High-Resolution Remote Sensing Image Classification And Application Using Multiple Kernel Extreme Learning Machine

In the past ten years,with the rapid development of aerospace and satellite sensing technology and the start of the implementation of major national high-level special projects,high-resolution remote sensing image data has increased dramatically,which provides information guarantee for national defense modernization in the fields of national censuses,road network design,crop production estimates,and disaster prevention and mitigation.However,due to the complex distribution of ground features in high-resolution remote sensing images,the traditional visual interpretation method requires a lot of manpower and material resources.Among them,in terms of manpower,there is also a problem of excessive subjectivity.For this reason,combining multiple features with neural networks to build classifiers has become the mainstream trend in remote sensing image classification.As a fast learning feedforward neural network,Extreme Learning Machine(ELM)provides an efficient and unified solution for clustering,regression,classification and other applications.Therefore,this thesis focuses on the application of extreme learning machines in remote sensing image classification.First,the background,significance of remote sensing image classification,research status,and problems faced are described.At the same time,the current status of ELM research is introduced.Second,the segmentation technology,classification algorithm,and ELM involved are introduced in detail.Aiming at the shortcomings of basic ELM in image classification,a multi-kernel ELM is constructed by introducing multiple kernel functions.Finally,to improve the classification accuracy and generalization ability of the classifier,based on a single classifier,a multi-kernel ELM integrated learning model based on the Ada Boost algorithm is proposed.The main research contents of this thesis are as follows:1.Aiming at the problem of the complexity of features in remote sensing images and the wide variety of ground objects,given ELM's fast learning ability and good generalization performance,a multi-feature multi-kernel ELM image classification algorithm is proposed.First,the image is initially segmented,and the typical ground feature objects are obtained through the area merge optimization algorithm.Second,the spectrum and spatial structure features are extracted and are weighted.Then,the cuckoo search algorithm is used to find the relevant kernel parameters of the multi-kernel ELM to obtain the optimal classification model.Finally,the constructed multi-kernel ELM classification model is applied to remote sensing image classification.The experimental results show that the ground category information was accurately distinguished.2.For the deficiency of single basis classifier classification accuracy and generalization performance,it is considered that the Ada Boost algorithm in integrated learning can generate several base classifiers.By combining the Ada Boost algorithm with multi-feature multi-kernel ELM and using the weighted voting mechanism,the final integrated strong classifier is obtained.The experimental results show that the integrated multi-kernel ELM improves the image classification accuracy and enhances the generalization performance.