Gaofen-1 Satellite Research Articles

D-FusionNet: road extraction from remote sensing images using dilated convolutional block

ABSTRACT Deep learning techniques have been applied to extract road areas from remote sensing images, leveraging their efficient and intelligent advantages. However, the contradiction between the effective receptive field and coverage range, as well as the conflict between the depth of the network and the density of geographic information, hinders further improvement in extraction accuracy. To address these challenges, we propose a novel semantic segmentation network called D-FusionNet. D-FusionNet integrates the Dilated Convolutional Block (DCB) module, which serves as a technique for expanding the receptive field and mitigating feature loss, resembling a residual mechanism during encoding. We evaluate the extraction capability of D-FusionNet using GF-2 (Gaofen-2) satellite datasets and Massachusetts aerial photography datasets. The experimental results demonstrate that D-FusionNet performs well in road extraction tasks. Compared to FCN, UNet, LinkNet, D-LinkNet, and FusionNet, D-FusionNet achieves an average improvement of 5.35% in F1-score, 7.12% in IoU (Intersection over Union), and 5.61% in MCC (Matthews Correlation Coefficient) on the GF-2 dataset. For the Massachusetts dataset, there is an average improvement of 2.48% in F1-score, 3.25% in IoU, and 2.25% in MCC. This study provides valuable support for road extraction from remote sensing images.

Crop Type Identification Using High-Resolution Remote Sensing Images Based on an Improved DeepLabV3+ Network

Remote sensing technology has become a popular tool for crop classification, but it faces challenges in accurately identifying crops in areas with fragmented land plots and complex planting structures. To address this issue, we propose an improved method for crop identification in high-resolution remote sensing images, achieved by modifying the DeepLab V3+ semantic segmentation network. In this paper, the typical crop area in the Jianghuai watershed is taken as the experimental area, and Gaofen-2 satellite images with high spatial resolutions are used as the data source. Based on the original DeepLab V3+ model, CI and OSAVI vegetation indices are added to the input layers, and MobileNet V2 is used as the backbone network. Meanwhile, the upper sampling layer of the network is added, and the attention mechanism is added to the ASPP and the upper sampling layers. The accuracy verification of the identification results shows that the MIoU and PA of this model in the test set reach 85.63% and 95.30%, the IoU and F1_Score of wheat are 93.76% and 96.78%, and the IoU and F1_Score of rape are 74.24% and 85.51%, respectively. The identification accuracy of this model is significantly better than that of the original DeepLab V3+ model and other related models. The proposed method in this paper can accurately extract the distribution information of wheat and rape from high-resolution remote sensing images. This provides a new technical approach for the application of high-resolution remote sensing images in identifying wheat and rape.

Generation of a 16 m/10-day fractional vegetation cover product over China based on Chinese GaoFen-1 observations: method and validation

ABSTRACTAs China has recently launched the GaoFen-1 satellite (GF-1) carrying on the wide-field view (WFV) sensor, it is a challenging task to make full use of its observations to produce the fractional vegetation cover (FVC). In light of this, our study presents a comprehensive algorithm to generate a 16 m/10-day FVC product by considering the vegetation types characteristics. For forests, considering the foliage clumping effect, FVC was estimated from the gap probability theory using GF-1 leaf area index (LAI) and clumping index (CI) as a priori knowledge; for non-forests, FVC was estimated from the dimidiate pixel model using GF-1 normalized difference vegetation index (NDVI). The performance of GF-1 FVC from 2018 to 2020 was evaluated using FVC ground measurements obtained from 7 sites for crops, grasslands, and forests in China. The direct validation indicated that the performance of the FVC product was satisfactory, as evidenced by R2 = 0.55, RMSE = 0.15 and BIAS = 0.01 for all vegetation types. Furthermore, the GF-1 FVC exhibited better performance compared to the GEOV3 FVC at a spatial resolution of 300 meters. Moreover, the 10-day temporal interval of GF-1 FVC product successfully facilitated the extraction of regional phenological information at a spatial resolution of 16 meters.

