Research On The Vegetation Automatic Identification Based On Visible Light Images Of UAV

Vegetation restoration is an important method to control high-intensity erosion.The survey and dynamic monitoring methods of vegetation restoration mainly have two types:Field survey and interpretation of satellite remote sensing image.The former method is time-consuming and laborious,while the latter one is unable to accurately identify the vegetation within a small area.This research studied the low altitude visible light images of UAV taken from the erosion areas within Anxi County and Changting County and constructed an automatic identification database after manual processing of visible light images.The computer learning algorithm and comparison algorithm were adopted to identify the vegetation in the images after analysis and then compare the results with the images of artificial identification to get the identification accuracy.The advantages and disadvantages of these two algorithms as well as the effects of vegetation automatic identification were discussed in the thesis.The main conclusions are as follows:(1)After artificial identification of more than 1,000 aerial visible light images,it is found that the best identification effects are at the height of 20 m and 30 m,while the effect differences of these two methods are not significant.The effect of artificial identification gradually decreases as the height increases,and the image definition declines sharply at the height of 60 m and above,the vegetation boundaries in which are too ambiguous to accurately identify the vegetation types.The height range of computer automatic identification is determined from 20-50 m;The actual coverage area of aerial visible light images under the height of 20m,30m,40m and 50 m are 750 m~2,1430 m~2,2280 m~2,2810 m~2 respectively by joining the reference in the process of sampling;The proportion of various vegetations in the obtained visible light images shows that the most abundant vegetations are arbor and fern when classified in terms of life from,Masson pine and Dicranopteris pedata when classified in terms of species.(2)After the automatic identification of the vegetations in visible light images through computer learning and comparison,it is found that the identification accuracy of vegetation main class is better than subclass.When the learning algorithm is applied to automatically identify the vegetation,it is easy to misjudge between arbor and fern,Masson pine and Dicranopteris pedata,which can finally influence the identification accuracy;the comparison algorithm has a low identification accuracy and a poor effect,and it is unable to determine the specific area of wrong identification.The identification effect of arbor and Dicranopteris pedata is well only in the learning algorithm,while in other cases,the classification and identification effects are unsatisfactory,and the identification process needs to be optimized.(3)The automatic identification accuracy can be significantly improved in the computer learning algorithm by adding 2-3 sample photos of the test area,while the identification accuracy in the comparison algorithm is not greatly improved.Learning algorithm is adopted to identify the visible light images,the identification accuracy of arbor,fern,Masson pine and Dicranopteris pedata can reach up to 92.00%,62.86%,65.87%and 87.12%,respectively;The identification accuracy of comparison algorithm is much lower than that of learning algorithm,with the identification accuracy of arbor and fern are only 53.13%and 37.13%,the identification accuracy of Masson pine and Dicranopteris pedata are only 45.99%and 30.48%.