Search

Supervised deep neural networks (DNNs) have been extensively used in diverse tasks. Generally, training such DNNs with superior performance requires a large amount of labeled data. However, it is time-consuming and expensive to manually label the data, especially for tasks in remote sensing, e.g., change detection. The situation motivates us to resort to the existing related images with labels, from which the concept of change can be adapted to new images. However, the distributions of the related labeled images (source domain) and unlabeled new images (target domain) are similar but not identical. It impedes a change detection model learned from source domains being well applied to the target domain. In this paper, we propose a transferred deep learning-based change detection framework to solve this problem. It consists of pretraining and fine-tuning stages. In the pretraining process, we propose two tasks to be learned simultaneously, namely, change detection for the source domain with labels and reconstruction of the unlabeled target data. The auxiliary task aims to reconstruct the difference image (DI) for the target domain. DI is an effective feature, such that the auxiliary task is of much relevance to change detection. The lower layers are shared between these two tasks in the training process. It mitigates the distribution discrepancy between the source and target domains and makes the concept of change from the source domain adapt to the target domain. In addition, we evaluate three modes of the U-net architecture to merge the information for a pair of patches. To fine-tune the change detection network (CDN) for the target domain, two strategies are exploited to select the pixels that have a high possibility of being correctly classified by an unsupervised approach. The proposed method demonstrates an excellent capacity for adapting the concept of change from the source domain to the target domain. It outperforms the state-of-the-art change detection methods via experimental results on real remote sensing data sets. [ABSTRACT FROM AUTHOR]

Deep convolutional neural networks (DCNNs) are driving progress in object detection of high-resolution remote sensing images. Region proposal generation, as one of the key steps in object detection, has also become the focus of research. High-resolution remote sensing images usually contain various sizes of objects and complex background, small objects are easy to miss or be mis-identified in object detection. If the recall rate of region proposal of small objects and multi-scale objects can be improved, it will bring an improvement on the performance of the accuracy in object detection. Spatial attention is the ability to focus on local features in images and can improve the learning efficiency of DCNNs. This study proposes a multi-scale spatial attention region proposal network (MSA-RPN) for high-resolution optical remote sensing imagery. The MSA-RPN is an end-to-end deep learning network with a backbone network of ResNet. It deploys three novel modules to fulfill its task. First, the Scale-specific Feature Gate (SFG) focuses on features of objects by processing multi-scale features extracted from the backbone network. Second, the spatial attention-guided model (SAGM) obtains spatial information of objects from the multi-scale attention maps. Third, the Selective Strong Attention Maps Model (SSAMM) adaptively selects sliding windows according to the loss values from the system's feedback, and sends the windowed samples to the spatial attention decoder. Finally, the candidate regions and their corresponding confidences can be obtained. We evaluate the proposed network in a public dataset LEVIR and compare with several state-of-the-art methods. The proposed MSA-RPN yields a higher recall rate of region proposal generation, especially for small targets in remote sensing images. [ABSTRACT FROM AUTHOR]

As a fundamental but challenging task in the interpretation of remote sensing images, scene classification plays an important role in various applications and has become an active research topic. Many previous works have demonstrated the remarkable performance of the deep convolutional neural networks (CNNs) for remote sensing scene classification. However, the progress made by CNN-based methods for scene classification has gradually reached saturation in recent years, due to the serious dependence on the pretrained CNN models, the limitations of manually designed network architecture and the disadvantages of existing data sets. In this article, a new paradigm to automatically design a suitable CNN architecture for scene classification is investigated. We propose an efficient architecture search framework to discover optimal network architectures in continuous search space with the gradient-based optimization method. Our framework consists of two stages: the search phase and the evaluation phase. During the search process, a greedy and progressive search strategy is introduced to search network building blocks (i.e., cells) through bilevel optimization. Besides, we propose a simple architecture regularization scheme to further improve the search efficiency and the robustness of discovered architectures. After the search process, the optimal cell architectures are determined and then repeatedly stacked to construct the final network for evaluation. For the data set, we propose a mergence strategy to build a new large-scale remote sensing scene image data set that contains rich scene categories and image diversity, making it feasible to find a new CNN model with strong generalization ability for scene classification. Extensive experiments demonstrate the efficiency of the proposed search strategies and the impressive classification performance of searched CNN architectures on seven public benchmark data sets, including four large-scale data sets and three small-scale data sets. [ABSTRACT FROM AUTHOR]