In-Terrestrial Aquaculture Fields Mapping from High Resolution Remote Sensing Images

Convolution neural networks are widely used for image processing in remote sensing. Aquacultures have an important role in food security and hence should be monitored. In this paper, a novel lightweight neural network for in-terrestrial aquaculture field retrieval from high-resolution remote sensing images is proposed. The structure of this pond segmentation network is based on the UNet architecture, providing higher training speed. Experiments are performed on Gaofen satellite datasets in Shanghai, China. The proposed network detects the inland aquaculture ponds in a shorter time than stateof-the-art neural network-based models and reaches an overall accuracy of about 90 %.

MULTI-TEMPORAL URBAN LAND-USE CHANGE DETECTION AND PREDICTION USING CNN-BASED CA-MARKOV MODEL FROM GAOFEN SATELLITE IMAGES

Abstract. The intelligent interpretation of land-use change has become a research frontier. Reasonably and effectively utilizing limited land resources and making scientific predictions to promote sustainable utilization of land resources is significant for establishing a resource-saving and environmentally friendly society. Remote sensing technology can efficiently complete multi-temporal and dynamic land-use change detection, especially using high-spatial resolution remote sensing images. However, the existing land-use change and prediction have not been combined. In addition, land-use change detection mainly relies on shallow feature design, resulting in low prediction accuracy and weak generalization performance. To solve the above problems, we proposed a CNN-based CA Markov model using multi-temporal GaoFen satellite remote sensing images for the change detection and prediction of land cover. Taking the city of Panzhihua in China as an example, the study constructed training sample data that includes a multi-temporal remote sensing training dataset from 2006, 2010, 2015, and 2021 using GaoFen satellite remote sensing images. Meanwhile, a multitemporal CNN land-use detection model was constructed to generate a land-use transfer matrix by training the dataset. Furthermore, the comprehensive driving factors were selected, including terrain factors (height and slope) and social factors (economic and population density). Then, the CA-Markov model was constructed to predict the land-use development trend in Panzhihua City after ten years. Compared with the traditional methods, experimental results demonstrate that the proposed model can improve the model's automatic interpretation ability and prediction accuracy with an increase of 24.6% in the FoM index and 4.37% in the Kappa coefficient.

High-Resolution Remote Sensing Image Segmentation Algorithm Based on Improved Feature Extraction and Hybrid Attention Mechanism

Segmentation of high-resolution remote sensing images is one of the hottest topics in deep learning. Compared to ordinary images, high-resolution remote sensing images possess characteristics such as higher intra-class diversity and lower inter-class separability. Additionally, the objects in these images are complex and have smaller sizes. Aiming at the classical segmentation network in remote sensing images, there are some problems, such as inaccurate edge object segmentation, inconsistent segmentation of different types of objects, low detection accuracy, and a high false detection rate. This paper proposes a new hybrid attention model (S-CA), a new coordinate efficient channel attention module (C-ECA), and a new small-target feature extraction network (S-FE). The S-CA model enhances important spatial and channel features in shallow layers, allowing for more detailed feature extraction. The C-ECA model utilizes convolutional layers to capture complex dependencies between variations, thereby better capturing feature information at each position and reducing redundancy in feature channels. The S-FE network can capture the local feature information of different targets more effectively. It enhances the recognition and classification capabilities of various targets and improves the detection rate of small targets. The algorithm is used for segmentation in high-resolution remote sensing images. Experiments were conducted on the public dataset GID-15 based on Gaofen-2 satellite remote sensing images. The experimental results demonstrate that the improved DeepLabV3+ segmentation algorithm for remote sensing images achieved a mean intersection over union (mIoU), mean pixel accuracy (mPA), and mean precision (mP) of 91.6%, 96.1%, and 95.5%, respectively. The improved algorithm is more effective than current mainstream segmentation networks.

