Enhancing Dense Small Object Detection in UAV Images Based on Hybrid Transformer

Transformer-based models have facilitated significant advances in object detection.However,their extensive computational consumption and suboptimal detection of dense small objects curtail their applicability in unmanned aerial vehicle(UAV)imagery.Addressing these limitations,we propose a hybrid transformer-based detector,H-DETR,and enhance it for dense small objects,leading to an accurate and efficient model.Firstly,we introduce a hybrid transformer encoder,which integrates a convolutional neural network-based cross-scale fusion module with the original encoder to handle multi-scale feature sequences more efficiently.Furthermore,we propose two novel strategies to enhance detection performance without incurring additional inference computation.Query filter is designed to cope with the dense clustering inherent in drone-captured images by counteracting similar queries with a training-aware non-maximum suppression.Adversarial denoising learning is a novel enhancement method inspired by adversarial learning,which improves the detection of numerous small targets by counteracting the effects of artificial spatial and semantic noise.Extensive experiments on the VisDrone and UAVDT datasets substantiate the effectiveness of our approach,achieving a significant improvement in accuracy with a reduction in computational complexity.Our method achieves 31.9%and 21.1%AP on the VisDrone and UAVDT datasets,respectively,and has a faster inference speed,making it a competitive model in UAV image object detection.

Research on fast detection method of infrared small targets under resourceconstrained conditions

Infrared small target detection is a common task in infrared image processing.Under limited computa⁃tional resources.Traditional methods for infrared small target detection face a trade-off between the detection rate and the accuracy.A fast infrared small target detection method tailored for resource-constrained conditions is pro⁃posed for the YOLOv5s model.This method introduces an additional small target detection head and replaces the original Intersection over Union(IoU)metric with Normalized Wasserstein Distance(NWD),while considering both the detection accuracy and the detection speed of infrared small targets.Experimental results demonstrate that the proposed algorithm achieves a maximum effective detection speed of 95 FPS on a 15 W TPU,while reach⁃ing a maximum effective detection accuracy of 91.9 AP@0.5,effectively improving the efficiency of infrared small target detection under resource-constrained conditions.

Compressive sensing for small moving space object detection in astronomical images

It is known that detecting small moving objects in as- tronomical image sequences is a significant research problem in space surveillance. The new theory, compressive sensing, pro- vides a very easy and computationally cheap coding scheme for onboard astronomical remote sensing. An algorithm for small moving space object detection and localization is proposed. The algorithm determines the measurements of objects by comparing the difference between the measurements of the current image and the measurements of the background scene. In contrast to reconstruct the whole image, only a foreground image is recon- structed, which will lead to an effective computational performance, and a high level of localization accuracy is achieved. Experiments and analysis are provided to show the performance of the pro- posed approach on detection and localization.

Oriented Bounding Box Object Detection Model Based on Improved YOLOv8

In the study of oriented bounding boxes(OBB)object detection in high-resolution remote sensing images,the problem of missed and wrong detection of small targets occurs because the targets are too small and have different orientations.Existing OBB object detection for remote sensing images,although making good progress,mainly focuses on directional modeling,while less consideration is given to the size of the object as well as the problem of missed detection.In this study,a method based on improved YOLOv8 was proposed for detecting oriented objects in remote sensing images,which can improve the detection precision of oriented objects in remote sensing images.Firstly,the ResCBAMG module was innovatively designed,which could better extract channel and spatial correlation information.Secondly,the innovative top-down feature fusion layer network structure was proposed in conjunction with the Efficient Channel Attention(ECA)attention module,which helped to capture inter-local cross-channel interaction information appropriately.Finally,we introduced an innovative ResCBAMG module between the different C2f modules and detection heads of the bottom-up feature fusion layer.This innovative structure helped the model to better focus on the target area.The precision and robustness of oriented target detection were also improved.Experimental results on the DOTA-v1.5 dataset showed that the detection Precision,mAP@0.5,and mAP@0.5:0.95 metrics of the improved model are better compared to the original model.This improvement is effective in detecting small targets and complex scenes.

