Development Of Globular Plants Count Software System Based On UAV Visible Light In Nursery Image

Flower nursery stock industry has a very important significance in the construction of garden cities,planting windbreaks and green belt.With the sustained development of economy and high attention to environmental issues in China in recent years,the demand for flowers and seedlings grow rapidly.However,the existing inventory statistics of nursery mainly relies on manual count,which has large labor costs and low efficiency,serious waste of human resources,and prone to statistical errors,also makes a serious waste of human resources,and prone to statistical errors.The purpose of this paper is to provide a statistical method of globular plant inventory to reduce the labor burden of front-line workers.This article specifically relates to the field of computer Digital Image Processing Technology.The contents of the study are:1.Collected nursery images.Used UAV(Phantom 4)collect the aerial images of the nursery in Beijing National Green belt located in the South Sixth,Beijing Xinanzhao nursery base in Fangshan,Beijing Xiaotangshan nursery base in Changping and Henan Yanling Longyuan flowers limited liability company,collection work was carried out in clear,windless weather.And recorded the relevant environmental parameters with the thermometer,hygrometer,and wind speed anemometer at the same time.2.Designed count algorithm of globular plants.First,the nursery image was cropped to remove irrelevant image edge information,such as sidewalks,street trees and so on.And then compared the OTSU and K-MEANS segmentation methods.Selected the K-MEANS clustering algorithm based on LAB space is selected.The plant targets and the background often randomly appeared due to the instability of the traditional K-MEANS segmentation,according to the total area of the plants was usually less than the background in the aerial nursery image,compared the sum of the black pixels and the white pixels in the image after dividing the image with the traditional K-MEANS,if the white pixels were more than black pixels,did negative transformation,otherwise stayed the same.And then used the the morphological method of median image filtering,hole filling,open operation to enhance the image,which could remove the noise and optimize the plant's spherical features.Finally,counted the segmented image with the algorithms of connected component labeling based on Dulmage-Mendelsohn decomposition and random Hough circle detection,and found that the algorithm of random Hough circle detection could better deal with the case of plant overlap,so it was used as the final count algorithm for the globular plants of the nursery image.Among them,the setting of the radius threshold in the random Hough circle detection algorithm was related to the image resolution,according to the experience,the threshold of the radius of the random Hough circle was set[9,88].By comparing the experimental results of 24 images with different resolution in 4 fields,determined that the sum of the optimal image pixels matched with the algorithm are around 120000,and the optimal resolution could be 310*372,450*263,310*397 and so on.After adjusting the best image resolution,counted the globular plants again.The experiment showed that the count result was relatively accurate for images with the single background and strong color contrast between the plants with background.For the overlap of plants in two lines and two lines below,the algorithm would effectively identify and count well,and could only count the plants on the border more than two lines.3.Nursery image stitching.The shooting area could be about 708 m2 when UAV flight height was 25m.For planting seedlings in large areas,a single aerial image couldn't describe the entire area required,so the large nursery image was collected with snake-like shooting method.Used the image stitching technique based on SIFT algorithm to connect the continuous aerial images of the large area nursery,and the number of its plants were obtained by counting the panorama.4.Development of count software system.Used the software engineering analysis method to develop the corresponding count software system.First,the demand analysis of the spherical plant counting software system was carried out,including investigation,requirement description and feasibility study.In the option of market survey,the market investigation of the globular plant counting software system had been conducted first,and listed the count defects of 'ImageJ' and 'IPP' image analysis software.Then the function,performance,economic feasibility and technical feasibility of the software was designed.Finally,the software of the counting software was developed.The software mainly integrated the plant counting algorithm which used the random Hough circle detection and the image stitching technique based on SIFT algorithm.Developed image import,image display,plant count,image stitching,image and data export functionIn summary,the count algorithm of globular plants in nursery image in this paper can be described as follows:Firstly,cropped the single UAV aerial image and adjust the image resolution to 400*300,and then used the improved K-MEANS clustering for image segmentation,used the median filter,hole filling,open operation for image enhancement,then the count algorithm of globular plants was realized by the random Hough circle detection algorithm.Secondly,used UAV collect the images of planting seedlings in large area with snake-like shooting method,used image mosaic algorithm based on SIFT to restore the panorama of planting seedlings in large area and counted it,the count result was the number of plants in large area.And finally develop software system of globular plant count in the nursery image.