Joint learning using multiscale attention-enhanced features for remote sensing image scene classification

Single Shot MultiBox Detector (SSD) method using multi-scale feature maps for object detection, showing outstanding performance in object detection task. However, as a one-stage detection method, it’s difficult for SSD methods to quickly notice significant areas of objects in the image. In the SSD network structure, feature maps of different scales are used to independently predict object, and there is a lack of interaction between low-level feature maps and high-level feature maps. In this paper we propose an enhanced SSD method using interactive multi-scale attention features (MA-SSD). Our method uses the attention mechanism to generate attention features of multiple scales and adds it to the original detection branch of the SSD method, which effectively enhances the feature representation ability and improves the detection accuracy. At the same time, the feature of different detection scales interacts with each other, and all the detection branches in our method have a parallel structure, which ensures the detection efficiency. Our proposed method achieves competitive performance on the public dataset PascalVOC.

Remote sensing image scene classification (RSISC) is an active task in the remote sensing community and has attracted great attention due to its wide applications. Recently, the deep convolutional neural networks (CNNs)-based methods have witnessed a remarkable breakthrough in performance of remote sensing image scene classification. However, the problem that the feature representation is not discriminative enough still exists, which is mainly caused by the characteristic of inter-class similarity and intra-class diversity. In this paper, we propose an efficient end-to-end local-global-fusion feature extraction (LGFFE) network for a more discriminative feature representation. Specifically, global and local features are extracted from channel and spatial dimensions respectively, based on a high-level feature map from deep CNNs. For the local features, a novel recurrent neural network (RNN)-based attention module is first proposed to capture the spatial layout information and context information across different regions. Gated recurrent units (GRUs) is then exploited to generate the important weight of each region by taking a sequence of features from image patches as input. A reweighed regional feature representation can be obtained by focusing on the key region. Then, the final feature representation can be acquired by fusing the local and global features. The whole process of feature extraction and feature fusion can be trained in an end-to-end manner. Finally, extensive experiments have been conducted on four public and widely used datasets and experimental results show that our method LGFFE outperforms baseline methods and achieves state-of-the-art results.

Remote sensing image scene classification has a high application value in the agricultural, military, as well as other fields. A large amount of remote sensing data is obtained every day. After learning the new batch data, scene classification algorithms based on deep learning face the problem of catastrophic forgetting, that is, they cannot maintain the performance of the old batch data. Therefore, it has become more and more important to ensure that the scene classification model has the ability of continual learning, that is, to learn new batch data without forgetting the performance of the old batch data. However, the existing remote sensing image scene classification datasets all use static benchmarks and lack the standard to divide the datasets into a number of sequential learning training batches, which largely limits the development of continual learning in remote sensing image scene classification. First, this study gives the criteria for training batches that have been partitioned into three continual learning scenarios, and proposes a large-scale remote sensing image scene classification database called the Continual Learning Benchmark for Remote Sensing (CLRS). The goal of CLRS is to help develop state-of-the-art continual learning algorithms in the field of remote sensing image scene classification. In addition, in this paper, a new method of constructing a large-scale remote sensing image classification database based on the target detection pretrained model is proposed, which can effectively reduce manual annotations. Finally, several mainstream continual learning methods are tested and analyzed under three continual learning scenarios, and the results can be used as a baseline for future work.

Remote sensing image scene classification is a challenging task. With the development of deep learning, methods based on convolutional neural networks (CNNs) have made great achievements in remote sensing image scene classification. Since the training of a CNN requires a large number of labeled samples, a generative adversarial network (GAN) for sample generation represents a new opportunity to solve the problem of the limited samples. However, most of the existing GAN-based sample generation methods can only generate unlabeled samples, instead of samples labeled with the corresponding scene category. In this article, to solve the problem, a supervised progressive growing generative adversarial network (SPG-GAN) is proposed for remote sensing image scene classification. The proposed method can generate labeled samples for the remote sensing image scene classification, significantly improving the classification accuracy in the case of limited samples. The SPG-GAN method has two main improvements. First, a conditional generative framework for labeled samples is proposed, in which the label information is added in the channel dimension as the input. By considering the constraints of the label information in the loss function, the network can be trained in the direction of a specific category. As a result, the network can generate remote sensing image scene classification samples with label categories. Second, a progressive growing sample generation method is introduced. In order to ensure that the generated samples have more spatial details, they are generated by progressively adding modules to the generator and discriminator, thereby ensuring that the generated sample is of better quality. After testing on two benchmark datasets and carrying out a large-scale experiment in the central area of the city of Wuhan in China, it was found that the proposed method can obtain a superior scene classification accuracy in the case of limited samples.

