7 February 2025 in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
In the domain of remote sensing imagery, minimizing the missed and false detection of small objects holds significant importance for the technology used in detecting small targets with unmanned aerial vehicles (UAVs). In small object detection, accurately localizing these objects becomes challenging due to the varying motion amplitudes required in different directions. Additionally, capturing the global dependency of features while keeping the network light poses a challenge. To tackle these challenges, we propose a novel object detection framework built upon the YOLO architecture, called αS-YOLO, which simultaneously tackles the problem of precise localization of small targets and maintaining the network's lightweight structure during long-range dependency capture. First, we design a new C2f_GE (Cross-convolution with 2 filters_ Global Context and Efficient Channel Attention) module, which aggregates global contextual features into the features of each pixel, enhancing the ability to model long-range dependencies while reducing network parameters to keep the network lightweight. Finally, to address the challenge of accurate small object localization, we propose a novel loss function, α-SIOU, which includes an adaptive angular control coefficient. This coefficient adjusts the distance loss in different directions based on the angular variation between the predicted and ground truth frames, adaptively converging the distance with the highest gradient. Experiments show that compared with the most advanced YOLO model, the accuracy of the proposed αS-YOLO method can be achieved in the detection of small targets on vehicles. On the VisDrone-DET2019 dataset, mAP and mAP50 compared to YOLOV8 improved 1% and 1.1%.
3 August 2023 in IEEE Transactions on Geoscience and Remote Sensing
Synthetic Aperture Radar (SAR) ship detection methods have achieved remarkable progress in recent years. However, unlike RGB images, the characteristics of SAR imaging will result in non-intuitive feature representations. Furthermore, due to the insufficient data of SAR images, existing methods relying on plenty of labeled SAR images may be hard to achieve promising performance. To address the aforementioned issues, a cross-modality feature transfer (CMFT) method is proposed in this paper, which enhances feature representations in the SAR modality by transferring rich knowledge in the RGB modality. First, we propose a multi-level modality alignment network (MMAN), which encourages the model to effectively learn modality-invariant features and alleviate the large cross-modality discrepancies by aligning features from multi-levels (scene level, local level, global level, and instance level). Second, to address the underperformance of samples with non-intuitive features in the modality alignment, we introduce a hard-sample supervision module (HSM) in the stage of feature extraction, which can thoroughly exploit the feature of hard-to-align samples by giving more optimization energy for them. Third, to enhance the discriminability of instance-level features, a feature complementary module (FCM) is customized to fully explore the potential complementary clues between instance-level features and context information for the instance-level feature alignment. Extensive experimental results demonstrate that the CMFT outperforms state-of-the-art detectors. Compared to the baseline model, CMFT improves the accuracy by 3.1% mAP on the SSDD dataset and 3.4% mAP on the HRSID dataset, demonstrating its superior SAR ship detection performance.
29 April 2024 in IEEE Signal Processing Letters
Video-based visible-infrared person re-identification (VI-ReID) aims to identify the same suspicious person captured by different sensors at different times and scenes. Overcoming the modality discrepancy caused by the different imaging principles of visible and infrared videos is a crucial challenge in this field. To address the above challenge, we propose a three-stage framework for video-based VI-ReID to fully exploit the identity information contained in different modality data, which named Decomposition-Mining- Aggregation framework. Our framework consists of three stages: decomposition, mining, and aggregation stage. Experimental results demonstrate that the combination of the three stages can effectively enhance the feature extraction network's ability to extract identity information from images with modality differences. In addition, we also illustrate the influence of video frame sampling and exploration strategy on the quality of VI-ReID model through experiments. Extensive experimental results on the HITSZ-VCM dataset demonstrate that our framework achieves the best accuracy in the VI-ReID task.
3 March 2025 in IEEE Transactions on Geoscience and Remote Sensing
In recent years, reconstruction-based methods have achieved excellent detection results in the field of hyperspectral anomaly detection (HAD). These methods predominantly operate on two aspects regarding their working principles: 1) reconstructing background pixels, and 2) suppressing anomalous pixels. However, most methods only tackle the HAD task from the spatial and spectral domains, making it challenging to effectively suppress anomalies. To eliminate these issues, this paper proposes a Fourier conditional mask based mixed attention (FCMMA) method. Firstly, we propose the Fourier conditional mask based mixed attention (FCMMA) method for hyperspectral anomaly detection. FCMMA generates a conditional mask that suppresses anomalous high-frequency information and preserves background low-frequency information in the frequency domain, optimizing the anomaly detection process. In addition, to achieve fine-grained hyperspectral anomaly detection, we propose the Fourier anomaly suppression filter (FASF). FASF uses Fourier techniques to manage background and anomalies, improving detection via precise frequency decoupling. Finally, a conditional mask (CMASK) network is designed to effectively suppress anomalies. The CMASK network integrated the FASF module and the spatial-spectral multilayer perceptual machine (SSMLP) module together to enhance the transformation and representation capabilities of the generated masks, which can also help suppress anomalies. The results on five different datasets show that the proposed method is more effective and superior when compared to nine state-of-the-art methods.
19 September 2024 in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
Semantic change detection (SCD) aiming to localize changes and identify their types is a complex pixel-level classification task. In SCD, detecting changes of narrow and long objects (NLOs) common in various scenarios is more challenging than detecting these of typical small objects, primarily because NLOs' narrowness makes them carry local details, while their length gives them a global view. The fusion of semantic with spatial information and the interaction of semantic with spatio-temporal information are crucial to accurately detecting changed NLOs. Insufficient utilization of these pieces of information carried by multi-layer semantic feature and change feature would limit the representation of the characteristics of changed NLOs, thereby leading to missed detection and false detection. We propose a new model that integrates semantic, spatial, and temporal information, named SSTNet, for semantic change detection. This model utilizes a designed multi-layer feature fusion (MFF) module to fuse features from different layers in the encoder, leveraging global semantic information and local spatial information. Additionally, the semantic change feature interaction (SCFI) module facilitates the interaction between semantic feature and change feature, enhancing the correlation between semantic information and spatial information. A series of experiments conducted on two public datasets demonstrate that SSTNet improves the detection accuracy of changed NLOs and outperforms state-of-the-art methods.