Target Multi-dimensional Association Mining And Deep Learning Method Based On Massive Infrared Video

Infrared thermal imaging technology uses the infrared electromagnetic wave band actively emitted by the object to perceive the difference between different targets and the background according to the difference of the surface temperature of the object for imaging.This imaging feature makes the infrared target detection system more adaptable to complex environments,and is suitable for long-distance observation in the military and aerospace fields.Massive infrared video data occupies a large proportion of the data collected by space-based satellite infrared ground detectors.With the full-time shooting and recording of the infrared ground detector video monitoring equipment,massive amounts of unstructured data have been generated.Research on mining massive infrared video data is of great help to storage of unstructured information and early warning analysis of infrared video targets.This paper analyzes the characteristics of massive infrared data,and then specifically classifies massive video processing as massive infrared data cleaning,target extraction based on video target detection,target multi-dimensional association mining,and the construction of a big data platform based on massive video processing.,And then carried out theoretical analysis,method research and simulation verification.The main research contents of this paper are as follows:(1)Aiming at the characteristics of a large number of redundant frames and sudden changes in the massive infrared video data,a data cleaning algorithm based on the massive infrared video is proposed.Firstly,the improved method of cleaning video blank and redundant frames is studied,and the detection is carried out on infrared surveillance video data set and self-made simulation data set.It is proved that the data cleaning method in this paper effectively reduces the redundancy of video.(2)Aiming at the slow detection speed and low accuracy of traditional detection methods in massive infrared video detection,by analyzing the advantages and disadvantages of the video target detection network model,research and design a key frame extraction algorithm based on 3dcnn fusion features,and propose Improved video target detection algorithm,after comparison experiments,this method can effectively improve the accuracy of target detection in infrared video,m AP can reach 0.714.(3)Based on the above process in the article,the target in the infrared mass video is extracted by multi-dimensional manual features and depth features of the target,and through analysis,the improved random forest method based on multi-dimensional features is used to perform multi-dimensional video data.Association mining.(4)In view of the large amount of infrared video data,it is difficult to store and the computing speed cannot meet the real-time needs,build a distributed storage system,use HDFS for storage management,and design a distributed data processing method based on the Spark platform,and design infrared In the process of video feature extraction,data parallelization and training model parallelization are realized to realize the distributed deployment of the algorithm in this paper;and a QT-based algorithm display platform is designed to visually display the algorithm in this paper.