Research On Hyperspectral Image Classification Method Inspired By Complementary Informatio

Hyperspectral images contain an abundance of spectral and spatial information,making them valuable in a wide range of applications such as land cover classification,precision agriculture,and military operations.Hyperspectral image classification serves as a fundamental step in these applications.Although existing deep learning-based hyperspectral image classification methods have achieved some progress,they still face several challenges.First,the high cost of hyperspectral image data acquisition and the time-consuming,specialized knowledge required for the annotation process make it difficult to obtain labeled samples.Consequently,the number of labeled samples is often insufficient for current classification tasks.Second,hyperspectral images are influenced by various factors such as observation conditions,feature internal structures,surface cover,and sensor characteristics.These factors may result in the " same object with different spectrum " and " different objects with the same spectral " phenomena,making it challenging to obtain ideal classification results using hyperspectral images alone.To address these two issues,this study has carried out the following work.First,to address the issue of insufficient labeled samples in hyperspectral image classification tasks,we propose a hyperspectral image classification method based on spectral complementary information.This method does not require labeling samples and only needs to model the complementary information between different spectral bands to achieve feature extraction.Specifically,since different spectral channels in hyperspectral images depict the same object’s response to varying electromagnetic bands,there must be a feature space that enables similar representation across channels.Inspired by this,we first divide the highdimensional spectral information into two groups,then use multi-layer convolution operations to extract features from each group of bands,and finally compare the features extracted from different samples to optimize the model through a contrastive loss function.To verify the effectiveness of this method,experiments were conducted on four public datasets.The results show that,with only 10 training samples per class,our unsupervised learning model achieves superior classification performance compared to common unsupervised models such as principal component analysis and autoencoders.Second,to address the issues of " same object with different spectrum" and " different objects with the same spectral " in hyperspectral images,we introduce Li DAR data as a supplement and proposes a hyperspectral image classification method based on multimodal complementary information.This method uses a dual-branch convolutional neural network,with each branch extracting features from hyperspectral images and Li DAR data through three convolutional layers.Firstly,a feature adjustment module is built on each convolutional layer,utilizing the complementary information between modalities to enhance the representation capability of convolutional features.Then,a feature fusion module is constructed at the output position of the first convolutional layer,dynamically allocating weights using the complementary information between modalities.Finally,after obtaining the features of the two data sets through the third convolutional layer,they are fused using a weighted summation.To verify the performance of this method,it is compared with advanced fusion models(including traditional models and deep learning-based models)on two public datasets.The experimental results show that the proposed method can achieve competitive classification results.