Based On Deep Learning Space Spectrum Joint Hyperspectral Image Classification Algorithm And System Implementation

The rapid development of hyperspectral remote sensing technology has brought new opportunities and challenges in several fields,such as agriculture,forestry,environmental monitoring and geological exploration.The classification of hyperspectral images has important application prospects for the analysis of object cover,precision agriculture and environmental monitoring using a few labeled samples.At the same time,involving processing and analysis of high-dimensional "space-spectrum unified" data,makes a great theoretical significance for machine learning and pattern recognitionAccording to the supervised classification of hyperspectral image(HSI)based on deep learning,this paper focuses on the heterogeneous deep neural network based classification algorithm using feature level and decision level fusion,which makes full use of spatial-spectral information to enhance the classification performance and robustness of the model under a small number of supervised samples.After reviewing deep learning based supervised classification of HSI,this paper mainly includes the following three aspects:1.A new HSI classification algorithm based on feature level fusion of heterogeneous neur al networks is proposed.This algorithm improves the fusion process of spatial-spectral feature of spatial-spectral classification algorithm.On the one hand,deep spectral feature is extracted by training a fully connected neural network model using spectral information.On the other hand,a deep convolution neural network(CNN)is used to extract spatial feature,along with deep spectral feature.The algorithm overcomes the feature incompatibility problem brought by feature splicing directly before classification.Experimental results show that the proposed algorithm is superior to the current deep learning based classification of HSI which splicing features directly before classification.2.A decision level fusion algorithm of HSI classification based on heterogeneous neural networks is proposed.In this algorithm,HSI spectral and spatial information is used to train a CNN and a fully connected neural network model of pixel level respectively.Then,proposed bilateral weighted decision fusing strategy is used to deal with soft probability classification maps generate by these models,to combine spatial-spectral classification.Compared with the previous single-neural-network classification algorithm,the classifier is more stable and robust.Compared with the HSI classification algorithm which splicing features directly before classification,proposed algorithm does not bring in high-dimensional data.Experimental results on the widely used HSI indicate that the proposed approach outperforms several well-known classification methods in terms of classification accuracy.3.A deep learning based HSI classification system is designed.The system includes visualization of HSI,HSI classification and pixel analysis module.It can realize several significant functions,such as classification,statistics and analysis of HSI.