Research On Key Techniques Of Image Processing Of UAVs For Vegetation Recognition

The traditional remote sensing vegetation image is mainly derived from aerial photography and satellite remote sensing.The image acquisition cost is high,the period is long,which is influenced greatly by the weather.And the acquired image data is lack of real-time because of the long period of remote sensing data acquisition.The visual interpretation of remote sensing image is still the most trusted method by many researchers,but the visual interpretation of remote sensing image requires the interpreter to have professional knowledge background and rich practical experience,and with the development of modern remote sensing technology,the acquired remote sensing data is growing rapidly,so it can be interpreted from such massive remote sensing data,not only the labor intensity is huge,the interpretation period is long,and the interpretation experience and familiarity degree to the interpreted region etc,all have an important impact on the accuracy of interpretation.In recent years,with the development of UAV low-altitude remote sensing technology,vegetation classification and recognition based on unmanned aerial vehicles(UAV)image provide an opportunity to solve the above problems.In this dissertation,the image processing system is used as the technology platform for vegetation classification and recognition.The characteristics and shortcomings of the existing civil unmanned aerial image processing system are analyzed and summarized.In the dissertation UAV image vegetation recognition is the goal,so this dissertation focuses on denoising,segmentation,splicing and vegetation recognition,which has a great impact on UAV vegetation classification.The main achievements and innovations of this dissertation are as follows:This dissertation summarizes the current research and development status of UAV image recognition at home and abroad.Based on the current situation of existing UAV image processing,the denoising,segmentation,splicing and recognition of UAV image vegetation recognition are carried out.The classical algorithms of UAV image denoising are studied and summarized.The principles,causes and types of noise problems in unmanned aerial vehicles(UAV)images are systematically studied and proposed an improved multi-objective particle swarm optimization algorithm for image denoising,the simulation results show that the noise of the UAV image is reduced obviously.Systematically research and summarize current image segmentation of the various types of classic and major algorithms.The image segmentation algorithm based on edge detection and the image segmentation algorithm based on threshold are studied emphatically.For the UAV color image edge detection,In this dissertation,a quaternion-based color image edge detection algorithm is proposed,which is a new approach for UAV color image processing.The color image segmentation algorithm based on the improved cuckoo algorithm trallis entropy is proposed for unmanned aerial vehicles(UAV)color images.For image vegetation recognition of unmanned aerial vehicles,a good segmentation algorithm is extremely important,because only excellent algorithm can extract the target object accurately,reduce unnecessary object interference,and provide important guarantee for the following vegetation accurate recognition.In order to improve the accuracy of vegetation identification,the quaternion is introduced into the color image of the unmanned aerial vehicle(UAV),the pixels of the color UAV image are represented by quaternions,so that in the image processing,a quaternion of a color pixel as a whole,significantly reduce the traditional image processing defects which each color channel is processed separately.In the existing artificial bee colony algorithm,employment bees,follow the bees,scout bees are using the same search formula,which leads to the slow convergence and low efficiency of the artificial bee colony algorithm,in order to improve these deficiencies,the corresponding search formula is given to the employment bees and the following bees respectively,then an improved artificial bee colony is proposed,the improved artificial bee colony algorithm is applied to image edge detection of UAV color image expressed by quaternion.Experiments show that achieved good results.In order to further improve the quality and speed of image segmentation,this dissertation studies image segmentation from the perspective of image threshold segmentation.This dissertation improves the cuckoo algorithm which is widely used,first the position of the parasitoid nest was changed,and the non-linear decreasing function was constructed respectively for the step factor and the finding probability in the cuckoo algorithm.On this basis,a color image segmentation algorithm based on improved cuckoo algorithm trallis entropy is proposed.Experimental results show that the quality and speed of color image threshold segmentation are obviously improved.The whole process of UAV image mosaic and various algorithms of unmanned aerial image mosaic are systematically studied.In order to improve the quality and speed of stitching,an important algorithm of image stitching,SIFT algorithm,is improved.Experimental results show that the feature points detected by the SIFT algorithm contain more edge response points.Therefore,this study uses an improved Sobel edge detection algorithm to extract the edge feature points and neighborhood points.Then the feature points extracted by SIFT algorithm are compared and the edge response points are removed.Based on this,a feature extraction method based on improved SIFT algorithm is proposed.The experimental results show that the feature extraction speed and efficiency are obviously improved.Based on the excellent performance of CNN in image recognition,the classic convolution neural network is introduced into the vegetation classification and identification,and the neural network is improved.At the same time,in order to improve the accuracy and efficiency of vegetation identification,the dissertation first prepares the CNN by using the K-means vegetation feature learning algorithm,and then uses the modified CNN to carry on the vegetation recognition.An improved CNN-based vegetation identification method is proposed.The experimental results show that the accuracy and speed of vegetation identification can be improved effectively.In this dissertation,the method,content and steps of UAV vegetation recognition simulation experiment are systematically designed and programmed according to the current development situation and research results.According to the pre-determined process,.in the MATLAB environment,realize the automatic simulation experiment of UAV vegetation recognition.