Remote sensing image scene classification plays an important role in a wide range of applications and hence has been receiving remarkable attention. During the past years, significant efforts have been made to develop various datasets or present a variety of approaches for scene classification from remote sensing images. However, a systematic review of the literature concerning datasets and methods for scene classification is still lacking. In addition, almost all existing datasets have a number of limitations, including the small scale of scene classes and the image numbers, the lack of image variations and diversity, and the saturation of accuracy. These limitations severely limit the development of new approaches especially deep learning-based methods. This paper first provides a comprehensive review of the recent progress. Then, we propose a large-scale dataset, termed "NWPU-RESISC45", which is a publicly available benchmark for REmote Sensing Image Scene Classification (RESISC), created by Northwestern Polytechnical University (NWPU). This dataset contains 31,500 images, covering 45 scene classes with 700 images in each class. The proposed NWPU-RESISC45 (i) is large-scale on the scene classes and the total image number, (ii) holds big variations in translation, spatial resolution, viewpoint, object pose, illumination, background, and occlusion, and (iii) has high within-class diversity and between-class similarity. The creation of this dataset will enable the community to develop and evaluate various data-driven algorithms. Finally, several representative methods are evaluated using the proposed dataset and the results are reported as a useful baseline for future research.

Due to the influence of the shooting angle of view and the flight height, the images taken by UAV often have complex backgrounds and contain a large number of small and unevenly distributed objects. In order to solve the problem that it is difficult to accurately locate and recognize small objects in UAV images under complex backgrounds, this paper proposes an multi-scale feature fusion algorithm D-A-FS SSD (Dilated-Attention-Feature Fusion SSD) based on the combination of dilated convolution and attention mechanism. In the process of feature extraction, the receptive field of the feature is expanded through the dilated convolution, which improves the network's feature expression of object distribution and scale difference information. And a attention network is used in our method to effectively suppresse the background information. In the multi-scale detection stage, our method fuses the low-level feature map responsible for detecting small objects with the high-level feature map which have much higher semantic information to improve the recognition accuracy of small objects. Experimental results show that our method effectively improves the accuracy of UAV image object detection.

With geospatial intelligence research developing rapidly, automatic road extraction is becoming a fundamental and challenging task. Due to the special geometric structure and spectral information of road networks, existing methods suffer from incomplete and fractured results. In this work, a novel road extraction convolutional neural network, incorporating the road boundary details and road junction information via a dual-branch multi-task structure, is proposed to learn synergistic feature representations and strengthen road connectivity. Firstly, a BiFPN-based feature aggregation module is utilised to bridge the semantic gap between low-level and high-level feature maps, allowing multi-scale spatial details to be fully fused. Secondly, the boundary auxiliary branch, using a U-shaped network with a spatial-channel attention module, captures residential information for the backbone to enhance the subtleties of road edges. Thirdly, the node inferring branch models the road junction position jointly with the road surface, aiming to strengthen the topology structure and reduce the fragmented road segments. We perform experiments on three diverse road datasets, namely the DeepGlobe dataset, Massachusetts dataset, and SpaceNet dataset. The results demonstrate that our model shows an overall performance improvement over some SOTA algorithms and the IoU indicator achieves 3.86%, 0.79%, and 1.71% improvements over Unet on the three datasets, respectively. • A novel Geometric Prior Assisted Neural Network (GA-Net) is proposed for automatic road extraction. The boundary information is introduced to the model, which strengthens the geometric details of road edges. The road junction information is jointly learned with the road surface information, which reduces the missing gap between road segments. • A feature fusion module is designed to bridge the semantic gap between multi-scale features maps. An attention module is designed to strengthen the road geometric information. The two light modules can be easily ensemble with other models. • We achieves better results than some SOTA algorithms for three popular datasets. And results of our model shows better completeness and connectivity.

This work proposes a remote sensing image scene classification method based on stackable attention structure. This method utilizes a stackable attention structure to represent the feature information in remote sensing images as a linear function, changing the previous processing methods for global and local information in remote sensing images to avoid the need for memory intensive attention maps. Through experiments, it has been shown that the scene classification method we proposed based on stackable attention structure has higher accuracy and smaller model size than ResNet152. This provides a new approach for constructing remote sensing image scene classification models, especially for processing large-sized remote sensing images.