A Study on Enhancing Chip Detection Efficiency Using the Lightweight Van-YOLOv8 Network

In pursuit of cost-effective manufacturing,enterprises are increasingly adopting the practice of utilizing recycled semiconductor chips.To ensure consistent chip orientation during packaging,a circular marker on the front side is employed for pin alignment following successful functional testing.However,recycled chips often exhibit substantial surface wear,and the identification of the relatively small marker proves challenging.Moreover,the complexity of generic target detection algorithms hampers seamless deployment.Addressing these issues,this paper introduces a lightweight YOLOv8s-based network tailored for detecting markings on recycled chips,termed Van-YOLOv8.Initially,to alleviate the influence of diminutive,low-resolution markings on the precision of deep learning models,we utilize an upscaling approach for enhanced resolution.This technique relies on the Super-Resolution Generative Adversarial Network with Extended Training(SRGANext)network,facilitating the reconstruction of high-fidelity images that align with input specifications.Subsequently,we replace the original YOLOv8smodel’s backbone feature extraction network with the lightweight VanillaNetwork(VanillaNet),simplifying the branch structure to reduce network parameters.Finally,a Hybrid Attention Mechanism(HAM)is implemented to capture essential details from input images,improving feature representation while concurrently expediting model inference speed.Experimental results demonstrate that the Van-YOLOv8 network outperforms the original YOLOv8s on a recycled chip dataset in various aspects.Significantly,it demonstrates superiority in parameter count,computational intricacy,precision in identifying targets,and speed when compared to certain prevalent algorithms in the current landscape.The proposed approach proves promising for real-time detection of recycled chips in practical factory settings.

Methods and Means for Small Dynamic Objects Recognition and Tracking

A literature analysis has shown that object search,recognition,and tracking systems are becoming increasingly popular.However,such systems do not achieve high practical results in analyzing small moving living objects ranging from 8 to 14 mm.This article examines methods and tools for recognizing and tracking the class of small moving objects,such as ants.To fulfill those aims,a customized You Only Look Once Ants Recognition(YOLO_AR)Convolutional Neural Network(CNN)has been trained to recognize Messor Structor ants in the laboratory using the LabelImg object marker tool.The proposed model is an extension of the You Only Look Once v4(Yolov4)512×512 model with an additional Self Regularized Non–Monotonic(Mish)activation function.Additionally,the scalable solution for continuous object recognizing and tracking was implemented.This solution is based on the OpenDatacam system,with extended Object Tracking modules that allow for tracking and counting objects that have crossed the custom boundary line.During the study,the methods of the alignment algorithm for finding the trajectory of moving objects were modified.I discovered that the Hungarian algorithm showed better results in tracking small objects than the K–D dimensional tree(k-d tree)matching algorithm used in OpenDataCam.Remarkably,such an algorithm showed better results with the implemented YOLO_AR model due to the lack of False Positives(FP).Therefore,I provided a new tracker module with a Hungarian matching algorithm verified on the Multiple Object Tracking(MOT)benchmark.Furthermore,additional customization parameters for object recognition and tracking results parsing and filtering were added,like boundary angle threshold(BAT)and past frames trajectory prediction(PFTP).Experimental tests confirmed the results of the study on a mobile device.During the experiment,parameters such as the quality of recognition and tracking of moving objects,the PFTP and BAT,and the configuration parameters of the neural network and boundary line model were analyzed.The results showed an increased tracking accuracy with the proposed methods by 50%.The study results confirmed the relevance of the topic and the effectiveness of the implemented methods and tools.

Aerial multi-spectral AI-based detection system for unexploded ordnance

Unexploded ordnance(UXO)poses a threat to soldiers operating in mission areas,but current UXO detection systems do not necessarily provide the required safety and efficiency to protect soldiers from this hazard.Recent technological advancements in artificial intelligence(AI)and small unmanned aerial systems(sUAS)present an opportunity to explore a novel concept for UXO detection.The new UXO detection system proposed in this study takes advantage of employing an AI-trained multi-spectral(MS)sensor on sUAS.This paper explores feasibility of AI-based UXO detection using sUAS equipped with a single(visible)spectrum(SS)or MS digital electro-optical(EO)sensor.Specifically,it describes the design of the Deep Learning Convolutional Neural Network for UXO detection,the development of an AI-based algorithm for reliable UXO detection,and also provides a comparison of performance of the proposed system based on SS and MS sensor imagery.