Radiometric Cross-Calibration of Wide-Field-of-View Cameras Based on Gaofen-1/6 Satellite Synergistic Observations Using Landsat-8 Operational Land Imager Images: A Solution for Off-Nadir Wide-Field-of-View Associated Problems

The Gaofen-1 satellite is equipped with four wide-field-of-view (WFV) instruments, enabling an impressive spatial resolution of 16 m and a combined swath exceeding 800 km. These WFV images have shown their valuable applications across diverse fields. However, achieving accurate radiometric calibration is an essential prerequisite for establishing reliable connections between satellite signals and biophysical, as well as biochemical, parameters. However, observations with large viewing angles (>20°) pose new challenges due to the bidirectional reflectance distribution function (BRDF) effect having a pronounced impact on the accuracy of cross-radiation calibrations, especially for the off-nadir WFV1 and WFV4 cameras. To overcome this challenge, a novel approach was introduced utilizing the combined observations from the Gaofen-1 and Gaofen-6 satellites, with Landsat-8 OLI serving as a reference sensor. The key advantage of this synergistic observation strategy is the ability to obtain a greater number of image pairs that closely resemble Landsat-8 OLI reference images in terms of geometry and observation dates. This increased availability of matching images ensures a more representative dataset of the observation geometry, enabling the derived calibration coefficients to be applicable across various sun–target–sensor geometries. Then, the geometry angles and bidirectional reflectance information were put into a Particle Swarm Optimization (PSO) algorithm incorporating radiative transfer modeling. This PSO-based approach formulates cross-calibration as an optimization problem, eliminating the reliance on complex BRDF models and satellite-based BRDF products that can be affected by cloud contamination. Extensive validation experiments involving satellite data and in situ measurements demonstrated an average uncertainty of less than eight percent for the proposed cross-radiation calibration scheme. Comparisons of top-of-atmosphere (TOA) results calibrated using our proposed scheme, the previous traditional radiative transfer modeling using MODIS BRDF products for BRDF correction (RTM-BRDF) method, and official coefficients reveal the superior accuracy of our method. The proposed scheme achieves a 36.99% decrease in root mean square error (RMSE) and a 38.13% increase in mean absolute error (MAE) compared to official coefficients. Moreover, it achieves comparable accuracy to the RTM-BRDF method while eliminating the need for MODIS BRDF products, with a decrease in RMSE exceeding 14% for the off-nadir WFV1 and WFV4 cameras. The results substantiate the efficacy of the proposed scheme in enhancing cross-calibration accuracy by improving image match-up selection, efficiently removing BRDF effects, and expanding applicability to diverse observation geometries.

Crop-Planting Area Prediction from Multi-Source Gaofen Satellite Images Using a Novel Deep Learning Model: A Case Study of Yangling District

Neural network models play an important role in crop extraction based on remote sensing data. However, when dealing with high-dimensional remote sensing data, these models are susceptible to performance degradation. In order to address the challenges associated with multi-source Gaofen satellite data, a novel method is proposed for dimension reduction and crop classification. This method combines the benefits of the stacked autoencoder network for data dimensionality reduction, and the convolutional neural network for classification. By leveraging the advantages of multi-dimensional remote sensing information, and mitigating the impact of dimensionality on the classification accuracy, this method aims to improve the effectiveness of crop classification. The proposed method was applied to the extraction of crop-planting areas in the Yangling Agricultural Demonstration Zone, using multi-temporal spectral data collected from the Gaofen satellites. The results demonstrate that the fusion network, which extracts low-dimensional characteristics, offers advantages in classification accuracy. At the same time, the proposed model is compared with methods such as the decision tree (DT), random forest (RF), support vector machine (SVM), hyperspectral image classification based on a convolutional neural network (HICCNN), and a characteristic selection classification method based on a convolutional neural network (CSCNN). The overall accuracy of the proposed method can reach 98.57%, which is 7.95%, 4.69%, 5.68%, 1.21%, and 1.10% higher than the above methods, respectively. The effectiveness of the proposed model was verified through experiments. Additionally, the model demonstrates a strong robustness when classifying based on new data. When extracting the crop area of the entire Yangling District, the errors for wheat and corn are only 9.6% and 6.3%, respectively, and the extraction results accurately reflect the actual planting situation of crops.

