18 December 2023 in IEEE Transactions on Geoscience and Remote Sensing
Infrared small target detection is widely used in remote sensing fields. However, the application scenes of space-based remote sensing imaging often lead to problems such as small target scale, weak energy, and serious influence by strong clutters. At present, traditional methods are often difficult to adapt to the change of target scale. And deep learning methods are often difficult to extract small target features, and the change in imaging characteristics also brings challenges to the generalization ability. To complement each other’s advantages, we propose an infrared small target detection method that combines the traditional methods with the deep learning methods. First, we construct a multi-stage feature extraction network for guiding the typical multi-scale traditional filtering results to progressively fuse. Secondly, we propose a multi-scale attention supervision module to adjust the semantic consistency of different stages, improving the network generalization ability. Next, a dynamic weight convolution module is utilized to obtain the optimal distribution of grayscale in the neighborhood. Finally, we use the background modelling results to suppress the background, effectively weakening the influence of background clutters and enhancing the target contrast. Experimental results show that the proposed method has good detection results for targets with different scales and signal-to-clutter ratios in a variety of complex scenes. Compared with the typical methods, our method has better detection performance and generalization ability.
29 July 2024 in IEEE Transactions on Geoscience and Remote Sensing
In the field of remote sensing, detection in dimly lit or shadowed areas has traditionally been difficult because of detector noise. Given that noise in real-world images of remote sensing exhibits spatial correlation, existing self-supervised methods encounter difficulties in reconciling the suppression of spatially correlated noise with the preservation of local texture details. To address this challenge, we propose a self-supervised model that combines blind spot feature extraction with diffusion-based texture generation to fine denoising of real-world images under adverse conditions. We first introduce a blind spot feature extraction structure based on the fusion of U-Net with Blind Spot Net (UBSN) and Blind Transformer (BTF). In UBSN, we integrate multi-stride Blind Spot Convolution(BSC) + Dilated Convolution (DC) feature extraction nodes and employ a Reshuffle strategy in skip-layer connections to maintain large-scale blind spot characteristics. Additionally, we design a transformer structure for blind spot between patches to remove the noise with spatial correlations while ensuring global feature acquisition. Subsequently, to restore texture details blurred by the blind spot structures, we introduce a texture generation diffusion structure during model training, achieving a balance between large-scale blind spot characteristics and local rich texture details. Experimental results demonstrate that our approach outperforms other self-supervised denoising methods, even some methods leveraging unpaired images, without the need for parameters related to on-orbit satellite detectors.
1 July 2025 in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
Image stitching aims to align two images from different perspectives. For infrared remote sensing images, the low resolution, lack of strong feature points, and the presence of large texture-less regions make it difficult to achieve effective feature matching and high-quality image stitching results. In the field of remote sensing stitching, the primary challenge is how to effectively extract features, reduce the influence of parallax, and improve registration accuracy. To improve the image stitching performance and obtain parallax-tolerant fine registration results, we propose a two-stage image stitching method based on unsupervised learning. Firstly, in the first stage, we use a multi-level feature extraction network to effectively extract image correlation features, progressively refining the registration from coarse to fine, thus ensuring performance under large-baseline conditions. Secondly, by utilizing a discrete feature detection module in the multi-level network, we remove anomalous feature regions and recombine effective local feature regions, enabling the fusion of detailed features with global features and improving registration accuracy. Finally, in the second stage, an image fine adjustment module is applied to process the image background and foreground, further eliminating parallax artifacts and improving registration accuracy. Compared with existing methods, our method has advantages in both registration accuracy and parallax tolerance. Extensive experiments demonstrate that our method effectively registers and stitches infrared remote sensing images on both the self-built infrared remote sensing dataset and the publicly available UDIS-D dataset, outperforming current state-of-the-art (SoTA) methods in terms of performance. The code is available at https://github.com/Shamfeeding/Image-Stitching.