Pre-locate net for object detection in high-resolution images

Small-object detection has long been a challenge.High-megapixel cameras are used to solve this problem in industries.However,current detectors are inefficient for high-resolution images.In this work,we propose a new module called Pre-Locate Net,which is a plug-and-play structure that can be combined with most popular detectors.We inspire the use of classification ideas to obtain candidate regions in images,greatly reducing the amount of calculation,and thus achieving rapid detection in high-resolution images.Pre-Locate Net mainly includes two parts,candidate region classification and behavior classification.Candidate region classification is used to obtain a candidate region,and behavior classification is used to estimate the scale of an object.Different follow-up processing is adopted according to different scales to balance the variance of the network input.Different from the popular candidate region generation method,we abandon the idea of regression of a bounding box and adopt the concept of classification,so as to realize the prediction of a candidate region in the shallow network.We build a high-resolution dataset of aircraft and landing gears covering complex scenes to verify the effectiveness of our method.Compared to state-of-the-art detectors(e.g.,Guided Anchoring,Libra-RCNN,and FASF),our method achieves the best m AP of 94.5 on 1920×1080 images at 16.7 FPS.

基于二阶段目标增强网络的低照度复杂环境下绝缘子故障检测方法

从低照度户外环境中航拍采集的绝缘子影像,存在照度低、背景复杂、绝缘子故障目标小等缺陷,严重影响低照度环境下绝缘子故障检测准确性。为解决上述问题,文章提出一种基于TOE-Net的低照度复杂环境下绝缘子故障检测方法,提出TOE-Net进行图像预处理方法,再使用YOLOv7-OL作为故障检测模块检测小目标绝缘子故障。在二阶段目标增强网络(two-stage object enhancement network,TOE-Net)中,设计零目标图像增强损失函数实现预增强网络(preparation enhancement network,PreEnNet)和深度增强网络(deep enhancement network,DeepEnNet)的无监督学习;使用信道级注意力模块跳跃式通道注意力机制(skip squeeze excitation networt,Skip_SENet)和跳跃式通道注意力机制(skip convolutional block attention module,Skip_CBAM)模块改进原始小目标特征增强单次多框检测算法(small object detection enhancement single shot multiBox detector,SDE-SSD),从而提升定位网络的小目标检测能力;设计弱监督机制使预增强网络根据小目标特征增强SSD的要求来提升图像增强能力,直到小目标特征增强SSD能够从增强图像中准确定位绝缘子串位置;使用深度增强网络深度增强绝缘子串区域,提升各类故障的特征显著性。故障检测模块中,将YOLOv7目标检测算法改进为面向小目标YOLOv7,在原模型中添加结合多尺度特征自适应融合网络的小目标检测通道,并将原始损失函数的CIOU改进为BIOU,从而提高模型的小目标检测性能。在低照度环境绝缘子故障检测实验中,该算法与5种目前常用目标检测算法相比具有较大优势,并且相较于低光目标检测算法IA-YOLO、GenISP with RetinaNet,m AP提升9.77%、10.35%,检测速度提升7.23%、10.16%,证明该算法适用于低照度复杂环境下小目标绝缘子故障检测任务;在正常光照绝缘子故障检测实验中该算法仍保持出色性能,证明该算法能够实现常规光照条件下绝缘子小目标故障检测。

基于改进YOLOv8的无人机航拍图像目标检测算法

针对现存无人机航拍图像目标检测算法检测精度较低、模型较为复杂的问题,提出一种改进YOLOv8的目标检测算法。在骨干网络引入多尺度注意力EMA,捕捉细节信息,以提高模型的特征提取能力;改进C2f模块,减小模型的计算量。提出了轻量级的Bi-YOLOv8特征金字塔网络结构改进YOLOv8的颈部,增强了模型多尺度特征融合能力,改善网络对小目标的检测精度。使用WIoU Loss优化原网络损失函数,引入一种动态非单调聚焦机制,提高模型的泛化能力。在无人机航拍数据集VisDrone2019上的实验表明,提出算法的mAP50为40.7%,较YOLOv8s提升了1.5%,参数量降低了42%,同时相比于其他先进的目标检测算法在精度和速度上均有提升,证明了改进算法的有效性和先进性。