Remote Sensing Inversion of Typical Water Quality Parameters of a Complex River Network: A Case Study of Qidong’s Rivers

The remote sensing inversion of the water quality parameters of a complex river network in the absence of historical ground data is a difficult problem in the field of remote sensing. In this paper, a sub-regional inversion method for typical water quality parameters is presented for a complex river network using Gaofen-1 satellite data. Qidong’s rivers were selected as the survey region, and different band combination models and datasets on different river sub-regions were used to perform the remote sensing inversion, which realized the inversion of the permanganate index (CODMn), ammonia nitrogen (NH3-N), total phosphorus (TP), and total nitrogen (TN) in the rivers. The results show that all the coefficients of determination (R^2) of the inversion models are larger than 0.5, indicating an increase of about 0.4 when compared with the inversion method of the whole region, indicating good relevance. Water quality data and satellite data collected at different times were used for validation, which showed good results. On the basis of the water quality inversion, the key polluted areas were extracted in combination with on-site surveys to find the pollution source in order to verify the results of the inversion. The sub-region inversion method proposed in this paper can be used for the remote sensing inversion of the water quality parameters of complex river networks in the absence of historical ground data.

Design of remote sensing image feature extraction method based on high score 2 wavelet transform

The feature extraction of Gaofen-2 Remote Sensing Image (RSI) has problems such as poor extraction accuracy and large noise reduction error. Therefore, this paper designs an RSI feature extraction method based on high score 2 wavelet transform (WT). The RSI is collected with the help of Gaofen-2 satellite and high-resolution remote sensing technology, the key points of the image are determined through the Gaussian difference scale space, and the key points of the edge are judged by the peak curvature value of the difference function at the edge junction, so as to complete the RSI acquisition. Specific filtering and spatial domain transformations are used to remove image noise and improve the quality of RSI. The mean shift (MS) algorithm is used to iteratively find the area with the most dense sample points in the RSI space, complete the image analysis, and realize the preprocessing of the high score 2 RSI. The linear features of the RSI are determined by the WT algorithm, and the image threshold is set for feature extraction of the high score 2 RSI. The experimental results show that in the RSI noise reduction error analysis of different methods, the noise reduction error curve of the sample RSI of the method proposed in this paper has the lowest trend, which is always lower than 2%. Compared with the two methods proposed before, the error is higher. At the same time, in the accuracy analysis of key point feature extraction, the proposed scheme has better accuracy. Therefore, it can be seen that this method has better comprehensive performance, and the proposed method can effectively improve the feature extraction accuracy of RSI and reduce the noise in RSI.

An improved method for retrieving aerosol optical depth over Ebinur Lake Basin from Gaofen-1

Lake Ebinur is the largest saltwater lake in Xinjiang. In recent years, the exposed dry lake bottom surface has become a remarkable source of salt and dust owing to the desiccation of the lake. Given the importance of the lake basin to human livelihoods, comprehensive monitoring of basin behavior is required. Remote sensing offers numerous novel approaches to monitoring the spatial and temporal distribution of aerosols. During this study, high spatial resolution aerosol optical depth (AOD) inversion at Lake Ebinur was still in the gap stage. The wide-field-of-view (WFV) data from four cameras on the GaoFen-1 (GF-1) satellite, launched by China in April 2013, offers high spatial and temporal resolution and excellent potential for AOD estimation. Owing to the shortwave infrared band (SWIR) deficiency and the complexity of surface reflectance, the coefficient of variation (CV) was proposed as a measure of surface reflectance to develop a surface reflectance database for land cover characteristics in the Ebinur Lake Basin. The significant improvements of the algorithm include (1) the design of an accurate and universal cloud detection algorithm through the blue and red bands. (2) The CV was used to assess the robustness of surface reflectance under long time series. After dividing the surface reflectance image pixels with low robustness, a reasonable value was assigned according to the surface reflectance variation characteristics of the image pixels. (3) Aerosol models were determined from the long-term measurements of the CIMEL solar photometer (CE318), and the validation of ground-based observations showed good correlation that WFV/GF-1 AOD was comparable to CE318 AOD (R = 0.705; root mean square error (RMSE) = 0.098; mean absolute error (MAE) = 0.078), and the smaller MAE and RMSE confirmed the feasibility of inversion of AOD by this method.

