Research On Remote Sensing Image Cloud And Cloud Shadow Detection Method Based On Convolutional Neural Networ

Cloud and cloud shadow detection in remote sensing image is the key to remote sensing image processing.The accuracy of detection results directly affects the application of subsequent remote sensing images,which has important research significance.In recent years,with the development of artificial intelligence technology,the cloud and cloud shadow detection method of remote sensing image based on the convolutional neural network has attracted much attention.Compared with the traditional method,the convolutional neural network-based method can achieve higher detection accuracy and better robustness.At the same time,it reduces the complexity of the algorithm and greatly improves detection efficiency.Aiming at different application scenarios of cloud and cloud shadow detection,this research realizes highprecision and lightweight detection based on the convolutional neural network.The main research contents and innovations are as follows:1.Aiming at the RGB color remote sensing images,cloud pixels have no obvious texture features,which is easy to produce false detection.A convolutional neural network model based on a dual attention mechanism(RDANet)is proposed,which is applied to the field of cloud and cloud shadow detection.The Da Block is introduced to effectively capture the global feature dependencies.The Res Block is used to avoid the degradation of the deep network.Combined with the R-ASPP module,the multi-scale features of the image are extracted without changing the size of the feature map.We trained and tested the Gaofen-1 WFV(Wide Field of View)remote sensing image dataset.The experimental results show that this method can effectively improve the accuracy of cloud and cloud shadow detection,and can still obtain better edge details of cloud and cloud shadow under complex conditions.2.It is difficult to detect small area cloud,thin cloud,and cloud shadow.A convolutional neural network model based on context information fusion(CIFNet)is proposed to improve the ability of context information extraction and fusion,and realize the accurate detection of cloud and cloud shadow in Gaofen-1 WFV remote sensing image.A Res Block-cloud is designed to capture global and local features and prevent network degradation.A GCF module is designed to fuse different levels of global context information through dense skip-connections and guide it to fuse with the decoder path features.Finally,an MCF module is designed to extract the multi-scale context relationship between cloud and cloud shadow.Experimental results show that this method can effectively detect small area cloud and thin cloud,and significantly improve the accuracy of cloud shadow detection.3.Aiming at the problem of the limited payload of microsatellites and limited computing power of hardware systems,and the high computing power requirements of existing methods limit the application of microsatellites in orbit cloud detection.A lightweight convolutional neural network model based on a depthwise separable convolution(L-MNet)is proposed.LMNet model is improved by M-Net model.The depthwise separable convolution is introduced and a DS-Conv Block is designed to reduce the complexity and computation of the algorithm.Experimental results show that under the premise of ensuring the detection accuracy,this method can reduce the size and parameter amount of the model,which is conducive to the task of microsatellites cloud detection in orbit.