改进YOLOv5s对病理学图像中猪只小肠绒毛的检测

为解决传统小肠绒毛需要专业人员手动检测耗时耗力且存在主观性和不稳定性等问题,同时提高在复杂病理学图像中小肠绒毛检测的准确率和效率,该研究提出基于改进YOLOv5s检测复杂病理学图像下猪只小肠绒毛的方法。首先,采用串联形式的混合池化对空间金字塔进行优化,增强特征提取与特征表达,提升检测精度;然后引入一种基于注意力机制的网络模块(simpleattentionmechanism,SimAM)与Bottleneck中的残差连接相结合,使用SimAM对Bottleneck中的特征图进行加权,得到加权后的特征表示,利用注意力机制加强模型对目标的感知。试验结果表明,该研究算法的平均精度(average precision)和每秒传输帧数(frame per second,FPS)达到92.43%和40帧/s。改进后的YOLOv5s在召回率和平均精度上相较改进前提高2.49和4.62个百分点,在不增加模型参数量的情况下,每帧图片的推理时间缩短1.04 ms。与经典的目标检测算法SSD、Faster R-CNN、YOLOv6s、YOLOX相比,平均精度分别提高15.16、10.56、2.03和4.07个百分点。结果表明,该方法能够实现病理学图像中猪只小肠绒毛自动化检测,保证复杂图像检测速度的同时,提高了小肠绒毛的检测精度。

MSH-YOLOv8:融合尺度重建的蘑菇小目标检测方法

[目的/意义]为了解决图像尺寸变化和目标尺度变换共存对小目标检测精度的影响问题,本研究提出了一种新的检测模型:Multi-Strategy Handling YOLOv8(MSH-YOLOv8)。[方法]该模型在YOLOv8的基础上增加一个检测头,以提高小尺度目标敏感度;引入Swin Transformer的检测结构到头部网络,以减少计算冗余;引入包含可变形卷积的C2f_Deformable Convolutionv4(C2f_DCNv4)结构和Swin Transformer编码器结构重构YOLOv8主干网络,优化并增强其特征传递和提取能力,提高小目标敏感度;采用基于规范化的注意力模块(Normalizationbased Attention Module,NAM)优化网络检测速度和准确性;用Wise-Intersection over Union Loss(WIoU)代替原损失函数,以提高训练效果和收敛速度;在后处理阶段应用分辨率动态训练、多尺度测试、软非极大值抑制算法(Soft-Non-Maximum Suppression,Soft-NMS)、加权边界框融合算法(Weighted Boxes Fusion,WBF)等方法,提高尺度变化下小目标检测效果。以蘑菇为研究对象,在开放数据集Fungi上开展实验。[结果和讨论]MSH-YOLOv8的平均正确率(Average Precision50,AP50)和AP@50-95分别达到了98.49%和75.29%,其中小目标检测指标值APs达39.73%。相较于主流模型YOLOv8,三项指标分别提高了2.34%,4.06%和8.55%;相较于优秀模型Transformer Prediction Heads-YOLOv5(TPH-YOLOv5),三项指标分别提高了2.14%,2.76%和6.89%。[结论]本研究提出的MSH-YOLOv8改进方法可在图像尺寸变化与目标尺度变化条件下有效提高小目标的检测效果。

