22 February 2023 in IEEE Geoscience and Remote Sensing Letters
Rotation object detection is of great importance in remote sensing imagery where the orientation is arbitrary and objects are densely distributed. However, there are several challenges that need to be overcome, such as the angular boundary problem in the regression-based methods and the square-like problem in the classification-based methods. For square-like object rotation detection, classification-based methods (e.g. CSL) suffer from the inconsistencies between angular coding and evaluation mechanisms due to the variation of aspect ratio. To address the angular inconsistencies of square-like object existing in current classification methods, we design a novel angular encoding mechanism based on aspect ratio. We make optimisations and improvements in the following two aspects: i) proposing an aspect ratio-based bidirectional coded label (AR-BCL) to replace circular smooth label (CSL) for angle coding of square-like object, which significantly improves detection accuracy for square-like object. ii) designing a cross-fusion decoupled head (CF-DH) based on angular classification to replace the existing coupled head (CH), which can help extract features that suitable for angular classification. Extensive experiments on DOTA, a large-scale public dataset for aerial images, demonstrate the effectiveness of our method for square-like object detection.
16 January 2021 in Computer Graphics Forum
20 February 2020 in arXiv
Precision mapping of landslide inventory is crucial for hazard mitigation. Most landslides generally co-exist with other confusing geological features, and the presence of such areas can only be inferred unambiguously at a large scale. In addition, local information is also important for the preservation of object boundaries. Aiming to solve this problem, this paper proposes an effective approach to fuse both local and non-local features to surmount the contextual problem. Built upon the U-Net architecture that is widely adopted in the remote sensing community, we utilize two additional modules. The first one uses dilated convolution and the corresponding atrous spatial pyramid pooling, which enlarged the receptive field without sacrificing spatial resolution or increasing memory usage. The second uses a scale attention mechanism to guide the up-sampling of features from the coarse level by a learned weight map. In implementation, the computational overhead against the original U-Net was only a few convolutional layers. Experimental evaluations revealed that the proposed method outperformed state-of-the-art general-purpose semantic segmentation approaches. Furthermore, ablation studies have shown that the two models afforded extensive enhancements in landslide-recognition performance.