改进机器学习在复杂场景无人机测绘图像分类中的应用OA

In unmanned aerial vehicle surveying and mapping of complex scenes,different categories of surface features may exhibit similar features in remote sensing images,resulting in poor separability between classes.This makes it difficult for classifiers to accurately distinguish between different categories of surface features,thereby reducing the accuracy of clas-sification.Therefore,research on UAV surveying and mapping image classification of complex scenes based on improved machine learning was conducted.Firstly,the fractal net evolution approach method(FNEA)was used to perform multi-scale segmentation on UAV surveying and mapping images.The segmentation objects were determined by calculating spec-tral heterogeneity and shape heterogeneity,where spectral heterogeneity was measured by band spectral standard deviation and layer weight values,and shape heterogeneity was measured based on smoothness and compactness characteristics.Subse-quently,principal component analysis network(PCAnet)was used for feature filtering,and a set of discriminative feature vectors was output through principal component analysis(PCA)convolutional layers and hash histograms.Finally,an improved support vector machine(SVM)decision tree classification method was adopted,and a genetic algorithm was intro-duced to optimize the decision tree structure.Through fitness function design and genetic operations,feature vectors were trained to generate the optimal decision tree,enhancing the separability between different categories.In the classification stage,SVM and k-nearest neighbor(KNN)algorithms were combined to achieve high-precision classification.The experi-mental results show that the proposed method has high data classification accuracy and stability,with a Kappa coefficient always above 0.75,indicating high practical application value.