Research Of Infected Pine Recognition In Remote Sensing Images Based On Support Vector Data Description

China is the most serious country in the forestry harmful biological disasters all around the world. In addition, pines are very important forest species resources with their high economic and practical value. In recent years, due to the invasion of Bursaphelenchus xylophilus disease, plenty of dead pines have been discovered on the trot in many regions. The disease will lead to the destruction of the whole pine forests for the reasons such as high propagation speed, strong pathopoiesis ability and wide parasitism range so that known as the head of the forestry harmful biological disasters. It does not only threaten the forest resources and ecological construction seriously, but also take enormous losses to the national economy. Therefore, research and realize the recognition and supervision of infected pines is very meaningful which can prevent the pines from insect pests, protect pine resources, promote the sustainable development of economy and so on.Traditional infected pine recognition depends on the artificial field investigation of plant diseases and insect pests experts. The method is very difficult to master the outbreak accurately and to be popularized on a large scale on account of the mountainous terrain and the responsibility of investigators. With the quick development of the remote sensing technology, it’s very effective to supervise mass dead pines through high-resolution remote sensing data acquired by the satellite monitoring platform. However, it’s very difficult for the satellite remote sensing technology to find the single infected pine on account of the satellite run cycle and the atmospheric environment. Besides, it’s not appropriate to be applied chronically on a large scale due to the constraint of the funding cost. In consequence, finding a kind of economic and effective way for the recognition and supervision has been becoming an important research project.In the dissertation, the small unmanned aerial vehicle(UAV)and the double spectrum camera were used respectively as the airborne platform and the remote sensing detection means to acquire the visible and near-infrared images of the pine forests. The recognition and supervision of infected pines was realized by the comprehensive application of the image processing, the pattern recognition and so on. The main research contents were as follows:1. The aim and significance, the domestic and foreign research status, the research area and data source for the recognition and supervision of the infected pines were elaborated. In addition, the basic theory and development status of support vector data description were introduced.2. The method of infected pine recognition in remote sensing images based on weighted support vector data description was proposed. Firstly, each color component for visible and near-infrared images was extracted as the color feature of the corresponding pixel on the basis of the difference of content information. Then the texture feature of the central pixel was acquired by extracting the gray level co-occurrence matrix of the adding window image block. Finally, the weighted support vector data description multi-classification model was constructed by building the weight function on the center distance of the training sample, in order to realize the multi-output classification of the objects in images, namely the infected pine recognition. Multi-group contrastive experiments proved the recognition veracity of the method.3. The method of infected pine recognition based on multi-feature and improved weighted support vector data description multi-classification was proposed. The feature was extracted based on single pixel and centre local area of each sample to the corresponding feature vector. On the basis of traditional support vector data description multi-classification, the samples located in fuzzy area were decided by the KNN membership function. And the weight function and the wave kernel function were beneficial to construct the weighted wave support vector data description multi-classification model for the infected pine recognition. The experiment results showed that the proposed method avoided the blind decision of samples located in fuzzy area in traditional method and tried to find the label owned by high density samples in the local samples, therefore it improved the recognition rate of infected pines. 4. The method of multilevel infected pine recognition based on feature sparse representation and weighted wave support vector data description was proposed. Firstly, the features were extracted in double spectrum images. Then, the learning dictionary was constructed by applying feature data set. According to the dictionary, the sparse coefficient of each sample was calculated and the sparse representation feature vector was acquired. Finally, the feature vectors were imported to weighted wave support vector data description multi-classification model. Thus, different level infected pines were recognized effectively. The experiment results indicated the proposed was feasible and effective not only in the visual sense but also in the quantitative evaluation.