Evaluation of the High-Resolution MuSyQ LAI Product over Heterogeneous Land Surfaces

In recent years, the retrieval and validation of remotely-sensed leaf area index (LAI) products over complex land surfaces have received much attention due to the high-precision land surface model simulations and applications in global climate change. However, most of these related researches mainly focus on coarse resolution products. This is because few products have been specifically designed for solving the problems derived from complex land surfaces in mountain areas until now. MuSyQ LAI is a new product derived from Gaofen-1 (GF-1) satellite data. This product is characterized with a temporal resolution of 10 days and a spatial resolution of 16 m. As is well known, high-resolution products have less uncertainties because of the homogeneities of sub-pixel. Therefore, to evaluate the precision and uncertainty of MuSyQ LAI, an up-scaling strategy was employed here to validate MuSyQ LAI for three mountain regions in Southwest China. The validation strategy can be divided into three parts. First, a regression model was built by in situ LAI measured by LAI-2200 and the normalized difference vegetation index (NDVI) from unmanned aerial vehicle (UAV) images to obtain a 0.5 m resolution LAI map. Second, an up-scaled LAI map with a spatial resolution consistent with MuSyQ LAI was calculated by the pixel-averaging method from the UAV-based LAI map. Third, the MuSyQ LAI was validated by the up-scaled UAV-based LAI in pixel scale. Simultaneously, the sources of uncertainty were analyzed and compared from the view of data source, retrieval model, and scale effects. The results suggested that MuSyQ LAI in the study areas are significantly underestimated by 53.69% due to the complex terrain and heterogeneous land cover. There are three main reasons for the underestimation. The differences between GF-1 reflectance and UAV-based reflectance employed to estimate LAI are the largest factors for the validation results, even accounting for 61.47% of the total bias. Subsequently, the scale effects led to about 28.44% bias. Last but not least, the models employed to retrieve LAI contributed merely 10.09% uncertainties to the total bias. In conclusion, the accuracy of MuSyQ LAI still has a large space to be improved from the view of reflectance over complex terrain. This study is quite important for applications of MuSyQ LAI products and also provides a reference for the improvement and application of other high-resolution remotely sensed LAI products.

Large-scale mapping of solifluction terraces in the southeastern Tibetan Plateau using high-resolution satellite images and deep learning

Solifluction terraces are a topographic and sedimentary signature for understanding the mechanism of solifluction occurrence and the past climate (e.g., paleoclimate), and have also impacts on hillslope stability. Traditional field surveys and aerial remote sensing have been primarily employed to map solifluction terraces, but they have challenges in large-scale mapping, which results in little knowledge of spatial distribution of solifluction terraces in many areas of the world. In this study, we made a first attempt at large-scale mapping of solifluction terraces based on high-resolution satellite remote sensing images and deep learning in the Luhuo and Xinduqiao areas, located in the southeastern Tibetan Plateau. Enriched training samples were firstly generated through manually delineated 1057 solifluction terrace and non-solifluction terrace polygons in Luhuo and data augmentation. The DeepLabv3+ model (Chen et al., 2018) was then trained by these diverse and abundant samples, and employed to Gaofen-2 satellite images for extraction of solifluction terraces in both the Luhuo and Xinduqiao areas. The performance estimation showed that the F1-score, precision, and recall reached 91.15 %, 84.56 %, and 98.85 % in the Luhuo area, and 84.30 %, 84.90 %, and 83.71 % in the Xinduqiao area, respectively. The extracted results of solifluction terraces were also validated by field investigations in the Xinduqiao area. Moreover, we found that the solifluction terraces were mainly distributed around the slope angle of 21° and concentrated in the altitude range of 3000–3900 m, but the slope orientations were heterogeneous in both areas. Our results demonstrated that the combination of high-resolution satellite images and deep learning had a significant potential for mapping solifluction terraces at regional and global scales.

