针对现有水下目标检测方法在成像退化类型多样与背景干扰等复杂场景中鲁棒性差以及在设备资源受限条件下难以兼顾检测精度与模型轻量化的问题,该文提出一种基于冗余信息优化的轻量化水下目标检测网络(UnderWater Faster YOLO network ba...针对现有水下目标检测方法在成像退化类型多样与背景干扰等复杂场景中鲁棒性差以及在设备资源受限条件下难以兼顾检测精度与模型轻量化的问题,该文提出一种基于冗余信息优化的轻量化水下目标检测网络(UnderWater Faster YOLO network based on redundancy information optimization,UWF-YOLO),并进一步构建了复杂场景水下目标检测数据集(Underwater Object Detection dataset with Complex Scene,CSUOD)。UWFYOLO采用FasterNet Block重构C2f模块优化主干和颈部网络,通过特征通道选择机制减少冗余特征,并引入Ghost卷积增强颈部网络的多尺度特征融合能力;同时,通过基于分组卷积的参数共享检测头降低计算开销;最后,应用结构化通道剪枝进一步压缩网络规模。CSUOD数据集通过收集真实水下图像进行标注和分辨率标准化处理,覆盖雾化、色偏、非均匀照明等各种退化类型,可用于复杂场景下水下目标检测模型的鲁棒性训练与性能评测。在DUO,RUOD和TrashCan数据集上进行实验分析,相较于YOLOv8s,所提方法在计算量、权重大小与参数量3个指标上分别降低了60.4%,77.3%和78.4%;与参数量相当的YOLOv9-tiny相比,mAP指标在3个数据集上分别提升了0.3%,2.3%和3.4%。同时,在自建CSUOD数据集上的主客观对比实验,进一步证实所提模型在实现显著轻量化的同时,能够有效避免背景干扰导致的误检、漏检等问题,特别在复杂水下环境中展现出优异的检测性能。此外,该文构建的复杂场景水下数据集将有助于推动水下目标检测方法的发展。展开更多
In the area of computer vision, deep learning has produced a variety of state-of-the-art models that rely on massive labeled data. However, collecting and annotating images from the real world is too demanding in term...In the area of computer vision, deep learning has produced a variety of state-of-the-art models that rely on massive labeled data. However, collecting and annotating images from the real world is too demanding in terms of labor and money investments, and is usually inflexible to build datasets with specific characteristics, such as small area of objects and high occlusion level. Under the framework of Parallel Vision, this paper presents a purposeful way to design artificial scenes and automatically generate virtual images with precise annotations.A virtual dataset named Parallel Eye is built, which can be used for several computer vision tasks. Then, by training the DPM(Deformable parts model) and Faster R-CNN detectors, we prove that the performance of models can be significantly improved by combining Parallel Eye with publicly available real-world datasets during the training phase. In addition, we investigate the potential of testing the trained models from a specific aspect using intentionally designed virtual datasets, in order to discover the flaws of trained models. From the experimental results, we conclude that our virtual dataset is viable to train and test the object detectors.展开更多
Space object recognition plays an important role in spatial exploitation and surveillance, followed by two main problems: lacking of data and drastic changes in viewpoints. In this article, firstly, we build a three-...Space object recognition plays an important role in spatial exploitation and surveillance, followed by two main problems: lacking of data and drastic changes in viewpoints. In this article, firstly, we build a three-dimensional (3D) satellites dataset named BUAA Satellite Image Dataset (BUAA-SID 1.0) to supply data for 3D space object research. Then, based on the dataset, we propose to recognize full-viewpoint 3D space objects based on kernel locality preserving projections (KLPP). To obtain more accurate and separable description of the objects, firstly, we build feature vectors employing moment invariants, Fourier descriptors, region covariance and histogram of oriented gradients. Then, we map the features into kernel space followed by dimensionality reduction using KLPP to obtain the submanifold of the features. At last, k-nearest neighbor (kNN) is used to accomplish the classification. Experimental results show that the proposed approach is more appropriate for space object recognition mainly considering changes of viewpoints. Encouraging recognition rate could be obtained based on images in BUAA-SID 1.0, and the highest recognition result could achieve 95.87%.展开更多
