Research On Remote Sensing Image Classification Based On Machine Learning

Machine learning is a fundamental and key issue in the field of image processing,especially in the field of remote sensing image processing.The machine learning method can separate the main features of the image from the complex data,so that the remote sensing image can be reasonably applied in various industries and fields.Machine learning-based image processing technology has been widely used in remote sensing image classification,segmentation,and recognition.It is a focus of attention and research in various fields.However,due to the complexity of remote sensing imagery feature distribution and different application backgrounds,the improvement of remote sensing image classification accuracy has become a key and difficult point.Therefore,how to improve the classification method to improve the classification accuracy and classification effect of remote sensing images is a very meaningful and difficult research topic.In view of the development trend of machine learning in remote sensing image processing applications,this paper is devoted to the research,exploration and improvement of some hotspot algorithms in machine learning methods,and applied to the classification of remote sensing images.Based on the existing machine learning classification methods,this paper combines improved genetic algorithm,patch-patching technology and fuzzy logic optimization techniques to study the improved SVM(Support Vector Machine)decision tree and RBF(Radial basis function)neural network algorithm.Then,an improved classification algorithm of CNN(convolutional neural network)architecture is proposed and applied to the classification of remote sensing images.The main work of this article is summarized as follows:A classification method based on the JM(Jeffries-Matusita)distance is proposed for the classification algorithm based on the combination of SVM and decision tree.This method is well solved the traditional support vector machine decision tree in the classification process ifa certain node is misclassified,this error will be derived to the next node,resulting in the more serious classification errors closer to the root node,reducing the accuracy of the classification,and improves image classification accuracy.For the classification of remote sensing images by RBF neural network,a fuzzy RBF neural network classification method based on genetic algorithm is proposed.This method applies the fuzzy theory to the RBF neural network and overcomes its problem of getting into a local extreme point.Then it uses the genetic algorithm to determine the optimal weights and thresholds of the RBF fuzzy neural network,and trains the network to improve the classification accuracy.Based on the CNN classification algorithm,an improved CNN architecture is proposed.The proposed method deals with remote sensing image classification in an end-to-end manner.The patch fragmentation method can be used to better extract the data features that are difficult to extract,such as corners of the image,so that the feature extraction of the image data can be more complete;In addition,optimizing the parameters of the CNN model from a small training set can alleviate the problem of overfitting of the neural network to some extent.Finally,the three improved methods proposed in this paper are compared and verified.The same experimental data is used to verify the three classification methods.