A Multi-spectral Image Saliency Detection System Based On UAV Platform

In recent years,UAV has developed rapidly in many fields because of its small size,light weight,and flexible maneuver.The unmanned aerial vehicle(UAV)on the vehicle can make the UAV quickly and flexibly handle the captured image and mark the area and location of the significant target.The paper designed and implemented a multi spectral image based on saliency detection system of UAV platform,the system uses the UAV equipped with infrared and visible light cameras by image acquisition,image transmission subsystem back to the ground,complete the image saliency detection in ground test subsystem.The ground saliency detection subsystem is an important part of the system.It mainly includes the preprocessing module composed of infrared image enhancement and super pixel segmentation,the saliency detection module formed by static and dynamic salient detection methods,and the salient object segmentation module.In the part of static saliency detection,this paper proposes a model of saliency detection based on two level random walk.It absorbs Markov chain by using partially absorbing random walk probability to get significant detection results.The experiment shows that the method obtains better performance curve in the public data set,and has good visual effect in the actual scene.The dynamic saliency detection algorithm based on motion information obtains the motion saliency image by calculating the contrast of the motion magnitude by the light flow histogram.Experiments show that this method can locate accurately in standard datasets and practical application scenarios,conforms to the significant goals of subjective evaluation of human eyes,and effectively provides motion significant information for the system.In order to improve the operation efficiency of the system,we use GPU based CUDA architecture to accelerate and optimize some modules in the system.After testing,the running speed of optimized parts is significantly improved,which ensures the practicality and effectiveness of the system.Based on the above research,we set up a multispectral image saliency detection system based on UAV platform,and verify the workflow and detection function of the system.The system has high running efficiency,significant detection results and complete functional modules.It can provide more accurate and significant target contour and motion information,improve the efficiency of UAV target detection,and has strong practicability and prospect.