Remote sensing image scene classification has become more and more popular in recent years. As we all know, it is very difficult and time-consuming to obtain a large number of manually labeled remote sensing images. Therefore, few-shot scene classification of remote sensing images has become an urgent and important research task. Fortunately, the recently proposed deep nearest neighbor neural network (DN4) has made a breakthrough in few-shot classification. However, due to the complex background in remote sensing images, DN4 is easily affected by irrelevant local features, so DN4 cannot be directly applied in remote sensing images. For this reason, a deep nearest neighbor neural network based on attention mechanism (DN4AM) is proposed to solve the few-shot scene classification task of remote sensing images in this paper. Scene class-related attention maps are used in our method to reduce interference from scene-semantic irrelevant objects to improve the classification accuracy. Three remote sensing image datasets are used to verify the performance of our method. Compared with several state-of-the-art methods, including MatchingNet, RelationNet, MAML, Meta-SGD and DN4, our method achieves promising results in the few-shot scene classification of remote sensing images. [ABSTRACT FROM AUTHOR]

Deep descriptors have advantages over handcrafted descriptors on local image patch matching. However, due to the complex imaging mechanism of remote sensing images and the significant differences in appearance between multi-modal images, existing deep learning descriptors are unsuitable for multi-modal remote sensing image registration directly. To solve this problem, this article proposes a deep feature correlation learning network (Cnet) for multi-modal remote sensing image registration. First, Cnet builds a feature learning network based on the deep convolutional network with the attention learning module, to enhance feature representation by focusing on meaningful features. Second, this article designs a novel feature correlation loss function for Cnet optimization. It focuses on the relative feature correlation between matching and nonmatching samples, which can improve the stability of network training and decrease the risk of overfitting. In addition, the proposed feature correlation loss with a scale factor can further enhance network training and accelerate network convergence. Extensive experimental results on image patch matching (Brown, HPatches), cross-spectral image registration (VIS-NIR), multi-modal remote sensing image registration, and single-modal remote sensing image registration have demonstrated the effectiveness and robustness of the proposed method. [ABSTRACT FROM AUTHOR]

During the past decades, convolutional neural network (CNN)-based models have achieved notable success in remote sensing image classification due to their powerful feature representation ability. However, the lack of explainability during the decision-making process is a common criticism of these high-capacity networks. Local explanation methods that provide visual saliency maps have attracted increasing attention as a means to surmount the barrier of explainability. However, the vast majority of research is conducted on the last convolutional layer, where the salient regions are unintelligible for partial remote sensing images, especially scenes that contain plentiful small targets or are similar to the texture image. To address these issues, we propose a novel framework called Prob-POS, which consists of the class-activation map based on the probe network (Prob-CAM) and the weighted probability of occlusion (wPO) selection strategy. The proposed probe network is a simple but effective architecture to generate elaborate explanation maps and can be applied to any layer of CNNs. The wPO is a quantified metric to evaluate the explanation effectiveness of each layer for different categories to automatically pick out the optimal explanation layer. Variational weights are taken into account to highlight the high-scoring regions in the explanation map. Experimental results on two publicly available datasets and three prevalent networks demonstrate that Prob-POS improves the faithfulness and explainability of CNNs on remote sensing images. [ABSTRACT FROM AUTHOR]

Small target detection is a challenging task in very-high-resolution (VHR) optical remote sensing imagery, because small targets occupy a minuscule number of pixels and are easily disturbed by backgrounds or occluded by others. Although current convolutional neural network (CNN)-based approaches perform well when detecting normal objects, they are barely suitable for detecting small ones. Two practical problems stand in their way. First, current CNN-based approaches are not specifically designed for the minuscule size of small targets (~15 or ~10 pixels in extent). Second, no well-established data sets include labeled small targets and establishing one from scratch is labor-intensive and time-consuming. To address these two issues, we propose an approach that combines Sig-NMS-based Faster R-CNN with transfer learning. Sig-NMS replaces traditional non-maximum suppression (NMS) in the stage of region proposal network and decreases the possibility of missing small targets. Transfer learning can effectively label remote sensing images by automatically annotating both object classes and object locations. We conduct an experiment on three data sets of VHR optical remote sensing images, RSOD, LEVIR, and NWPU VHR-10, to validate our approach. The results demonstrate that the proposed approach can effectively detect small targets in the VHR optical remote sensing images of about $10\times 10$ pixels and automatically label small targets as well. In addition, our method presents better mean average precisions than other state-of-the-art methods: 1.5% higher when performing on the RSOD data set, 17.8% higher on the LEVIR data set, and 3.8% higher on NWPU VHR-10. [ABSTRACT FROM AUTHOR]

