concept of U-shaped structure element was introduced to verify harbor region finally.Selective attention mechanism of human visual has been applied in the field of target recognition recently because of their excellent performance in object perception. This paper presented a detection approach of small tactical targets based on selective visual attention model. Discussion on previous knowledge, a selective visual attention hypothesis was improved according to feature integration theory. In visual attention, using data-driven strategy, the region with significant characteristics was searched from image and regarded as the focus of attention (FOA). While in region of interest (ROI) analysis, recognition of oil depots, aircrafts and missile positions were leaded by prior-knowledge. In paper, salience map fusion module receives feedback from inhibition of return and location for enhanced, then target detection cycled. Such a process is more in keeping with human visual attention.According to military applications, a typical target recognition system was established to achieve efficient management of remote sensing images, analysis and extraction of the military targets, creation of target feature models and the target database. System integrated software and hardware framework. It also offered fast and convenient way of human-computer interaction for assisting interpreters. The experimental analysis showed that the methods for automatically recognizing some strategic and tactical targets, such as airports, bridges, harbors, oil depots, aircrafts, and missile positions, in high resolution remote sensing images with complicated backgrounds were fast and effective.