Drone-based small object detection is of great significance in practical applications such as military actions, disaster rescue, transportation, etc. However, the severe scale differences in objects captured by drones...Drone-based small object detection is of great significance in practical applications such as military actions, disaster rescue, transportation, etc. However, the severe scale differences in objects captured by drones and lack of detail information for small-scale objects make drone-based small object detection a formidable challenge. To address these issues, we first develop a mathematical model to explore how changing receptive fields impacts the polynomial fitting results. Subsequently, based on the obtained conclusions, we propose a simple but effective Hybrid Receptive Field Network (HRFNet), whose modules include Hybrid Feature Augmentation (HFA), Hybrid Feature Pyramid (HFP) and Dual Scale Head (DSH). Specifically, HFA employs parallel dilated convolution kernels of different sizes to extend shallow features with different receptive fields, committed to improving the multi-scale adaptability of the network;HFP enhances the perception of small objects by capturing contextual information across layers, while DSH reconstructs the original prediction head utilizing a set of high-resolution features and ultrahigh-resolution features. In addition, in order to train HRFNet, the corresponding dual-scale loss function is designed. Finally, comprehensive evaluation results on public benchmarks such as VisDrone-DET and TinyPerson demonstrate the robustness of the proposed method. Most impressively, the proposed HRFNet achieves a mAP of 51.0 on VisDrone-DET with 29.3 M parameters, which outperforms the extant state-of-the-art detectors. HRFNet also performs excellently in complex scenarios captured by drones, achieving the best performance on the CS-Drone dataset we built.展开更多
面向高能同步辐射光源(High Energy Photon Source,HEPS)的高性能像素阵列探测器(HEPS-BPIX4)的数据获取系统(Data Acquisition,DAQ)需满足高实时性要求。通过在线压缩图像数据,可有效降低后续传输与存储的压力。针对传统压缩算法在压...面向高能同步辐射光源(High Energy Photon Source,HEPS)的高性能像素阵列探测器(HEPS-BPIX4)的数据获取系统(Data Acquisition,DAQ)需满足高实时性要求。通过在线压缩图像数据,可有效降低后续传输与存储的压力。针对传统压缩算法在压缩率和实时性方面的不足,本文提出了一种基于深度学习目标检测的图像数据在线压缩方法。采用端到端的YOLO(You Only Look Once)目标检测算法,对深度学习模型进行高效训练,并验证了其在HEPS-BPIX4 DAQ数据流中实现在线数据压缩的可行性。实验结果表明,该方法的图像数据在线压缩平均压缩比达到5.88。同时,设计了高效的部署方案,并对性能进行了测试,单线程下的压缩处理速率可达GB∙s^(−1)量级。此外,进一步提出了适用于HEPS-BPIX4 DAQ框架的多线程部署方案,以满足更高的压缩性能需求,为缓解HEPS-BPIX4 DAQ系统高带宽图像数据处理压力提供了新思路。展开更多
小而杂乱的物体交织在一起,在遥感图像中尤为常见,给目标检测带来了巨大挑战。在旋转目标检测任务中这个困难更加突出。鉴于此,本文提出了基于提案增强的解耦特征挖掘旋转检测器(decoupled feature mining rotational detector based on...小而杂乱的物体交织在一起,在遥感图像中尤为常见,给目标检测带来了巨大挑战。在旋转目标检测任务中这个困难更加突出。鉴于此,本文提出了基于提案增强的解耦特征挖掘旋转检测器(decoupled feature mining rotational detector based on proposal enhancement,PDMDet)。首先,采用单阶段检测器取代两阶段检测器的提案生成网络,通过生成高质量提案以减少背景冗余。其次,在相同维度使用自注意力,不同维度使用交叉注意力,通过对相同维度特征增强,不同维度特征交错融合提升检测器对不同尺寸目标的识别能力。最后,鉴于分类和定向边界框回归任务对特征的敏感性不同,本文提出解耦特征细化处理两个不同任务。通过实验,PDMDet在DOTA-v1.0、DOTA-v1.5和HRSC2016这3个数据集上分别取得单尺度78.37%、72.35%、98.60%的平均精度均值,检测准确率高于其他算法,在复杂的旋转目标检测场景具有一定的竞争力。展开更多
目的自动引导运输小车(automatic guided vehicles,AGV)在工厂中搬运货物时会沿着规定路线运行,但是在靠近障碍物时只会简单地自动停止,无法感知障碍物的具体位置和大小,为了让AGV小车在复杂的工业场景中检测出各种障碍物,提出了一个点...目的自动引导运输小车(automatic guided vehicles,AGV)在工厂中搬运货物时会沿着规定路线运行,但是在靠近障碍物时只会简单地自动停止,无法感知障碍物的具体位置和大小,为了让AGV小车在复杂的工业场景中检测出各种障碍物,提出了一个点云多尺度编码的单阶段3D目标检测网络(multi-scale encoding for single-stage 3D object detector from point clouds,MSE-SSD)。方法首先,该网络通过可学习的前景点下采样模块来对原始点云进行下采样,以精确地分割出前景点。其次,将这些前景点送入多抽象尺度特征提取模块进行处理,该模块能够分离出不同抽象尺度的特征图并对它们进行自适应地融合,以减少特征信息的丢失。然后,从特征图中预测出中心点,通过多距离尺度特征聚合模块将中心点周围的前景点按不同距离尺度进行聚合编码,得到语义特征向量。最后,利用中心点和语义特征向量一起预测包围框。结果MSE-SSD在自定义数据集中进行实验,多个目标的平均精度(average precision,AP)达到了最优,其中,在困难级别下空AGV分类、简单级别下载货AGV分类比排名第2的IASSD(learning highly efficient point-based detectors for 3D LiDAR point clouds)高出1.27%、0.08%,在简单级别下工人分类比排名第2的SA-SSD(structure aware single-stage 3D object detection from point cloud)高出0.71%。网络运行在单个RTX 2080Ti GPU上检测速度高达77帧/s,该速度在所有主流网络中排名第2。将训练好的网络部署在AGV小车搭载的开发板TXR上,检测速度达到了8.6帧/s。结论MSE-SSD在AGV小车避障检测方面具有较高的精确性和实时性。展开更多
基金supported by the National Natural Science Foundation of China(Nos.62276204 and 62203343)the Fundamental Research Funds for the Central Universities(No.YJSJ24011)+1 种基金the Natural Science Basic Research Program of Shanxi,China(Nos.2022JM-340 and 2023-JC-QN-0710)the China Postdoctoral Science Foundation(Nos.2020T130494 and 2018M633470).