Remote sensing image scene classification faces challenges, such as the difference in semantic granularity of different scene categories and the imbalance of the number of samples, which cause the wrong features learning for deep convolutional networks (DCNs). This article proposes a multiple granularity semantic learning network (MGSN), including multiple granularity semantic learning (MGSL) and nonuniform sampling augmentation (NUA) modules. Specifically, the MGSL module makes full use of different granularities of semantic information of scenes, guiding the network to learn global and local features simultaneously. And, the relationship between semantic features of different granularity has been explored, based on which the learning of coarse-grained features helps to improve the learning of fine-grained semantic features. It shows that learning fine-grain semantics can inhibit learning coarse-grain semantic features. The NUA module combines sampling and sample augmentation to balance the sample distribution, which can avoid overfitting caused by oversampling. The proposed MGSN achieved state-of-the-art classification accuracy on two large-scale remote sensing image scene classification datasets, Million-AID and NWPU-RESISC45. Under 10<inline-formula><tex-math notation="LaTeX">$\%$</tex-math></inline-formula> and 20<inline-formula><tex-math notation="LaTeX">$\%$</tex-math></inline-formula> training samples of the NWPU-RESISC45 dataset, MGSN achieves 91.92<inline-formula><tex-math notation="LaTeX">$\%$</tex-math></inline-formula> and 94.33<inline-formula><tex-math notation="LaTeX">$\%$</tex-math></inline-formula> top-1 accuracy, respectively. In experiments conducted on the Million-AID dataset, the proposed MGSN performed best among 18 DCNs. In comparison to the baseline, FixEfficientNet, MGSN improved the accuracy of top-1 and top-5 by 10.63<inline-formula><tex-math notation="LaTeX">$\%$</tex-math></inline-formula> and 5.47<inline-formula><tex-math notation="LaTeX">$\%$</tex-math></inline-formula>, respectively, with low complexity costs.

The performance of deep neural networks in object detection is impressive, however, direct use of existingalgorithms may not obtain satisfactory resultsdue to the characteristics of underwater images. In this paper, a multi-scale feature fusion Single Shot Multibox Detector (MFFSSD) based on the SSD algorithm is proposed to improve the accuracy of underwater object detection. There are two main improvements: introducing attention module and perform multiscale feature fusion. By introducing attention module, the features of the input image obtained through convolution are weighted in both channel and space aspect, so as to highlight the useful informationto overcomes the problem of small and blur objects which are common in underwater images. By extracting a lower level feature map forward, more details of the lower level image can be obtained to adapt to the underwater object with simple shape, so that low-level features have a higher contribution to the recognition object; at the same time, the high-level feature map is enlarged by using deconvolution, and then fused with the low-level feature map to introduce more contextual information. The fusion of the features extracted from the shallow and high level networks ensures the efficiency of the features to the greatest extent. Experiments on URPC show that MFFSSD significantly improved the underwater object recognitionaccuracy compared to the original algorithm.

Unsupervised domain adaptation (UDA) is an important solution to reduce the bias between the labeled source domain and the unlabeled target domain. It has attracted more attention for optical remote sensing image scene classification and retrieval. Currently, most of the previous work is devoted to closed-set UDA. In fact, the target domain often contains unknown classes. Moreover, some open UDA methods mine structural information of the target domain directly from the type knowledge of the source domain, and less directly from the unlabeled data of the target domain. In this paper, we propose a new self-supervised-driven open-set UDA method combining contrastive self-supervised learning with consistency self-training for optical remote sensing scene classification and retrieval. Specifically, a contrastive self-supervised learning network is introduced to learn discriminative features from the unlabeled target domain data. Moreover, a novel open-set class learning module is developed based on two-level confidence rules and the consistency self-training strategy, which can obtain reliable unknown class samples for co-training. Finally, an open-set dataset including six cross-domain scenarios is constructed based on three public datasets and several experiments are conducted with eleven state-of-the-art domain adaptation methods. Experimental results demonstrate that our proposed method achieves superior performances on the six open-set cross-domain scenarios in both scene classification and retrieval. Especially, our method improves the overall classification accuracies by 9.72% to 24.06% and improve mean average retrieval precisions by 8.06% to 16.21% on the complex UCMD (source domain) → NWPU (target domain) scenario, compared with the other eleven state-of-the-art methods.