摘要针对现有水下目标检测方法在成像退化类型多样与背景干扰等复杂场景中鲁棒性差以及在设备资源受限条件下难以兼顾检测精度与模型轻量化的问题,该文提出一种基于冗余信息优化的轻量化水下目标检测网络(UnderWater Faster YOLO network based on redundancy information optimization,UWF-YOLO),并进一步构建了复杂场景水下目标检测数据集(Underwater Object Detection dataset with Complex Scene,CSUOD)。UWFYOLO采用FasterNet Block重构C2f模块优化主干和颈部网络,通过特征通道选择机制减少冗余特征,并引入Ghost卷积增强颈部网络的多尺度特征融合能力;同时,通过基于分组卷积的参数共享检测头降低计算开销;最后,应用结构化通道剪枝进一步压缩网络规模。CSUOD数据集通过收集真实水下图像进行标注和分辨率标准化处理,覆盖雾化、色偏、非均匀照明等各种退化类型,可用于复杂场景下水下目标检测模型的鲁棒性训练与性能评测。在DUO,RUOD和TrashCan数据集上进行实验分析,相较于YOLOv8s,所提方法在计算量、权重大小与参数量3个指标上分别降低了60.4%,77.3%和78.4%;与参数量相当的YOLOv9-tiny相比,mAP指标在3个数据集上分别提升了0.3%,2.3%和3.4%。同时,在自建CSUOD数据集上的主客观对比实验,进一步证实所提模型在实现显著轻量化的同时,能够有效避免背景干扰导致的误检、漏检等问题,特别在复杂水下环境中展现出优异的检测性能。此外,该文构建的复杂场景水下数据集将有助于推动水下目标检测方法的发展。
基金supported by the National Natural Science Foundation of China(61533019,71232006)
摘要In the area of computer vision, deep learning has produced a variety of state-of-the-art models that rely on massive labeled data. However, collecting and annotating images from the real world is too demanding in terms of labor and money investments, and is usually inflexible to build datasets with specific characteristics, such as small area of objects and high occlusion level. Under the framework of Parallel Vision, this paper presents a purposeful way to design artificial scenes and automatically generate virtual images with precise annotations.A virtual dataset named Parallel Eye is built, which can be used for several computer vision tasks. Then, by training the DPM(Deformable parts model) and Faster R-CNN detectors, we prove that the performance of models can be significantly improved by combining Parallel Eye with publicly available real-world datasets during the training phase. In addition, we investigate the potential of testing the trained models from a specific aspect using intentionally designed virtual datasets, in order to discover the flaws of trained models. From the experimental results, we conclude that our virtual dataset is viable to train and test the object detectors.
基金National Natural Science Foundation of China (60776793,60802043)National Basic Research Program of China (2010CB327900)
摘要Space object recognition plays an important role in spatial exploitation and surveillance, followed by two main problems: lacking of data and drastic changes in viewpoints. In this article, firstly, we build a three-dimensional (3D) satellites dataset named BUAA Satellite Image Dataset (BUAA-SID 1.0) to supply data for 3D space object research. Then, based on the dataset, we propose to recognize full-viewpoint 3D space objects based on kernel locality preserving projections (KLPP). To obtain more accurate and separable description of the objects, firstly, we build feature vectors employing moment invariants, Fourier descriptors, region covariance and histogram of oriented gradients. Then, we map the features into kernel space followed by dimensionality reduction using KLPP to obtain the submanifold of the features. At last, k-nearest neighbor (kNN) is used to accomplish the classification. Experimental results show that the proposed approach is more appropriate for space object recognition mainly considering changes of viewpoints. Encouraging recognition rate could be obtained based on images in BUAA-SID 1.0, and the highest recognition result could achieve 95.87%.