With the progress and development of remote sensing science, the technology of acquiring high precision satellite video is becoming more and more sophisticated. Along with the acquisition of high-precision real-time surface data, satellite video is applied to the detection and observation of target objects, and provides intuitive and new research ideas, which is also of great significance for observation, judgment, and continuous tracking of abnormal events. For the algorithm and research means of satellite video anomaly events, scholars want to apply anomaly detection of surveillance video to satellite video. However, due to the influence of factors such as positioning accuracy and recognition rate, the application level of anomaly detection of satellite video is not high at the present stage. This paper explores the application of the algorithm based on sparse reconstruction in abnormal event detection and completes the abnormal event detection of satellite video by comparing surveillance videos. Compared with other existing anomaly detection algorithms, this algorithm innovatively uses satellite video as training and detection data to improve the detection ability of the algorithm in global events and has good recognition ability for crowded sections and all kinds of targets. This paper also has advantages in processing method and computational efficiency. Sparse coding is used to construct the reconstruction error of abnormal judgment, and combinatorial learning method is introduced to improve detection efficiency.

The research ideas in this paper are divided into five stages: 1.  Compare and determine the algorithms; 2. Feature representation; 3. Dictionary optimal selection; 4. Judge the abnormality; 5. Video data application verification. Based on the inherent structure of video, an efficient sparse combination learning framework was proposed to represent features by multi-scale optical flow histogram and find the optimal representation combination of data set using normal dictionary. The feasibility of the algorithm is experimentally verified on surveillance video and then applied to satellite video to get the final result. The abnormal event detection ability of the system can well meet the needs of processing satellite video observation targets and has good performance in accuracy and efficiency.