Overlap-prior Based Stitching Method For UAV Farmland Thermal Infrared Remote Sensing Image

Farmland thermal infrared images effectively reflect crop canopy temperature,and are widely used in crop drought stress,precision irrigation,crop growth,disease,lodging and yield monitoring.Therefore,obtaining high-quality thermal infrared panoramic image is a key step in the refinement of the research area.Researchers usually use UAV remote sensing technology and image mosaic technology to obtain the thermal infrared panoramic image of the research area.However,the thermal infrared remote sensing image is easily affected by wind and temperature drift,and the quality is poor,which leads to the difficulty of thermal infrared image stitching,which often leads to dislocation,distortion,or even failure.At the same time,due to the single feature of farmland scene,the task of stitching is more challenging.In order to solve the above problems,this paper proposes a simple and effective mosaic method of thermal infrared remote sensing image based on niswgsp image mosaic model and POS(position and orientation sytem)data recorded by UAV during flight.The main contents of this paper are as follows:(1)Making UAV farmland thermal infrared remote sensing image data set.At present,the test data sets for image mosaic model at home and abroad are all visible light data.In order to objectively evaluate the performance of the mosaic algorithm,based on the four research areas of Hetao Irrigation Area in Inner Mongolia,the thermal infrared remote sensing data of UAV farmland were taken in May,July and September of 2018 and 2019 respectively.The ground feature information mainly includes sunflower,corn,wheat and plastic mulched farmland.At the same time,in order to ensure the integrity and diversity of the data set,we set up different experimental environment for each study area,including sunny,cloudy,light rain three kinds of meteorological environment,as well as the morning,noon,afternoon three time periods to obtain thermal infrared data.In this paper,24 groups of UAV farmland thermal infrared remote sensing image mosaic data sets with different growth periods and different meteorological environments are constructed.(2)The model suitable for large-scale thermal infrared remote sensing image mosaic is selected.Due to the lack of research on thermal infrared remote sensing image mosaic,it is necessary to select the mosaic model suitable for thermal infrared image.In this paper,three classic image mosaic models,autostitch,APAP and niswgsp,are selected to test the mosaic effect of the three models on the constructed data set.Through the experiment,it is found that niswgsp image mosaic model is better than autostitch and APAP in mosaic efficiency,mosaic effect and model scalability.APAP will cause serious distortion problems when there are many stitched images.(3)The stitching model of thermal infrared remote sensing image based on the prior of overlap ratio.Firstly,according to the POS information and internal parameters of thermal infrared camera in the process of UAV remote sensing image acquisition,the overlapping ratio of adjacent remote sensing images is estimated,and multiple image matching pairs with high overlapping ratio and high image matching degree are constructed for each image.Secondly,in the niswgsp image mosaic model,the constructed image matching pairs are added to the local alignment term of the model to ensure that there are multiple image matching pairs constraints in different directions for each image and prevent the mosaic algorithm from converging to the local optimal solution.The quality of image mosaic is greatly improved,and the stitched results are better than the classical image stitching method and popular commercial software on the data set.