Land Cover Classification by Gaofen Satellite Images Based on CART Algorithm in Yuli County, Xinjiang, China

High-resolution remote-sensing images can be used in human activity analysis and criminal activity monitoring, especially in sparsely populated zones. In this paper, we explore the applicability of China’s Gaofen satellite images in the land cover classification of Xinjiang, China. First of all, the features of spectral reflectance and a normalized radar cross section (NRCS) for different types of land covers were analyzed. Moreover, the seasonal variation of the NRCS in SAR (Synthetic Aperture Radar) images for the study area, Dunkuotan Village of Yuli County, China, was demonstrated by the GEE (Google Earth Engine) platform accordingly. Finally, the CART (classification and regression trees) algorithm of a DT (decision tree) was applied to investigate the classification of land cover in the western area of China when both optical and SAR images were employed. An overall classification accuracy of 83.15% with a kappa coefficient of 0.803 was observed by using GF-2/GF-3 images (2017–2021) in the study area. The DT-based classification procedure proposed in this investigation proved that Gaofen series remote-sensing images can be engaged to effectively promote the routine workflow of the administrative department.

Enabling country-scale land cover mapping with meter-resolution satellite imagery

High-resolution satellite images can provide abundant, detailed spatial information for land cover classification, which is particularly important for studying the complicated built environment. However, due to the complex land cover patterns, the costly training sample collections, and the severe distribution shifts of satellite imageries caused by, e.g., geographical differences or acquisition conditions, few studies have applied high-resolution images to land cover mapping in detailed categories at large scale. To fill this gap, we present a large-scale land cover dataset, Five-Billion-Pixels. It contains more than 5 billion labeled pixels of 150 high-resolution Gaofen-2 (4 m) satellite images, annotated in a 24-category system covering artificial-constructed, agricultural, and natural classes. In addition, we propose a deep-learning-based unsupervised domain adaptation approach that can transfer classification models trained on labeled dataset (referred to as the source domain) to unlabeled data (referred to as the target domain) for large-scale land cover mapping. Specifically, we introduce an end-to-end Siamese network employing dynamic pseudo-label assignment and class balancing strategy to perform adaptive domain joint learning. To validate the generalizability of our dataset and the proposed approach across different sensors and different geographical regions, we carry out land cover mapping on five megacities in China and six cities in other five Asian countries severally using: PlanetScope (3 m), Gaofen-1 (8 m), and Sentinel-2 (10 m) satellite images. Over a total study area of 60,000 km2, the experiments show promising results even though the input images are entirely unlabeled. The proposed approach, trained with the Five-Billion-Pixels dataset, enables high-quality and detailed land cover mapping across the whole country of China and some other Asian countries at meter-resolution.

Black and Odorous Water Detection of Remote Sensing Images Based on Improved Deep Learning

Black and odorous water seriously affects the ecological balance of rivers and the health of people living nearby. Satellite remote sensing technology with its advantages of a large range, long-time series, low cost, and high efficiency, has provided a new area for water quality detection. Much archived remote sensing satellite data can be further processed and used as a data source for black and odorous water detection. In this paper, Gaofen-2 remote sensing data with a spatial resolution of 1 m is leveraged as the data source. To enrich the data source in the northern coastal zone of China, we have built a high-quality remote sensing dataset, called the remote sensing images for black and odorous water detection (RSBD) dataset, which is collected from the Gaofen-2 satellite in Yantai, China. In addition, we propose a network with an encoder-decoder discriminant structure for black and odorous water detection. In the network, an augmented attention module is designed to capture a more comprehensive semantic feature representation. Further, the median balancing loss function is adopted to solve the imbalance issues. Experimental results demonstrate that the network is superior to other state-of-the-art semantic segmentation methods on our dataset.

