Few-shot object detection is a recently emerging branch in the field of computer vision. Recent research studies have proposed several effective methods for object detection with few samples. However, their performances are limited when applied to remote sensing images. In this article, we specifically analyze the characteristics of remote sensing images and propose a few-shot fine-tuning network with a shared attention module (SAM) to adapt to detecting remote sensing objects, which have large size variations. In our SAM, multi-attention maps are computed in the base training stage and shared with the feature extractor in the few-shot fine-tuning stage as prior knowledge to help better locate novel class objects with few samples. Moreover, we design a new few-shot fine-tuning stage with a balanced fine-tuning strategy (BFS), which helps in mitigating the severe imbalance between the number of novel class samples and base class samples caused by the few-shot settings to improve the classification accuracy. We have conducted experiments on two remote sensing datasets (NWPU VHR-10 and DIOR), and the excellent results demonstrate that our method makes full use of the advantages of few-shot learning and the characteristics of remote sensing images to enhance the few-shot detection performance.

As the application scenarios of remote sensing imagery (RSI) become richer, the task of ship detection from an overhead perspective is of great significance. Compared with traditional methods, the use of deep learning ideas has more prospects. However, the Convolutional Neural Network (CNN) has poor resistance to sample differences in detection tasks, and the huge differences in the image environment, background, and quality of RSIs affect the performance for target detection tasks; on the other hand, upsampling or pooling operations result in the loss of detailed information in the features, and the CNN with outstanding results are often accompanied by a high computation and a large amount of memory storage. Considering the characteristics of ship targets in RSIs, this study proposes a detection framework combining an image enhancement module with a dense feature reuse module: (1) drawing on the ideas of the generative adversarial network (GAN), we designed an image enhancement module driven by object characteristics, which improves the quality of the ship target in the images while augmenting the training set; (2) the intensive feature extraction module was designed to integrate low-level location information and high-level semantic information of different resolutions while minimizing the computation, which can improve the efficiency of feature reuse in the network; (3) we introduced the receptive field expansion module to obtain a wider range of deep semantic information and enhance the ability to extract features of targets were at different sizes. Experiments were carried out on two types of ship datasets, optical RSI and Synthetic Aperture Radar (SAR) images. The proposed framework was implemented on classic detection networks such as You Only Look Once (YOLO) and Mask-RCNN. The experimental results verify the effectiveness of the proposed method.

The varying Polarimetric Orientation Angles (POAs) result in scattering diversity, leading to ambiguity in the interpretation of Polarimetric Synthetic Aperture Radar (PolSAR) images. Exploring the scattering characteristics in the Polarimetric Rotation Domain (PRD) and the complementary features can help overcome the ambiguity. To address this, we propose a novel PolSAR image classification algorithm called Learning Scattering similarity and Texture-based attention with Convolutional Neural Networks (LSTCNN). Three strategies are included in the proposed method. Firstly, a pixel-level Scattering Similarity Learning (SSL) module is proposed to analyze the scattering components of radar targets by learning the mapping from PolSAR data in the PRD to typical scattering models, with rotation angles as learnable parameters to utilize scattering diversity and avoid ambiguity. Secondly, a neighborhood-level Texture-based Attention (TA) module is proposed to learn the spatially enhanced features of PolSAR images, with the attention module design guided by the physical meaning of texture and consideration of channel and position importance. Finally, the proposed LSTCNN, which includes the SSL module, TA module, and classification module, combines pixel-level scattering features in the PRD and neighborhood-level texture features to increase the discriminability of features. The experimental results on three PolSAR images acquired by AIRSAR and E-SAR demonstrate the robustness and excellence of LSTCNN.

With the rapid development of artificial intelligence, how to take advantage of deep learning and big data to classify polarimetric synthetic aperture radar (PolSAR) imagery is a hot topic in the field of remote sensing. As a key step for PolSAR image classification, feature extraction technology based on target decomposition is relatively mature, and how to extract discriminative spatial features and integrate these features with polarized information to maximize the classification accuracy is the core issue. In this context, this paper proposes a PolSAR image classification algorithm based on fully convolutional networks (FCNs) and a manifold graph embedding model. First, to describe different types of land objects more comprehensively, various polarized features of PolSAR images are extracted through seven kinds of traditional decomposition methods. Afterwards, drawing on transfer learning, the decomposed features are fed into multiple parallel and pre-trained FCN-8s models to learn deep multi-scale spatial features. Feature maps from the last layer of each FCN model are concatenated to obtain spatial polarization features with high dimensions. Then, a manifold graph embedding model is adopted to seek an effective and compact representation for spatially polarized features in a manifold subspace, simultaneously removing redundant information. Finally, a support vector machine (SVM) is selected as the classifier for pixel-level classification in a manifold subspace. Extensive experiments on three PolSAR datasets demonstrate that the proposed algorithm achieves a superior classification performance.

Due to device limitations, small networks are necessary for some real-world scenarios, such as satellites and micro-robots. Therefore, the development of a network with both good performance and small size is an important area of research. Deep networks can learn well from large amounts of data, while manifold networks have outstanding feature representation at small sizes. In this paper, we propose an approach that exploits the advantages of deep networks and shallow Grassmannian manifold networks. Inspired by knowledge distillation, we use the information learned from convolutional neural networks to guide the training of the manifold networks. Our approach leads to a reduction in model size, which addresses the problem of deploying deep learning on resource-limited embedded devices. Finally, a series of experiments were conducted on four remote sensing scene classification datasets. The method in this paper improved the classification accuracy by 2.31% and 1.73% on the UC Merced Land Use and SIRIWHU datasets, respectively, and the experimental results demonstrate the effectiveness of our approach.

Recently, object detection in natural images has made a breakthrough, but it is still challenging in oriented ship detection for remote sensing imagery. Considering some limitations in this task, such as uncertain ship orientation, unspecific features for locating and classification in the complex optical environment, and multiplicative speckle interference of synthetic aperture radar (SAR), we propose an oriented ship detector based on the pairwise branch detection head and adaptive SAR feature enhancement. The details are as follows: (1) Firstly, the ships with arbitrary directions are described with a rotated ground truth, and an oriented region proposal network (ORPN) is designed to study the transformation from the horizontal region of interest to the rotated region of interest. The ORPN effectively improved the quality of the candidate area while only introducing a few parameters. (2) In view of the existing algorithms that tend to perform classification and regression prediction on the same output feature, this paper proposes a pairwise detection head (PBH) to design parallel branches to decouple classification and locating tasks, so that each branch can learn more task-specific features. (3) Inspired by the ratio-of-average detector in traditional SAR image processing, the SAR edge enhancement (SEE) module is proposed, which adaptively enhances edge pixels, and the threshold of the edge is learned by the channel-shared adaptive thresholds block. Experiments were carried out on both optical and SAR datasets. In the optical dataset, PBH combined with ORPN improved recall by 5.03%, and in the SAR dataset, the overall method achieved a maximum F1 score improvement of 6.07%; these results imply the validity of our method.