基于ResNet18 U-Net模型的SAR图像海面溢油检测

Sea Surface Oil Spill Detection of SAR Images Based on ResNet18 U-Net

摘要:

针对合成孔径雷达(SAR)图像海面溢油检测问题，采用U-Net模型、ResNet18-UNet模型、ResNet34-UNet模型、ResNet50-UNet模型等四种卷积神经网络模型进行了SAR图像海面溢油检测的研究与实验，挑选出更适合此次SAR图像海面溢油检测研究与实验的溢油检测模型；通过利用四种卷积神经网络模型对同一数据集进行SAR图像海面溢油检测的方式，对比分析了不同卷积神经网络模型之间的性能差异和溢油检测效果；实验结果表明，在此次SAR图像海面溢油检测研究与实验中，ResNet18-UNet模型的性能在U-Net模型的基础上有了一定的优化，而且与另外两种ResNet-UNet模型相比，ResNet18-UNet模型在此次海面溢油检测实验中拥有最高的性能和最好的溢油检测效果，获得了更高的检测精度和检测效率。

Abstract:

Aiming at the problem of oil spill detection in synthetic aperture radar (SAR) images, four convolutional neural network models are used to study and experiment on oil spill detection in SAR images, including U-Net model, ResNet18-UNet model, ResNet34-UNet model, and ResNet50-UNet model. A more suitable mode for oil spill detection was selected for this research and experiment. By using the four convolutional neural network models for SAR image sea surface oil spill detection on the same dataset, the performance differences and oil spill detection effects among different convolutional neural network models were compared and analyzed. The experimental results show that for this oil spill detection using the same SAR dataset, the performance of the ResNet18-UNet model has been optimized on the basis of the U-Net model, and compared with the other two ResNet-UNet models, the ResNet18-UNet model has the highest performance and the best oil spill detection effect for the SAR dataset used in this study, achieving higher detection accuracy and efficiency.