文摘Drone-based small object detection is of great significance in practical applications such as military actions, disaster rescue, transportation, etc. However, the severe scale differences in objects captured by drones and lack of detail information for small-scale objects make drone-based small object detection a formidable challenge. To address these issues, we first develop a mathematical model to explore how changing receptive fields impacts the polynomial fitting results. Subsequently, based on the obtained conclusions, we propose a simple but effective Hybrid Receptive Field Network (HRFNet), whose modules include Hybrid Feature Augmentation (HFA), Hybrid Feature Pyramid (HFP) and Dual Scale Head (DSH). Specifically, HFA employs parallel dilated convolution kernels of different sizes to extend shallow features with different receptive fields, committed to improving the multi-scale adaptability of the network;HFP enhances the perception of small objects by capturing contextual information across layers, while DSH reconstructs the original prediction head utilizing a set of high-resolution features and ultrahigh-resolution features. In addition, in order to train HRFNet, the corresponding dual-scale loss function is designed. Finally, comprehensive evaluation results on public benchmarks such as VisDrone-DET and TinyPerson demonstrate the robustness of the proposed method. Most impressively, the proposed HRFNet achieves a mAP of 51.0 on VisDrone-DET with 29.3 M parameters, which outperforms the extant state-of-the-art detectors. HRFNet also performs excellently in complex scenarios captured by drones, achieving the best performance on the CS-Drone dataset we built.
文摘小而杂乱的物体交织在一起,在遥感图像中尤为常见,给目标检测带来了巨大挑战。在旋转目标检测任务中这个困难更加突出。鉴于此,本文提出了基于提案增强的解耦特征挖掘旋转检测器(decoupled feature mining rotational detector based on proposal enhancement,PDMDet)。首先,采用单阶段检测器取代两阶段检测器的提案生成网络,通过生成高质量提案以减少背景冗余。其次,在相同维度使用自注意力,不同维度使用交叉注意力,通过对相同维度特征增强,不同维度特征交错融合提升检测器对不同尺寸目标的识别能力。最后,鉴于分类和定向边界框回归任务对特征的敏感性不同,本文提出解耦特征细化处理两个不同任务。通过实验,PDMDet在DOTA-v1.0、DOTA-v1.5和HRSC2016这3个数据集上分别取得单尺度78.37%、72.35%、98.60%的平均精度均值,检测准确率高于其他算法,在复杂的旋转目标检测场景具有一定的竞争力。
文摘目的自动引导运输小车(automatic guided vehicles,AGV)在工厂中搬运货物时会沿着规定路线运行,但是在靠近障碍物时只会简单地自动停止,无法感知障碍物的具体位置和大小,为了让AGV小车在复杂的工业场景中检测出各种障碍物,提出了一个点云多尺度编码的单阶段3D目标检测网络(multi-scale encoding for single-stage 3D object detector from point clouds,MSE-SSD)。方法首先,该网络通过可学习的前景点下采样模块来对原始点云进行下采样,以精确地分割出前景点。其次,将这些前景点送入多抽象尺度特征提取模块进行处理,该模块能够分离出不同抽象尺度的特征图并对它们进行自适应地融合,以减少特征信息的丢失。然后,从特征图中预测出中心点,通过多距离尺度特征聚合模块将中心点周围的前景点按不同距离尺度进行聚合编码,得到语义特征向量。最后,利用中心点和语义特征向量一起预测包围框。结果MSE-SSD在自定义数据集中进行实验,多个目标的平均精度(average precision,AP)达到了最优,其中,在困难级别下空AGV分类、简单级别下载货AGV分类比排名第2的IASSD(learning highly efficient point-based detectors for 3D LiDAR point clouds)高出1.27%、0.08%,在简单级别下工人分类比排名第2的SA-SSD(structure aware single-stage 3D object detection from point cloud)高出0.71%。网络运行在单个RTX 2080Ti GPU上检测速度高达77帧/s,该速度在所有主流网络中排名第2。将训练好的网络部署在AGV小车搭载的开发板TXR上,检测速度达到了8.6帧/s。结论MSE-SSD在AGV小车避障检测方面具有较高的精确性和实时性。
