Multi-Element Features Extraction from High Spatial Resolution Remote Sensing Image Based on Semantic Segmentation Network

The high spatial resolution remote sensing image contains rich special information with a great detail. Traditional segmentation methods have low classification accuracy, the complexity of high spatial resolution remote sensing images also requires higher the classification technology of remote sensing images. In recent years, neural networks have made remarkable achievements in the field of image segmentation. Our method was tested on the High-Resolution Remote Sensing Segmentation Dataset, comparative semantic segmentation models such as FCN (Fully Convolutional Networks), SegNet and DeepLab-V3+. Experimental results showed that the overall accuracy of FCN, SegNet and DeepLab-V3+ methods were 82.8%, 85.5% and 87.4 %. The analysis results show that the DeepLab-V3+algorithm has higher iterative efficiency in model training; SegNet is the second, and the FCN algorithm has the worst iterative efficiency. The DeepLab-V3+algorithm has the highest accuracy and the fastest iteration efficiency in the extraction of multi-element features from high spatial resolution remote sensing images, which achieved a balance between accuracy and computational resource consumption.