基于YOLOv7改进的夜间樱桃检测方法:YOLOv7-Cherry

针对樱桃检测算法无法对夜晚环境下的樱桃进行成熟度识别的问题,提出一种改进的YOLOv7算法:YOLOv7-Cherry。使用一种将夜间樱桃图像和白天相同位置的樱桃图像相融合的图像预处理方法,保留夜间樱桃图像高空间分辨信息的同时加强其光谱分辨率。在YOLOv7-Cherry中,将CBAM注意力机制插入到骨干网络中,利用注意力机制强化神经网络的表征能力,强调重要特征,忽略次要特征,加强对樱桃目标特征的提取;为了加强目标检测算法对图像中小樱桃的识别,增加小目标检测层;改进了原始网络的初始检测框大小;为了减少遮挡对樱桃目标造成的损失,对检测框使用了Soft-NMS方法进行冗余去除。实验结果表明,YOLOv7-Cherry可以有效地识别出夜晚环境下的成熟樱桃和未成熟樱桃,与YOLOv3、Faster-RCNN、YOLOv4、YOLOv5和原YOLOv7相比,YOLOv7-Cherry的mAP提高了26.88、25.05、22.51、17.11和7.66个百分点,其中,识别精度、召回率、mAP和F1为93.9%、94.7%、97.4%、94.3%。

改进Faster R-CNN的遥感图像小目标检测算法

遥感图像目标检测是目标检测领域的一个关键问题,目前利用深度学习检测目标的算法大多在单向特征融合过程中添加注意力机制,一视同仁地去增强各类型的目标,并不能突出小目标。为了取得更好的检测效果,通过引入非对称高低层调制机制,构造兼顾低层细节信息和高层语义信息的特征图,以达到增强小目标特征检测的目的;同时使用DIoU损失函数代替原算法SmoothL1损失函数以提升算法检测精度与收敛速度;并且在感兴趣区域分类任务中引入灵活上下文信息以提高小目标分类准确性。实验结果表明,该算法在DIOR和NWPU VHR-10数据集上均取得了良好的表现。

基于YOLOv8n的航拍图像小目标检测算法

针对航拍图像小目标检测中存在目标密集和相互遮挡问题,提出一种基于YOLOv8n的航拍图像小目标检测算法。在主干网络末段,置换C2f中Bottleneck为改进后的FasterNet,保持通道数并提升收敛速度;替换SPPF中CBS激活函数SiLU为ReLU使输入负值置零,在CBS后引入SE注意力机制扩张感受野,保留更多小目标特征。输出端检测头前嵌入高效多尺度注意力机制EMA获取更多细节信息,进一步提高小目标关注度。将基线网络损失函数CIoU替换成Wise IoU,提供增益分配策略,专注普通质量锚框,提高网络泛化能力。在数据集VisDrone2021和RSOD上做消融实验和对比实验,相较于基线算法,mAP@0.5分别提升5.1和7.2个百分点,mAP@0.5:0.95分别提升4.4和2.1个百分点,表明检测精度指标显著提升;在公开数据集VOC2007+2012上做泛化实验,mAP@0.5提升3.8个百分点,表明具有良好的鲁棒性。

航拍图像小目标检测算法设计

针对传统检测算法在航拍图像小目标检测上准确率低,并存在误检、漏检等问题,提出一种基于YOLOv5的改进算法RBN-YOLOv5。设计基于RepVGG模块的C3RepBlock特征提取模块,增加小目标检测层更具判别性的浅层特征,通过局部和全局信息的联合表征获得更大的感受野;引入BiFormer注意力机制,提升模型检测精度,并基于归一化Wasserstein距离改进损失函数,增强小目标定位能力。在VisDrone2019数据集上的训练结果表明,RBN-YOLOv5相较于YOLOv5在检测精度上提高了9.8%,而且模型参数量大幅降低。