Continuous Change Detection of Flood Extents With Multisource Heterogeneous Satellite Image Time Series

Flood monitoring is of crucial importance for protecting lives and properties. Change detection (CD) methods on multi-source remote sensing images have been widely used for flood extent monitoring. In this paper, we propose a spatiotemporal fusion CD (STFCD) algorithm, exploiting the spatial dependence and temporal interaction of multi-source heterogeneous (MSH) satellite image time series (SITS), to realize improved flood CD performance in comparison with existing methods. The proposed STFCD algorithm mainly contains two steps, i.e., spatial clustering and temporal fusion. In the spatial clustering step, we propose a sparse Markov random field (MRF)-based strategy to exploit contextually spatial features in each image of MSH-SITS, which provides a larger local receptive field than the commonly used MRF. In the temporal fusion step, the historical information of flood detection results is employed as constraints to effectively reduce the effects of terrain shadows in synthetic aperture radar (SAR) images and cloud shadows and topography shadows in optical images on flood CD results of existing methods in accordance with the temporal dependence among MSH-SITS. Experiments on real MSH-SITS (containing Gaofen-1, Gaofen-3, Gaofen-6, Sentinel-1, and Sentinel-2 satellite images) covering Chinese Amur and Huma Rivers show that the overall flood CD accuracy of our proposed STFCD algorithm is higher than the other commonly used algorithms for CD of flood extents and demonstrate the robustness of our proposed STFCD algorithm.

Extraction of Pig Farms From GaoFen Satellite Images Based on Deep Learning

Extraction of Pig Farms From GaoFen Satellite Images Based on Deep Learning

Progressive Task-Based Universal Network for Raw Infrared Remote Sensing Imagery Ship Detection

Infrared remote sensing images are becoming increasingly popular due to their superior penetration and resistance to light interference. However, challenges still remain when applying them in real-world applications: 1) raw infrared images suffer from severe stripes interference, and the preprocessing techniques used to obtain standard image products for subsequent detection tasks tend to be time-consuming, which fails to meet the application requirements; 2) current destriping techniques may inevitably weaken the local contrast between some objects and the local background since they need to consider the gray consistency of the overall image; 3) in low-resolution images, dim and small infrared targets are challenging to discriminate, resulting in high false alarms. To address these challenges, we proposed a progressive task-based universal network for raw infrared image ship detection while simultaneously removing stripes. First, we built an integrated network consisting of two components: the stripe denoising component (SDC) and the object detection component (ODC). We also designed a feedback loss adjustment mechanism to enhance the focus of the SDC on the target area. Second, a directed two-branch network was constructed for efficient stripe noise removal, including an <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">x</i> -direction branch for feature enhancement and a <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">y</i> -direction branch for grayscale smoothing. Finally, a parallel network with two labels was designed to extract the inherent features of the target and the background, as well as their relationship features, to achieve refined ship detection. We conducted experiments on a self-assembled dataset from the GaoFen-1 satellite to validate our approach. The experimental results demonstrated that the proposed method outperformed other state-of-the-art methods in infrared image ship detection.

Building Detection From Panchromatic and Multispectral Images With Dual-Stream Asymmetric Fusion Networks

Building detection from panchromatic (PAN) and Multi-spectral (MS) images is an essential task for many practical applications. In this paper, a dual-stream asymmetric fusion network is proposed, named DAFNet. DAFNet can achieve effective information fusion at the feature level. It obtains better building detection performance from three perspectives: 1) A two-stream network structure is designed to guarantee the ability to extract information from PAN and MS images; 2) An asymmetric feature fusion (AFF) module is proposed to fuse features efficiently and concisely; 3) Two consistency regularization losses, i.e., PAN information preservation (PiP) loss and cross-modal semantic consistency (CSC) loss are applied to further explore the consistency between features for better fusion. The experiments are conducted on a challenging building detection dataset collected from GaoFen-2 satellite images. Comprehensive evaluations on 12 popular detection methods demonstrate the superiority of our DAFNet compared with the existing state-of-the-art fusion methods. We reveal that feature-level fusion is more suitable for building detection from PAN-MS images. Code is released at <uri xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">https://github.com/floatingstarZ/DAFNet</uri>