The detection of arbitrary-oriented and multi-scale objects in satellite optical imagery is an important task in remote sensing and computer vision. Despite significant research efforts, such detection remains largely unsolved due to the diversity of patterns in orientation, scale, aspect ratio, and visual appearance; the dense distribution of objects; and extreme imbalances in categories. In this paper, we propose an adaptive dynamic refined single-stage transformer detector to address the aforementioned challenges, aiming to achieve high recall and speed. Our detector realizes rotated object detection with RetinaNet as the baseline. Firstly, we propose a feature pyramid transformer (FPT) to enhance feature extraction of the rotated object detection framework through a feature interaction mechanism. This is beneficial for the detection of objects with diverse patterns in terms of scale, aspect ratio, visual appearance, and dense distributions. Secondly, we design two special post-processing steps for rotated objects with arbitrary orientations, large aspect ratios and dense distributions. The output features of FPT are fed into post-processing steps. In the first step, it performs the preliminary regression of locations and angle anchors for the refinement step. In the refinement step, it performs adaptive feature refinement first and then gives the final object detection result precisely. The main architecture of the refinement step is dynamic feature refinement (DFR), which is proposed to adaptively adjust the feature map and reconstruct a new feature map for arbitrary-oriented object detection to alleviate the mismatches between rotated bounding boxes and axis-aligned receptive fields. Thirdly, the focus loss is adopted to deal with the category imbalance problem. Experiments on two challenging satellite optical imagery public datasets, DOTA and HRSC2016, demonstrate that the proposed ADT-Det detector achieves a state-of-the-art detection accuracy (79.95% mAP for DOTA and 93.47% mAP for HRSC2016) while running very fast (14.6 fps with a 600 × 600 input image size).

Recent studies in remote sensing object detection have made excellent progress and shown promising performance. However, most current detectors only explore rotation-invariant feature extraction but disregard the valuable spatial and semantic prior knowledge in remote sensing images (RSIs), which limits the detection performance when encountering blurred or heavy occluded objects. To address this issue, we propose a mask-reconstruction relation learning (MRRL) framework to learn such prior knowledge among objects and a consistency-reasoning transformer over relation proposals (CTRP) to recognize objects with limited visual features via consistency reasoning. Specifically, MRRL framework applies random mask to some objects in the training dataset and performs masked objects reconstruction to guide the network to learn the distribution consistency of objects. CTRP is the core component of the MRRL framework, which models the interaction between spatial and semantic priors, and uses easy detected objects to reason hard detected objects. The trained CTRP can be integrated into the existing detector to improve the ability of object detection with limited visual features in RSIs. Extensive experiments on widely-used datasets for two distinct tasks, namely remote sensing object detection task and occluded object detection task, demonstrate the effectiveness of the proposed method. Source code is available at https://github.com/sunpeng96/CTRP_mmrotate.

Oriented object detection has attained remarkable progress in addressing the challenges associated with rotating invariant feature extraction. However, most existing object detection most existing encounter serious performance degradation when processing objects with tiny size, blurring and occlusion. One of the reasons is that existing methods mainly focus on local features, while overlooking the correlation among objects and common sense knowledge, which is inconsistent with the human visual system. To address this issue, we propose a dual-stream global relationship learning (DGRL) method consisting of two modules: a dynamic correlation learning (DCL) method and a global knowledge mapping (GKM) module. The DCL method can construct a region-to-region dynamic relationship graph based on feature correlations and implicitly guides the detection network to learn more powerful class representation by updating nodes of the graph. The GKM module generates a class-to-class global semantic relationship graph via external knowledge and achieve more stable representation learning by dynamically global relational mapping. Extensive experiments are performed on oriented object detection datasets DOTA, HRSC2016, DIOR-R as well as horizontal object detection datasets NWPU VHR-10, RSOD. The results demonstrate that the proposed method achieves the state-of-the-art detection accuracy.

Arbitrary-oriented object detection in remote sensing optical images is an emerging yet challenging task due to the diversity of the visual appearance of objects. Although considerable progress has been made, existing methods often suffer from insufficient positive samples and critical feature misalignment caused by objects with considerable diversity. In this paper, we propose a critical feature learning (CFL) method to address the above issues. First, to obtain a more accurate location, we introduce the spatial transform to model the rotating bounding box regression and construct a spatial transform selection (STS) strategy, which is used to implement dynamic sample selection metrics to guarantee sufficient positive samples. Then, we propose a scale-controlled smooth L ₁ loss (SC-Loss) to change the form of regression loss function based on the statistics of proposals so that the training can focus more on high-quality samples. Finally, to capture critical features, we propose a critical feature sampling (CFS) module to perform location refinement of classification feature samples to extract accurate features. Comprehensive and extensive experimental results based on three benchmarks, DOTA, FAIR1M-1.0, and HRSC2016, demonstrate that the proposed method is superior to existing methods in terms of accuracy and effectiveness.