With the development of remote sensing imaging technology, remote sensing images with high-resolution and complex structure can be acquired easily. The classification of remote sensing images is always a hot and challenging problem. In order to improve the performance of remote sensing image classification, we propose an adaptive multiscale deep fusion residual network (AMDF-ResNet). The AMDF-ResNet consists of a backbone network and a fusion network. The backbone network including several residual blocks generates multiscale hierarchy features, which contain semantic information from low to high levels. In the fusion network, the adaptive feature fusion module proposed can emphasize useful information and suppress useless information by learning the weights, which represent the importance of the features. The AMDF-ResNet can make full use of the multiscale hierarchy features and the extracted feature is discriminative. In addition, we propose a samples selection method named important samples selection strategy (ISSS). Based on superpixels segmentation result, gradient information and spatial distribution are used as two references to determine the selection numbers and select samples. Compared with the random selection strategy, training samples selected by ISSS are more representative and diverse. The experimental results on four data sets demonstrate that the AMDF-ResNet and ISSS are effective. [ABSTRACT FROM AUTHOR]

Object detection has played an important role in the remote sensing (RS) image understanding domain, but it still suffers from challenges, including the complexity of the image background, the variety of objects and differences of target scales. Although many studies applying deep learning methods to object detection have been presented, effective methods of multiscale object detection in RS images are still lacking. In this paper, we propose a novel object detection approach based on the Multiscale-SELU-DenseNet (MSE-DenseNet) and the dynamic anchor assignment (DAA) strategy. We fuse a highly modified DenseNet and a feature pyramid network into the MSE-DenseNet to extract multi-level feature descriptions of RS images, and then we use them to detect objects on various scales. To further improve the detection performance, we propose a DAA strategy for assigning prior anchor boxes. The experimental results show that our proposed method outperforms state-of-the-art detection methods on the VHR-10 dataset and maintains a real-time detection speed. [ABSTRACT FROM AUTHOR]

Object detection in remote sensing images has been widely used in military and civilian fields and is a challenging task due to the complex background, large-scale variation, and dense arrangement in arbitrary orientations of objects. In addition, existing object detection methods rely on the increasingly deeper network, which increases a lot of computational overhead and parameters, and is unfavorable to deployment on the edge devices. In this paper, we proposed a lightweight keypoint-based oriented object detector for remote sensing images. First, we propose a semantic transfer block (STB) when merging shallow and deep features, which reduces noise and restores the semantic information. Then, the proposed adaptive Gaussian kernel (AGK) is adapted to objects of different scales, and further improves detection performance. Finally, we propose the distillation loss associated with object detection to obtain a lightweight student network. Experiments on the HRSC2016 and UCAS-AOD datasets show that the proposed method adapts to different scale objects, obtains accurate bounding boxes, and reduces the influence of complex backgrounds. The comparison with mainstream methods proves that our method has comparable performance under lightweight. [ABSTRACT FROM AUTHOR]

High-resolution optical remote sensing image classification is an important research direction in the field of computer vision. It is difficult to extract the rich semantic information from remote sensing images with many objects. In this paper, a multiscale self-adaptive attention network (MSAA-Net) is proposed for the optical remote sensing image classification, which includes multiscale feature extraction, adaptive information fusion, and classification. In the first part, two parallel convolution blocks with different receptive fields are adopted to capture multiscale features. Then, the squeeze process is used to obtain global information and the excitation process is used to learn the weights in different channels, which can adaptively select useful information from multiscale features. Furthermore, the high-level features are classified by many residual blocks with an attention mechanism and a fully connected layer. Experiments were conducted using the UC Merced, NWPU, and the Google SIRI-WHU datasets. Compared to the state-of-the-art methods, the MSAA-Net has great effect and robustness, with average accuracies of 94.52%, 95.01%, and 95.21% on the three widely used remote sensing datasets. [ABSTRACT FROM AUTHOR]

Traditional change detection (CD) methods operate in the simple image domain or hand-crafted features, which has less robustness to the inconsistencies (e.g., brightness and noise distribution, etc.) between bitemporal satellite images. Recently, deep learning techniques have reported compelling performance on robust feature learning. However, generating accurate semantic supervision that reveals real change information in satellite images still remains challenging, especially for manual annotation. To solve this problem, we propose a novel self-supervised representation learning method based on temporal prediction for remote sensing image CD. The main idea of our algorithm is to transform two satellite images into more consistent feature representations through a self-supervised mechanism without semantic supervision and any additional computations. Based on the transformed feature representations, a better difference image (DI) can be obtained, which reduces the propagated error of DI on the final detection result. In the self-supervised mechanism, the network is asked to identify different sample patches between two temporal images, namely, temporal prediction. By designing the network for the temporal prediction task to imitate the discriminator of generative adversarial networks, the distribution-aware feature representations are automatically captured and the result with powerful robustness can be acquired. Experimental results on real remote sensing data sets show the effectiveness and superiority of our method, improving the detection precision up to 0.94–35.49%. [ABSTRACT FROM AUTHOR]