A Real-Time Small Target Vehicle Detection Algorithm with an Improved YOLOv5m Network Model

To address the challenges of high complexity,poor real-time performance,and low detection rates for small target vehicles in existing vehicle object detection algorithms,this paper proposes a real-time lightweight architecture based on You Only Look Once(YOLO)v5m.Firstly,a lightweight upsampling operator called Content-Aware Reassembly of Features(CARAFE)is introduced in the feature fusion layer of the network to maximize the extraction of deep-level features for small target vehicles,reducing the missed detection rate and false detection rate.Secondly,a new prediction layer for tiny targets is added,and the feature fusion network is redesigned to enhance the detection capability for small targets.Finally,this paper applies L1 regularization to train the improved network,followed by pruning and fine-tuning operations to remove redundant channels,reducing computational and parameter complexity and enhancing the detection efficiency of the network.Training is conducted on the VisDrone2019-DET dataset.The experimental results show that the proposed algorithmreduces parameters and computation by 63.8% and 65.8%,respectively.The average detection accuracy improves by 5.15%,and the detection speed reaches 47 images per second,satisfying real-time requirements.Compared with existing approaches,including YOLOv5m and classical vehicle detection algorithms,our method achieves higher accuracy and faster speed for real-time detection of small target vehicles in edge computing.

Short-time maritime target detection based on polarization scattering characteristics

In this paper,a detection method combining Cameron decomposition based on polarization scattering characteristics in sea clutter background is proposed.Firstly,the Cameron decomposition is exploited to fuse the radar echoes of full polarization channels at the data level.Due to the artificial material structure on the surface of the target,it can be shown that the non-reciprocity of the target cell is stronger than that of the clutter cell.Then,based on the analysis of the decomposition results,a new feature with scattering geometry characteristics in polarization domain,denoted as Cameron polarization decomposition scattering weight(CPD-SW),is extracted as the test statistic,which can achieve more detailed descriptions of the clutter scattering characteristics utilizing the difference between their scattering types.Finally,the superiority of the proposed CPD-SW detector over traditional detectors in improving detection performance is verified by the IPIX measured dataset,which has strong stability under short-time observation in threshold detection and can also improve the separability of feature space zin anomaly detection.

Improved Weighted Local Contrast Method for Infrared Small Target Detection

In order to address the problem of high false alarm rate and low probabilities of infrared small target detection in complex low-altitude background,an infrared small target detection method based on improved weighted local contrast is proposed in this paper.First,the ratio information between the target and local background is utilized as an enhancement factor.The local contrast is calculated by incorporating the heterogeneity between the target and local background.Then,a local product weighted method is designed based on the spatial dissimilarity between target and background to further enhance target while suppressing background.Finally,the location of target is obtained by adaptive threshold segmentation.As experimental results demonstrate,the method shows superior performance in several evaluation metrics compared with six existing algorithms on different datasets containing targets such as unmanned aerial vehicles(UAV).

Infrared Small Target Detection Algorithm Based on ISTD-CenterNet

This paper proposes a real-time detection method to improve the Infrared small target detection CenterNet(ISTD-CenterNet)network for detecting small infrared targets in complex environments.The method eliminates the need for an anchor frame,addressing the issues of low accuracy and slow speed.HRNet is used as the framework for feature extraction,and an ECBAM attention module is added to each stage branch for intelligent identification of the positions of small targets and significant objects.A scale enhancement module is also added to obtain a high-level semantic representation and fine-resolution prediction map for the entire infrared image.Besides,an improved sensory field enhancement module is designed to leverage semantic information in low-resolution feature maps,and a convolutional attention mechanism module is used to increase network stability and convergence speed.Comparison experiments conducted on the infrared small target data set ESIRST.The experiments show that compared to the benchmark network CenterNet-HRNet,the proposed ISTD-CenterNet improves the recall by 22.85%and the detection accuracy by 13.36%.Compared to the state-of-the-art YOLOv5small,the ISTD-CenterNet recall is improved by 5.88%,the detection precision is improved by 2.33%,and the detection frame rate is 48.94 frames/sec,which realizes the accurate real-time detection of small infrared targets.

