The rapid development of unmanned aerial vehicle (UAV) technology has contributed to the increasing sophistication of UAV-based object-detection systems, which are now extensively utilized in civilian and military sectors. However, object detection from UAV images has numerous challenges, including significant variations in the object size, changing spatial configurations, and cluttered backgrounds with multiple interfering elements. To address these challenges, we propose SOD-YOLO, an innovative model based on the YOLOv8 model, to detect small objects in UAV images. The model integrates the receptive field convolutional block attention module (RFCBAM) in the backbone network to perform downsampling, improving feature extraction efficiency and mitigating the spatial information sparsity caused by downsampling. Additionally, we developed a novel neck architecture called the balanced spatial and semantic information fusion pyramid network (BSSI-FPN) designed for multi-scale feature fusion. The BSSI-FPN effectively balances spatial and semantic information across feature maps using three primary strategies: fully utilizing large-scale features, increasing the frequency of multi-scale feature fusion, and implementing dynamic upsampling. The experimental results on the VisDrone2019 dataset demonstrate that SOD-YOLO-s improves the mAP50 indicator by 3% compared to YOLOv8s while reducing the number of parameters and computational complexity by 84.2% and 30%, respectively. Compared to YOLOv8l, SOD-YOLO-l improves the mAP50 indicator by 7.7% and reduces the number of parameters by 59.6%. Compared to other existing methods, SODA-YOLO-l achieves the highest detection accuracy, demonstrating the superiority of the proposed method.

Unmanned aerial vehicle (UAV) image object detection has extensive applications across both civilian and military domains. However, the traditional YOLOv8 detection algorithm faces significant challenges in detecting small objects in UAV imagery, primarily due to a high missed detection rate and an excessive number of parameters. To address these issues, this paper introduces an enhanced small object detection approach, called Small-Size Object Detection Algorithm Based on Improved YOLOv8 for UAV Imagery (SS-YOLOv8). Firstly, considering the difficulties and stringent real-time requirements in detecting small objects in UAV aerial images, this work streamlines the model by eliminating the large object detection head and its associated redundant network layers, significantly limiting the parameter count. Furthermore, a specialized tiny object detection head is proposed. It is custom-designed for detecting small objects. To enhance the model’s capacity for extracting fine features of small objects and mitigate information loss during feature extraction, the convolution module is replaced in the backbone with space-to-depth convolution (SPD-Conv). Moreover, in the neck section, the self created GCU module is incorporated as it effectively amalgamates deep semantic features with shallow positional ones. Finally, while maintaining consistent parameter costs, multi-scale features are meticulously reused and incorporated to achieve a more comprehensive and sophisticated feature fusion. The experimental results on the VisDrone2019 dataset showed that compared with YOLOv8, SS-YOLOv8 improved its MAP index by 6.9% on the validation set and 5.8% on the test set, while reducing the number of parameters and model size by 65.93% and 64.85%, respectively. Compared with the best performing comparison algorithm, SS-YOLOv8 has the highest detection accuracy, indicating the superiority of this method.

Ship detection in Synthetic Aperture Radar (SAR) images has garnered widespread attention due to its significant applications. However, few researchers have focused on the challenges posed by the unique limitations of SAR imaging mechanisms. These challenges include issue discontinuity and incompleteness of ship instances caused by internal hollowing or fracturing, the conventional IoU metric possesses limitations in evaluating the similarity between two boxes, and the self-attention mechanism considers only absolute position information. To address these challenges, we propose the Mixed Receptive Field and Multi-Position Encoding for Scale-Sensitive SAR Ship Detection with Transformer (M²S-DETR), which consists of four key components: the Mixed Receptive Field Module (MRFM), Multi-Position Encoding (MPE), Multi-Scale Feature Aggregation Unit (MFAU), and Scale-Sensitive IoU (S2-IoU). The MRFM, based on dilated convolution, complements missing ship structure information and allows for a larger receptive field. To more accurately capture position relationships within the self-attention mechanism, we incorporate learnable relative position encoding in the MPE, which reveals more nuanced relative position information to supplement absolute position encoding. In the multi-scale feature extraction and fusion phase, the MFAU utilizes a dual-path top-down and bottom-up strategy to better represent features across all scales. Additionally, we introduce Scale-Sensitive IoU (S2-IoU) into the classification branch to mitigate the phenomenon of inconsistent prediction pairs, thereby enabling the transformer decoder to perform high-quality initial query selection. Our method surpasses several CNN-based and Transformer-based detectors on the commonly used benchmark datasets SSDD and HRSID, respectively. Code is available at: https://github.com/douling843/M²S-DETR.

Synthetic aperture radar (SAR) is an active coherent microwave remote sensing system. SAR systems working in different bands have different imaging results for the same area, resulting in different advantages and limitations for SAR image classification. Therefore, to synthesize the classification information of SAR images into different bands, an SAR image fusion classification method based on the decision-level combination of multi-band information is proposed in this paper. Within the proposed method, the idea of Dempster–Shafer evidence theory is introduced to model the uncertainty of the classification result of each pixel and used to combine the classification results of multiple band SAR images. The convolutional neural network is used to classify single-band SAR images. Calculate the belief entropy of each pixel to measure the uncertainty of single-band classification, and generate the basic probability assignment function. The idea of the term frequency-inverse document frequency in natural language processing is combined with the conflict coefficient to obtain the weight of different bands. Meanwhile, the neighborhood classification of each pixel in different band sensors is considered to obtain the total weight of each band sensor, generate weighted average BPA, and obtain the final ground object classification result after fusion. The validity of the proposed method is verified in two groups of multi-band SAR image classification experiments, and the proposed method has effectively improved the accuracy compared to the modified average approach.

High resolution remote sensing imageries contain a wealth of information about ground object features. Semantic segmentation of high resolution remote sensing imageries has emerged as a prominent area of research. However, many existing models have become complex in pursuit of accuracy, making it challenging to achieve a balance between efficiency and precision. To address this issue, we propose LiteSeger, a lightweight semantic segmentation model that combines the strengths of convolutional neural network (CNN) and transformer architectures. Initially, the MobileViT backbone network, which integrates CNN and Transformer, is employed to rapidly extract multi-level features. Subsequently, we devised a Global Attention mechanism and a Local Attention mechanism, respectively, for low-level and high-level features, which we incorporated into the decoder as the Main Feature Fusion (MFF) module. Local Attention enhances the representation of detailed features in both spatial and channel dimensions. Global Attention strengthens global interconnectivity through long-range semantic information. Particularly, Our proposed Distance-aware Grouped Attention demonstrates a significant reduction in the computational complexity of attention computation, without compromising segmentation accuracy. In comparison to other light-weight models, our proposed LiteSeger achieves the best MIoU and Ave.F1 on both the Vaihingen and Potsdam datasets.On the Vaihingen dataset, it achieves an MIoU of 71.22 and an Ave.F1 of 82.97. On the Potsdam dataset, the MIoU and Ave.F1 reach 72.00 and 83.34, respectively. At the same time, it has significant advantages in inference efficiency. LiteSeger contains only 7.0 M parameters and under the experimental settings of this study, it achieves an inference speed of 50 FPS. The source codes are available at https://github.com/HIGISX/LiteSeger.