Research On Classification And Recognition Methods Of Objects Based On Multi-temporal Remote Sensing Images In Beibu Gulf Of The South China Sea

With the continuous development of remote sensing technology,remote sensing data are more and more abundant.There are multi-temporal remote sensing image data sets in the same locality based on massive remote sensing data.The improvement of temporal resolution means that there are more and more information of time dimension in the multi-temporal remote sensing images.And the increase of time dimension information expands remote sensing data application fields.How to use existing well-rounded technique approaches and mine the temporal dimension information embedded in the remote sensing data to improve remote sensing image classification and recognition accuracy has become a non-evasive problem commonly concerned by researchers in remote sensing technology application.In order to resolve the problem,this paper built the classification and recognition algorithms of multi-temporal remote sensing images in two ways(one is to improve and optimize the existing remote sensing image classification algorithms,and the other is to set up an new classification and recognition algorithms of remote sensing images)by studying how to mine the temporal dimension information embedded in the remote sensing data based on multiple classifier combination(MCC),random forest algorithm and fingerprint identification technology.In detail,firstly,the paper developed an improved algorithm of multiple classifier combination using the stable weight value through the analysis of multiple classifier combination and random forest algorithm,and the new algorithm was applied to multi-temporal remote sensing images.Secondly,the paper brought forward a classification and recognition algorithms using multi-temporal remote sensing data pixels image based on fingerprint identification.At last,the experiments were given to prove the validity of these algorithms.The main contents in this paper are as follow:(1)The algorithm of multiple classifier combination using weighted vote based modified weight(MCC_WVA-MW)was put forward.Firstly,the classification results of multiple classifier combination using different vote models were comparatively analyzed.Secondly,the paper brought forward the concept of stable weight and revised the stable weight algorithm in order to deal with the instability problem caused by accuracy weight.At last,this paper presented the multiple classifier combination using weighted vote algorithm based modified weight.The algorithm could not only avoid the instability problem caused by accuracy weight,but also prominently improve the capability of the classifiers and the accuracy of classification results.According to the testing,the overall accuracy of the MCC_WVA-MW is raised form 52.76%-58.23%to 61.54%.(2)This paper proposed the algorithm of multiple classifiers iterative combination using weighted vote based modified weight(MCIC_WVA-MW).In this paper,we improved the MCC_WVA-MW above using the principle of iteration,and presented the algorithm of MCIC_WVA-MW.This algorithm could adequately dig out the potential of the MCC algorithm,and improve the classification accuracy.(3)By analyzing the three mainly considered influence factors in the process of remote sensing image supervised classification,the author presented the algorithm of multi-temporal remote sensing images multi-samples multiple classifier combination using weighted vote algorithm based modified weight(MS_MT_MCC_WVA-MW)by using the principle of random forest algorithm in this paper.The algorithm of MS_MT_MCC_WVA-MW greatly improves the accuracy of classification results,because it enhances the capability of the classifiers,avoids the instability problem caused by random training sample,and reduces the influence of different object with the same spectra characteristics and same object with the different spectra characteristics in the process of classification of remote sensing.The test results show that the overall accuracy of the classification algorithm is raised form 42.30%-69.08%to 83.09%.The accuracy of the algorithm is improved significantly.(4)The classification and recognition algorithms using multi-temporal remote sensing data pixels image based on fingerprint identification have presented by using the principle of fingerprint identification technology in this paper.These algorithms improve the classification accuracy and the operating efficiency because the information of multi-temporal remote sensing images is used adequately and it figures out the problem of low efficiency caused high dimensional remote sensing data in the process of classification.The test results show that the overall accuracy of the multi-temporal remote sensing images classification algorithm based on fingerprint identification is up to 80%.And the Accuracy of mangrove and eucalyptus forest identification result by the object recognition algorithm using multitemporal remote sensing data pixels image based on fingerprint identification is up to 79.63%and 75.54%.The classification and recognition algorithm has achieved satisfactory results.This paper proposed the algorithms of multiple classifier combination using weighted vote algorithm based modified weight(MCC_WVA-MW)and its iteration algorithm,and applied it to multi-temporal remote sensing image data sets,then we established the algorithm of MS_MT_MCC_WVA-MW.The methods improve the capability of the classifiers and the accuracy of classification results,because it weaken the mainly negative influences in the process of remote sensing image supervised classification.But the algorithm has many problems,for example,an extremely complicated and tedious process and low-efficiency.And the classification algorithms of multi-temporal remote sensing data pixels image based on the theory of digital imaging and pattern recognition have presented in this paper.To improve the classification efficiency,the method mines the temporal dimension information of multi-temporal remote sensing images and improves the effect and accuracy of remote sensing classification by using the digital image compression technology.And then the classification algorithm was translated into recognition algorithm by setting certain threshold.The recognition algorithm can be used to the object recognition of remote sensing image effectively.The classification and recognition algorithms of multi-temporal remote sensing images suggested in two ways in the paper have their own advantages and disadvantages.