面向无人机图像场景的小目标检测模型

无人机因其飞行高度和拍摄角度的独特性,采集的遥感图像中存在大量小目标。小目标物体像素小、语义信息少,容易受背景信息干扰和出现聚集遮挡,是当前检测模型性能不佳的主要原因之一。提出一种面向小目标的无人机图像目标检测模型UAIDet(unmanned aerial vehicles images detector),从解决信息冲突和检测框回归难入手,提升模型的检测性能。其一,构建自适应的通道融合模块,在特征融合阶段动态学习通道权重以过滤不同尺度特征之间的信息冲突,抑制特征融合时的尺度不一致性,提高小目标物体的检测能力;其二,设计误差敏感定位损失函数,在小目标物体检测框的收敛阶段提出偏移敏感损失项以解决小目标对几何误差的敏感性,提高定位损失函数的鲁棒性,优化小目标物体的检测精度。在数据集Visdrone2022上对文章方法进行实验,mAP(means average precision)和AP50(average precision at IOU threshold 50%)分别达到了22.0%和37.1%,相较于基准模型分别提高3和4.7个百分点。TinyPerson数据集上的mAP和AP50为9.9%和29.1%,分别提高了4.29和4.2个百分点,证明UAIDet模型的有效性和鲁棒性。

难点注意力感知红外小目标检测网络

随着飞行器机动性能的提升,多帧红外小目标检测方法不足以满足检测要求。近年来,基于深度学习的单帧红外小目标检测方法取得了巨大成功。然而,红外小目标通常缺少形状特征,而且边界与背景模糊不清,给准确分割带来了一定的挑战。针对上述问题,本文提出难点注意力感知红外小目标检测网络。通过基于点的区域建议模块获取目标潜在区域,同时滤除多余背景。为实现高质量分割、细化掩码边界模块、判断粗掩码中无序、非局部难以分辨点,融合这些难点的多尺度特征,进行逐像素注意力建模。最后,由点检测头对难点注意力感知特征重新预测,生成精细分割掩码。在公开数据集NUDT-SIRST和IRDST上进行测试,平均精度均值mAP达到87.4和63.4,F值达到0.8935和0.7056。本文提出的难点注意力感知红外小目标检测网络可在多检测场景、多目标形态下实现准确分割,抑制误报信息,同时控制计算开销。

高分辨率遥感图像的目标检测

卫星遥感图像的分辨率高且目标在图像内的相对尺寸小,因此难以同时确保检测准确率和运行速度。为解决高像素遥感图像的目标检测问题,本文提出了一种结合滑动窗口分割和小目标检测器的检测方法。首先使用滑动窗口法将图像分割成多个子图,滑动步长略小于窗口的大小以使每个子图之间具有一定的重叠部分,并采用较大的分割窗口以降低子图数量。之后对子图进行压缩,使用目标检测算法处理压缩后的图片,降低算法运行时间。最后合并检测结果并采用非极大化抑制策略以去除在重叠部分重复检测的目标。在检测算法方面,本文以YOLOv8n为基础,使用SPD卷积核对网络结构进行改进,基于NWD方法调整正负样本匹配策略,并改进特征金字塔结构以提升算法对小目标的检测性能,从而使算法能够适应在更大尺寸下压缩的子图以减少图像分割数量,提升检测速度。实验证明,在图像平均分辨率为4000×4000的车辆检测数据集上,该方法对目标检测的平均准确率为55.7%,平均每张图片的计算时间约为47.5 ms,准确率比YOLOv8n提升16%,比YOLOv5s提升15%,比YOLOv6s提升7.6%。本文方法的运行效率满足实时化要求,能够以更高精度实时检测卫星遥感图像中的目标。

一种改进的基于深度学习的小目标检测方法

提出了一种改进的基于深度学习的小目标检测方法,用于解决当前主流算法针对小目标进行检测时输入图像需为小尺寸照片且模型参数过多等缺点的问题。为了解决这些问题,首先对图像进行预处理,将一张较大尺寸的图像按一定规则拆分成多张小尺寸图像后送入网络,克服了以往算法需要小尺寸图像才能进行检测的问题。对DNANet网络结构进行改进,减少其网络层数,提高了网络推断速度。使用TverskyLoss为像素分割的损失函数对损失函数进行优化,并采用渐进式学习法训练模型,使网络从普通目标到小目标的检测过程更为稳定。实验结果表明,该方法有效提升了深度学习在小目标大尺寸图像方面的收敛速度,改进后的网络对大尺寸图像的预测准确率提升了5%,预测时间缩短了25%。综上所述,提出的基于深度学习的小目标检测方法,可以方便地应用于工程实践中,并具有较高的实际应用价值。