Research On The Vegetation Identification Method Based On UAV Image Acquisition

Vegetation interpretation and classification is basic works in large area forestrysurvey or ecological monitoring used remote sensing. Vegetation remote sensingimages mainly generate from satellite remote sensing images and aerial photographyimages, which was following many problems such as long period of images getting,poor real-time performance, sensitive affected by cloudiness and expensive cost etc.On the one hand, at present the most reliable method of remote sensing interpretationstill rely on visual interpretation, the people who executing visual interpretation weredemanded to possess wealth of geoscience background knowledge and interpretationexperience of long-term accumulation. This method has a lot of subjectivity andfollowing many problems such as strength work intensity of interpretation, longperiod of data acquisition, interpret accuracy was restrict by factors as visualinterpreter’s experience and familiarity in region of interpretation. On the other hand,how to get feature interpret data economical and quickly from massive remote sensingdata is a problem to professional staff while Earth observation technologies rapidlyevolving. In recent years, remote sensing system based on UAV platform provides anopportunity to solve the problem with the gradually opening up of China’s lowairspace and the progress of UAV technology. Therefore, to explore the vegetationimage interpretation method which was suitable for the characteristics of the UAV is anecessary complement to the traditional means of remote sensing.UAV remote sensing system as a technology platform to obtain image data ofvegetation was selected in thesis. Function features and limitations of commercialize image processing software in UAV was analyze and summarized, against the problemof vegetation interpretation and classification with images collected from UAV,engineering ideas were used, systematic study was carried out on aspects about thebasic principle, the basic methods and techniques to achieve, process framework onUAV image recognition technology was designed, the computer interpreting system inUAV remote sensing images were improved. Viable technologies and solutions wereproposed and simulation was executed to solve many key technologies and problemsabout vegetation type identification method from UAV images.The main research and innovation as follows:1st. The whole package of UAV image with "get-processing-interpret"technical processes was researched and summarized. Via analyzed and researchedthe existing commercial UAV image processing software flow, found that most imageprocessing also focused on image correction, image stitching, and other applicationsto generate3D simulation map, the applications of image recognition andclassification is not concerned, limits the utilization of image resources in UAV toenhance. Theoretical system for the identification and classification of UAV imagewas complemented and extended, many effective explorations and experimentationswere completed around the mountain vegetation remote sensing recognition methodsand techniques.2nd. The main UAV images stitching algorithm was researched and summarized,SIFT algorithm has been improved, UAV images quickly stitching was realized.Surrounding achieve UAV massive image sequence rapid synthesis problem, forexisting issues in UAV image stitching algorithm such as long running time, largefootprint and slow stitching, proposed that improved SIFT algorithm to realize UAVimages quickly stitching. The number of matching feature points were reduced bymatching algorithm based on color images, using image color information, and increased extreme determination pixels.3rd. HSV-T features model suitable for UAV image recognition was proposedbased on computer vision and principles of human cognitive psychology, and nearestneighbor method was chosen to achieve the image feature matching. The traditionalvisual interpretation mechanism was simulated from the features as image color,texture, shape, position relationship, etc., two prominent visual features about colorand texture was selected as an observation point to design the HSV (color)-T (Tamuratexture) feature model. The problem of the existing image classification using singlefeature identification and classification resulting in the low recognition rate wassolved, the accuracy of image recognition was effectively improved, and effectivenessof HSV-T model was verified by experiments.4th. The method flow of vegetation types recognition suitable for UAV imageswere researched and improved, UAV images rough classification of vegetation wasinitial realized. A set of process frameworks on vegetation automatic identificationand classification with UAV images were proposed, the validity and reliability of theprocess flow was verified through experiments according to the computer imagerecognition processes and combined with the characteristics of the UAV images.5th. Content, steps and methods of the simulation experiments in vegetationrecognition with UAV image were practiced and summarized. According to apredetermined procedure for pretreatment about UAV image correction, stitching,feature extraction and feature training, and nearest neighbor pattern recognition wasused for feature matching, simulation experiments with good results of the automaticidentification of vegetation types of UAV images by MATLAB environment wasachieved.