Water detection is a salient feature in conservation of water and for crisis management. This work avails of Synthetic-Aperture Radar (SAR) images, using various methods of deep learning for detection of water. The adhoc for candidate’s submission are on the basis of the changes in the detection of SAR images from multiple passes in the same place. It is allocated to the target neural network system to examine the presence of water. The method proposed applies the pre-training system that exists and is trained to classify natural red-green-blue (RGB) images. SAR images represent the non-standard images. The adoption of SAR images has proposed an approach that RGB images are used in pre-training networks. Subsequently the pre-trained system is well-tuned for differentiate of water on its appearing from the SAR image candidate area. This work uses five deep learning methods for water detection, such as AlexNet, ZFNet, VGGNet, Google Net and DenseNet. As compared with all networks, the DenseNet gives the best result. The sensitivity, specificity and accuracy of DenseNet is 97.15, 96.25 and 98.20%

水检测是节约用水和危机管理的一个显着特征。这项工作利用合成孔径雷达 (SAR) 图像，使用各种深度学习方法来检测水。候选人提交的adhoc是基于同一地点多次通过SAR图像检测的变化。它被分配给目标神经网络系统来检查水的存在。所提出的方法应用现有的预训练系统，并经过训练对自然红绿蓝 (RGB) 图像进行分类。SAR 图像代表非标准图像。SAR图像的采用提出了一种将RGB图像用于预训练网络的方法。随后，预训练的系统被很好地调整以区分水从 SAR 图像候选区域中的出现。这项工作使用了五种深度学习方法进行水检测，例如 AlexNet、ZFNet、VGGNet、Google Net 和 DenseNet。与所有网络相比，DenseNet 给出了最好的结果。DenseNet 的灵敏度、特异性和准确度分别为 97.15、96.25 和 98.20%