RT-YOLO:A Residual Feature Fusion Triple Attention Network for Aerial Image Target Detection

In recent years,target detection of aerial images of unmannedaerial vehicle(UAV)has become one of the hottest topics.However,targetdetection of UAV aerial images often presents false detection and misseddetection.We proposed a modified you only look once(YOLO)model toimprove the problems arising in object detection in UAV aerial images:(1)A new residual structure is designed to improve the ability to extract featuresby enhancing the fusion of the inner features of the single layer.At the sametime,triplet attention module is added to strengthen the connection betweenspace and channel and better retain important feature information.(2)Thefeature information is enriched by improving the multi-scale feature pyramidstructure and strengthening the feature fusion at different scales.(3)A newloss function is created and the diagonal penalty term of the anchor frame isintroduced to improve the speed of training and the accuracy of reasoning.The proposed model is called residual feature fusion triple attention YOLO(RT-YOLO).Experiments showed that the mean average precision(mAP)ofRT-YOLO is increased from 57.2%to 60.8%on the vehicle detection in aerialimage(VEDAI)dataset,and the mAP is also increased by 1.7%on the remotesensing object detection(RSOD)dataset.The results show that theRT-YOLOoutperforms other mainstream models in UAV aerial image object detection.

CAFUNeT:A small infrared target detection method in complex backgrounds

Small infrared target detection has widespread applications in various fields including military,aviation,and medicine.However,detecting small infrared targets in complex backgrounds remains challenging.To detect small infrared targets,we propose a variable-structure U-shaped network referred as CAFUNet.A central differential convolution-based encoder,ASPP,an Attention Fusion module,and a decoder module are the critical components of the CAFUNet.The encoder module based on central difference convolution effectively extracts shallow detail information from infrared images,complemented by rich contextual information obtained from the deep features in the decoder module.However,the direct fusion of the shallow detail features with semantic features may lead to feature mismatch.To address this,we incorporate an Attention Fusion(AF)module to enhance the network performance further.We performed ablation studies on each module to evaluate its effectiveness.The results show that our proposed algorithm outperforms the state-of-the-art methods on publicly available datasets.

LF-CNN:Deep Learning-Guided Small Sample Target Detection for Remote Sensing Classification

Target detection of small samples with a complex background is always difficult in the classification of remote sensing images.We propose a new small sample target detection method combining local features and a convolutional neural network(LF-CNN)with the aim of detecting small numbers of unevenly distributed ground object targets in remote sensing images.The k-nearest neighbor method is used to construct the local neighborhood of each point and the local neighborhoods of the features are extracted one by one from the convolution layer.All the local features are aggregated by maximum pooling to obtain global feature representation.The classification probability of each category is then calculated and classified using the scaled expected linear units function and the full connection layer.The experimental results show that the proposed LF-CNN method has a high accuracy of target detection and classification for hyperspectral imager remote sensing data under the condition of small samples.Despite drawbacks in both time and complexity,the proposed LF-CNN method can more effectively integrate the local features of ground object samples and improve the accuracy of target identification and detection in small samples of remote sensing images than traditional target detection methods.

Infrared Image Small Target Detection Based on Bi-orthogonal Wavelet and Morphology

An image multi-scale edge detection method based on anti-symmetrical bi-orthogonal wavelet is given in theory. Convolution operation property and function as a differential operator are analyzed,which anti-symmetrical bi-orthogonal wavelet transform have. An algorithm for wavelet reconstruction in which multi-scale edge can be detected is put forward. Based on it, a detection method for small target in infrared image with sea or sky background based on the anti-symmetrical bi-orthogonal wavelet and morphology is proposed. The small target detection is considered as a process in which structural background is removed, correlative background is suppressed, and noise is restrained. In this approach, the multi-scale edge is extracted by means of the anti-symmetrical bi-orthogonal wavelet decomposition. Then, module maximum chains formed by complicated background of clouds, sea wave and sea-sky-line are removed, and the image background becomes smoother. Finally, the morphology based edge detection method is used to get small target and restrain undulate background and noise. Experiment results show that the approach can suppress clutter background and detect the small target effectively.

Novel detection method for infrared small targets using weighted information entropy

This paper presents a method for detecting the small infrared target under complex background. An algorithm, named local mutation weighted information entropy （LMWIE）, is proposed to suppress background. Then, the grey value of targets is enhanced by calculating the local energy. Image segmentation based on the adaptive threshold is used to solve the problems that the grey value of noise is enhanced with the grey value improvement of targets. Experimental results show that compared with the adaptive Butterworth high-pass filter method, the proposed algorithm is more effective and faster for the infrared small target detection.