The classification of aerial scenes has been extensively studied as the basic work of remote sensing image processing and interpretation. However, the performance of remote sensing image scene classification based on deep neural networks is limited by the number of labeled samples. In order to alleviate the demand for massive labeled samples, various methods have been proposed to apply semi-supervised learning to train the classifier using labeled and unlabeled samples. However, considering the complex contextual relationship and huge spatial differences, the existing semi-supervised learning methods bring different degrees of incorrectly labeled samples when pseudo-labeling unlabeled data. In particular, when the number of labeled samples is small, it affects the generalization performance of the model. In this article, we propose a novel semi-supervised learning method with early labeled and small loss selection. First, the model learns the characteristics of simple samples in the early stage and uses multiple early models to screen out a small number of unlabeled samples for pseudo-labeling based on this characteristic. Then, the model is trained in a semi-supervised manner by combining labeled samples, pseudo-labeled samples, and unlabeled samples. In the training process of the model, small loss selection is used to further eliminate some of the noisy labeled samples to improve the recognition accuracy of the model. Finally, in order to verify the effectiveness of the proposed method, it is compared with several state-of-the-art semi-supervised classification methods. The results show that when there are only a few labeled samples in remote sensing image scene classification, our method is always better than previous methods.

The effective use of multi-scale features remains an open problem for object detection tasks. Recently, proposed object detectors have usually used Feature Pyramid Networks (FPN) to fuse multi-scale features. Since Feature Pyramid Networks use a relatively simple feature map fusion approach, it can lead to the loss or misalignment of semantic information in the fusion process. Several works have demonstrated that using a bottom-up structure in a Feature Pyramid Network can shorten the information path between lower layers and the topmost feature, allowing an adequate exchange of semantic information from different layers. We further enhance the bottom-up path by proposing a multi-scale residual aggregation Feature Pyramid Network (MSRA-FPN), which uses a unidirectional cross-layer residual module to aggregate features from multiple layers bottom-up in a triangular structure to the topmost layer. In addition, we introduce a Residual Squeeze and Excitation Module to mitigate the aliasing effects that occur when features from different layers are aggregated. MSRA-FPN enhances the semantic information of the high-level feature maps, mitigates the information decay during feature fusion, and enhances the detection capability of the model for large objects. It is experimentally demonstrated that our proposed MSRA-FPN improves the performance of the three baseline models by 0.5–1.9% on the PASCAL VOC dataset and is also quite competitive with other state-of-the-art FPN methods. On the MS COCO dataset, our proposed method can also improve the performance of the baseline model by 0.8% and the baseline model’s performance for large object detection by 1.8%. To further validate the effectiveness of MSRA-FPN for large object detection, we constructed the Thangka Figure Dataset and conducted comparative experiments. It is experimentally demonstrated that our proposed method improves the performance of the baseline model by 2.9–4.7% on this dataset and can reach up to 71.2%.

Fine-tuning of pretrained convolutional neural networks (CNNs) has been proven to be an effective strategy for remote sensing image scene classification, particularly when a limited number of labeled data sets are available for training purposes. However, such a fine-tuning process often needs that the input images are resized into a fixed size to generate input vectors of the size required by fully connected layers (FCLs) in the pretrained CNN model. Such a resizing process often discards key information in the scenes and thus deteriorates the classification performance. To address this issue, in this paper, we introduce a scale-free CNN (SF-CNN) for remote sensing scene classification. Specifically, the FCLs in the CNN model are first converted into convolutional layers, which not only allow the input images to be of arbitrary sizes but also retain the ability to extract discriminative features using a traditional sliding-window-based strategy. Then, a global average pooling (GAP) layer is added after the final convolutional layer so that input images of arbitrary size can be mapped to feature maps of uniform size. Finally, we utilize the resulting feature maps to create a new FCL that is fed to a softmax layer for final classification. Our experimental results conducted using several real data sets demonstrate the superiority of the proposed SF-CNN method over several well-known classification methods, including pretrained CNN-based ones.

Person re-identification (re-ID), which aims at spotting a person of interest across multiple camera views, has gained more and more attention in computer vision community. In this paper, we propose a novel deep Siamese architecture based on convolutional neural network (CNN) and multi-level similarity perception. According to the distinct characteristics of diverse feature maps, we effectively apply different similarity constraints to both low-level and high-level feature maps, during training stage. Therefore, our network can efficiently learn discriminative feature representations at different levels, which significantly improves the re-ID performance. Besides, our framework has two additional benefits. Firstly, classification constraints can be easily incorporated into the framework, forming a unified multi-task network with similarity constraints. Secondly, as similarity comparable information has been encoded in the network's learning parameters via back-propagation, pairwise input is not necessary at test time. That means we can extract features of each gallery image and build index in an off-line manner, which is essential for large-scale real-world applications. Experimental results on multiple challenging benchmarks demonstrate that our method achieves splendid performance compared with the current state-of-the-art approaches.