In the field of remote sensing, panchromatic sharpening technology integrates spatial data from panchromatic images with spectral data from multispectral images to generate high-resolution multispectral images. Precise mapping from multispectral to single-band panchromatic images greatly impacts the quality of fusion images. This paper introduces the spatial-spectral transformer network (SSTNet). SSTNet combines the spatial constraints of the difference model in the gradient domain and the intensity constraints in the spectral domain to quantitatively express multispectral imagery into panchromatic imagery in a many-on-one form, laying the groundwork for designing the loss function in image fusion. compared to the original method, SSTNet is applied to P2Sharpen. Experiments on the Quickbird dataset demonstrate that the evaluation indexes of the SSTNet-based P2Sharpen method are improved in both reduced and full resolution resolution, requiring fewer training data and epoch.
Rapid, accurate extraction of rural residential areas is of great significance to rural planning and urbanization. On the basis of the improved YOLOv8 object detection algorithm, this paper puts forward a technical method for accurately extracting rural residential areas from multi-scale remote sensing images. The rapid extraction of rural blocks comes true via improving the retrieval mechanism of YOLOv8 algorithm: First, the feature extraction module based on ECA local crosschannel interaction attention mechanism is designed to deeply dig the detailed features with inconsistent scales in the detection of residential areas. Efficient channel interaction pays more attention to the positive sample feature information in the feature map, and meanwhile, it reduces the complexity of the model. Second, Swish activation function is proposed to avoid gradient disappearance and poor activation effect caused by over-fitting. Third, DIoU loss is introduced to accurately show the real distance error between two predicted residential areas and enhance the performance of multitarget detection. In the end, ablation experiments and comparative experiments are conducted on CBDV1.0 building data set. The experimental results show that this method can extract rural residential areas from multi-scale remote sensing images, which provides support for large-scale remote sensing image mapping of rural residential areas.
Based on YOLOX network, this paper presented an algorithm for extracting point-like independent houses from remote sensing images. First, an Adaptively Spatial Feature Fusion (ASFF) network was added to the feature extraction module PANet to deeply mine the detailed features of small target houses with different scales. Second, a feature extraction module based on ECA local cross-channel interaction attention mechanism was designed. Efficient channel interaction paid more attention to the positive sample feature information in the feature map and lowered the complicacy of the model. Finally, the Swish activation function was used to avert poor activation effect. Experiments were conducted on the point-like independent houses data set, and the optimum mechanism and effectiveness of the improved method were validated by qualitative analysis of ablation experiments and quantitative analysis of comparison experiments. On the premise of adding ASFF mechanism and ECA attention mechanism and optimizing Swish activation function, the mAP precision of the improved network model was up to 94.83%, 11.16% higher than that of the original network. The robustness and effectiveness of the improved method were quantitatively verified by conducting comparative experiments with widely used detection algorithms.
In recent years, deep learning dense matching techniques have developed rapidly. Among the various training methods such as supervised, unsupervised and semi-supervised, the matching accuracy of supervised training methods is still much higher than other training methods. However, this training method requires the use of manually labelled parallax maps as samples. The accuracy of the labelling will directly affect the matching accuracy of the trained network model. Therefore, it needs to be analyzed and studied. In this paper, the noise immunity of the deep learning supervised method is studied by simulating systematic error, random error and gross error, and the experimental results show that: (i) The deep learning method has anti-noise ability within a certain range and has a better anti-noise effect on random noise, but the matching accuracy decreases at an accelerated rate as the noise increases. (ii) The pre-training method by transfer learning can effectively improve the matching accuracy, increase the noise immunity, and make the original non-converging network converge.
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.