Cloud and snow detection of remote sensing images based on improved Unet3+

Abstract

Cloud detection is an important step in remote sensing image processing and a prerequisite for subsequent analysis and interpretation of remote sensing images. Traditional cloud detection methods are difficult to accurately detect clouds and snow with very similar features such as color and texture. In this paper, the features of cloud and snow in remote sensing images are deeply extracted, and an accurate cloud and snow detection method is proposed based on the advantages of Unet3+ network in feature fusion. Firstly, color space conversion is performed on remote sensing images, RGB images and HIS images are used as input of Unet3+ network. Resnet 50 is used to replace the Unet3+ feature extraction network to extract remote sensing image features at a deeper level, and add the Convolutional Block Attention Module in Resnet50 to improve the network’s attention to cloud and snow. Finally, the weighted cross entropy loss is constructed to solve the problem of unbalanced sample number caused by high proportion of background area in the image. The results show that the proposed method has strong adaptability and moderate computation. The mPA value, mIoU value and mPrecision value can reach 92.76%, 81.74% and 86.49%, respectively. Compared with other algorithms, the proposed method can better eliminate all kinds of interference information in remote sensing images of different landforms and accurately detect cloud and snow in images.

A deep learning method for optimizing semantic segmentation accuracy of remote sensing images based on improved UNet

“…Since CNN has caused many breakthroughs in computer vision tasks of natural images, it has been widely employed in the work of the semantic segmentation of RSI 35 .Some researchers have used CNN for some specific applications on RSI. These tasks have included but have not been limited to the extraction of multiple classes of geo-objects in the image, as in this paper, but also the extraction of only a single class of geo-objects, such as building extraction 36 , 37 , road extraction 38 – 40 , cloud and snow detection 41 , and urban village mapping 42 . Some models were developed, such as AWNet 43 , HA-MPPNet 44 and HED-UNet 45 .…”

A brief analysis of remote sensing image cloud extraction method based on deep learning

“…If the color space is converted into HIS color space, this colour space is based on the human visual system. It describes the image with hue, intensity and saturation, and expresses the color category and degree of homogeneity in a deeper level, For highly similar objects such as white clouds and snow, the types of objects can be distinguished in more detail [57] . Using multi angle information as features in remote sensing images will also bring great help.…”

Enhanced Tropical Cyclone Precipitation Prediction in the Northwest Pacific Using Deep Learning Models and Ensemble Techniques

“…Another enhancement, UNet3+, employs full-scale skip connections to combine low-level detail with high-level semantics from feature maps at different scales, maximizing the use of full-scale feature maps. This approach has demonstrated promise in weather monitoring applications [35][36][37]. Despite their potential, there have been few studies conducted on the application of SE-UNet and UNet3+ in improving precipitation forecasting.…”