The Automatic Monitoring Method Study Of Mikania Micrantha Kunth Based On Computer Vision

Mikania micrantha Kunth(M.micrantha)is widely distributed in various provinces in the south of China,which seriously damages the ecological environment of forests and farmland and causes serious economic losses.To prevent the further spread of M.micrantha,and protect the biodiversity,it’s necessary to automatically monitor the M.micrantha.In the wild,M.micrantha and phorophyte have low visual identification under visible light,irregular distribution,and the environment complex and changeable.Therefore,it is difficult to realize real-time and accurate monitoring of M.micrantha by general detection methods.In this paper,hyperspectral images and high-definition color images of M.micrantha were taken as research objects,and the identification methods of M.micrantha in the wild were studied by using computer vision technology.The main research work were as follows:(1)Recognition of M.micrantha based on hyperspectral analysis:When working with hyperspectral data,preprocessing,dimension reduction,and classifier are fundamental to achieving reliable recognition accuracy and efficiency.Five preprocessing methods,principal component analysis(PCA)dimension reduction,and three classifiers are used in different combinations to process the 138-dimensional hyperspectral data of M.micrantha,which was collected from the field.The combination method of Savitzky-Golay(SG)convolution smoothing,PCA and a random forest(RF)achieved an accuracy(A)of 88.71%,an average accuracy(AA)of 88.68%,and a Kappa of 0.7740 with an execution time of 9.647 ms.In comparison,the combination method of SG smoothing,PCA and a support vector machine(SVM)resulted in lower values of A(84.68%),AA(84.66%),and Kappa(0.6934)with a shorter execution time(1.318 ms).According to specific identification accuracy requirements and time cost,SG-PCA-RF and SG-PCA-SVM are two potential methods for recognizing M.micrantha in the wild.(2)Recognition of M.micrantha based on high-definition color image:In order to meet the demand of the high precision of M.micrantha and Invasive alien plants(IAPs)monitoring in the wild.We acquired color images of the monitoring area in the wild environment using an unmanned aerial vehicle and proposed a novel network-MmNet-based on a deep convolutional neural network(CNN)to identify M.micrantha in the images,and a improved network-IAPsNet-based on MmNet to identify IAPs in the images.After training and testing,the identification of 400 testing samples of M.micrantha by MmNet is very good,with accuracy of 94.50%and time cost of 10.369 s.And the identification of 893 testing samples of IAPs by IAPsNet with accuracy of 93.39%and time cost of 1.8846 s.Moreover,in quantitative comparative analysis,compared with recently popular CNNs,the proposed MmNet and IAPsNet have higher accuracy and efficiency.Moreover,IAPsNet has successfully applied to monitor the M.micrantha and IAPs in the real-time.The above research provides some theoretical basis for the development of intelligent monitoring systems for M.micrantha and IAPs.