Research Of Cloud Detection Algorithm For Landsat Image Based On Feature Fusion

Landsat satellite remote sensing image has been widely applied in environmental monitoring,resource exploration,disaster analysis,forestry investigation,agricultural production and other fields.First,it will affect the accurate interpretation of the image if the observation area is covered by cloud layer.Secondly,if the useless data of cloud region is deleted by cloud detection when satellite image data back to the ground,the utilization ratio of broadband will be increased.Therefore,accurate cloud detection for Landsat images is of great importance.This paper research and realize the application of Gabor wavelet and convolution neural network in Landsat image cloud detection.Then the features of Landsat image extracted by Gabor wavelet and convolution neural network are fused,and the KECA algorithm is introduced to reduce the image features of fusion,completed the cloud detection experiments and comparison analysis.The innovation of this paper is lies in: Proposed the Landsat image cloud detection algorithm based on the convolution neural network,the experimental results show the superiority of the algorithm in the detection precision.Proposed the cloud detection algorithm based on fusion the features extracted by Gabor wavelet and convolution neural networks,and introduced the KECA algorithm to improving the accuracy of cloud detection and reduce the time complexity of the algorithm.is introduced to improve the cloud effectively.The accuracy of detection and the time complexity of algorithm are reduced.The main contents of this paper are as follows:(1)Research the Landsat image cloud detection algorithm based on Gabor and SVM.Based on the analysis of the principle of Gabor wavelet algorithm and the features of Landsat image,extracted the 160 dimension feature of Landsat sample image by setting the Gabor filter of different directions and scales.In order to solve the problem of large SVM operation,introduce the LSSVM classifier to classify the extracted features,completed the experiment of of the Landsat image cloud detection.Finally,the feasibility of the algorithm is verified by experimental comparison.(2)Research and propose a cloud detection algorithm based on convolutional neural network for Landsat images.Based on the analysis of the principle and solution of the convolution neural network model,completed selection of the activation function,the number of network layers,the number of convolution kernel and the receptive field size through the cloud detection experiment,constructed a cloud detection model based on convolution neural network.Finally,experimental results verify the effectiveness of cloud detection algorithm based on convolutional neural network.(3)In order to further improve the effect of cloud detection,propose the cloud detection algorithm that fusing the image features extracted by Gabor wavelet and convolutional neural network based on the previous research.That is to combine the extracted features into a lossless new feature vector through serial combination,and use LSSVM to classify the fused features,completed the cloud detection experiments after feature fusion.After that,introduce KECA algorithm to optimize the fusion feature vectors,and used LSSVM to classify and detect the image features after dimension reduction.Experimental results show that our algorithm can effectively improve the accuracy of cloud detection and reduce the time complexity of the algorithm.