Wide Area Remote Sensing Image On Orbit Target Extraction and Identification Method

Aiming at the characteristics of target sparsity of remote sensing satellite image, the method for extracting and identifying the target of wide area remote sensing image on orbit is proposed, which is mainly realized by the technique of morphological matching & deep learning. The visual enhancement technology is used to extract the suspected target slices quickly and reduce the amount of processed data greatly, and the deep learning method is adopted to reduce the extraction false alarm rate greatly and realize the recognition and fineness of the target. In the system function and performance verification test, real-time and correct rate indicators were tested on 487 slices of GF-2 satellites with more than 64 scenes. The system can extract and identify 60 targets that the length is more than 100 meters per second. The extraction method based on traditional morphological matching targets has an efficiency of no more than 50%, and the recall rate reaches over 90%. The target classification and identification method based on deep learning increases the precision to over 95%. The system has high integration, excellent real-time performance and expansible, which can meet the real-time processing requirements of current remote sensing satellite payload.