With the development of satellite remote sensing technology, the quality and quantity of remote sensing images are constantly improved. Remote sensing feature classification is also playing an increasingly important role in urban planning, resource exploration and other fields. In the early stage of remote sensing feature classification, machine learning algorithms such as SVM and K-means are mainly used. Nowadays, with the expansion of deep learning, various kinds of research in the computer vision field emerge in an endless manner. Remote sensing images are also mostly classified by different neural networks. According to the characteristics and advantages of U-NET, channel attention mechanism, ResNet, large convolution kernel and structural reparameterization, this paper proposes a network structure called RA-UNET. This paper uses the remote sensing ground object classification dataset LoveDA to conduct experiments. The results show that the network classification effect of this paper is better, with mIoU reaching 59.4% and mPA reaching 72.6%. And use the network in this paper and the four mainstream neural networks of FCN, SegNet, PSPNet and UNet to conduct comparative experiments. The comparative experimental results show that the classification effect of the network in this paper is better than the above four mainstream neural networks.
For the high-resolution remote sensing building image, the object shape is irregular and the texture characteristics are similar to those in the foreground and background. Meanwhile, traditional manual methods spend considerable time extracting features. In this paper, a Deeplabv3+ algorithm is modified and improved to better meet the accuracy, massive, and real-time requirements in semantic segmentation of high-resolution remote-sensing building objects. First of all, the lightweight network MobileNetv2 is implemented to replace Xception as the backbone to extract the network, so as to reduce the number of network parameters and improve the training speed. Secondly, ASPP (Atrus Spatial Pyramid Pooling) obtains the depth effective feature layer and integrates three parallel channel attention mechanisms to improve the extraction effect of depth features. The newly proposed algorithm is tested on the Massachusetts dataset. The results show that the IOU and accuracy of evaluation indicators are increased by 8.31% and 2.11%, respectively, compared with the original Deeplabv3+ model. While reducing the number of model parameters and training time, the segmentation effect of the new algorithm has also been partially optimized.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.