A New Method of Small Moving Target Detection and Its Performance Analysis

This paper describes a new method of small moving target detection and analyzes the performance of this algorithm. The method is based on multi-level threshold decision-making and sliding trajectory confidence testing technology. The parameters of the algorithm are also given. Experiments have been conducted, the results show that the algorithm has advantages of high detection probability, simple structure, and excellent real-time performance.

Dim Moving Small Target Detection by Local and Global Variance Filtering on Temporal Profiles in Infrared Sequences

In this paper, the temporal different characteristics between the target and background pixels are used to detect dim moving targets in the slow-evolving complex background. A local and global variance filter on temporal profiles is presented that addresses the temporal characteristics of the target and background pixels to eliminate the large variation of background temporal profiles. Firstly, the temporal behaviors of different types of image pixels of practical infrared scenes are analyzed.Then, the new local and global variance filter is proposed. The baseline of the fluctuation level of background temporal profiles is obtained by using the local and global variance filter. The height of the target pulse signal is extracted by subtracting the baseline from the original temporal profiles. Finally, a new target detection criterion is designed. The proposed method is applied to detect dim and small targets in practical infrared sequence images. The experimental results show that the proposed algorithm has good detection performance for dim moving small targets in the complex background.

A Novel SAR Image Ship Small Targets Detection Method

To satisfy practical requirements of high real-time accuracy and low computational complexity of synthetic aperture radar (SAR) image ship small target detection, this paper proposes a small ship target detection method based on the improved You Only Look Once Version 3 (YOLOv3). The main contributions of this study are threefold. First, the feature extraction network of the original YOLOV3 algorithm is replaced with the VGG16 network convolution layer. Second, general convolution is transformed into depthwise separable convolution, thereby reducing the computational cost of the algorithm. Third, a residual network structure is introduced into the feature extraction network to reuse the shallow target feature information, which enhances the detailed features of the target and ensures the improvement in accuracy of small target detection performance. To evaluate the performance of the proposed method, many experiments are conducted on public SAR image datasets. For ship targets with complex backgrounds and small ship targets in the SAR image, the effectiveness of the proposed algorithm is verified. Results show that the accuracy and recall rate improved by 5.31% and 2.77%, respectively, compared with the original YOLOV3. Furthermore, the proposed model not only significantly reduces the computational effort, but also improves the detection accuracy of ship small target.

复杂背景下基于YOLOv7-tiny的图像目标检测算法

“黑飞”无人机一旦带有炸弹等物品,会对人们带来威胁。对在公园、游乐场、学校等复杂背景下“黑飞”的无人机进行目标检测是十分必要的。前沿算法YOLOv7-tiny属于轻量级网络,具有更小的网络结构和参数,更适合检测小目标,但在识别小目标无人机时出现特征提取能力弱、回归损失大、检测精度低的问题;针对此问题,提出了一种基于YOLOv7-tiny改进的无人机图像目标检测算法YOLOv7-drone。首先,建立无人机图像数据集;其次,设计一种新的注意力机制模块SMSE嵌入到特征提取网络中,增强对复杂背景下无人机目标的关注度;然后,在主干网络中融入RFB结构,扩大特征层的感受野,丰富特征信息以增强特征提取的鲁棒性;然后,改进网络中的特征融合机制,通过新增小目标检测层,增加对小尺度目标的检测精度;然后,改变损失函数提高模型的收敛速度,减少损失以增强模型的鲁棒性;最后,引入可变形卷积(Deformable convolution, DCN),更好的根据目标本身形状进行特征提取,提升了检测精度。在PASCAL VOC公共数据集上进行对比实验,结果表明改进后的算法YOLO7-drone相比于YOLOv7-tiny,平均精度(map@0.5)提升了6%;在自制无人机数据集上进行实验,结果表明YOLOv7-drone与原算法相比,平均精度(map@0.5)提高了6.1%,并且检测速度为72帧/s;与YOLOv5l、YOLOv7目标检测算法进行对比实验,结果表明改进后的算法在平均精度(map@0.5)上分别高于对比算法4%、3.1%,验证了文中算法的可行性。

基于WGAN-GP-CNN的海面小目标检测

针对传统基于统计理论的海面小目标检测方法在复杂海面环境中性能不高的问题,该文提出了一种改进的检测方法。首先通过分析海杂波和目标回波的特征,将检测问题转化为特征空间的分类任务。鉴于海面小目标样本数量有限,存在样本不平衡的问题,该文引入了一种基于梯度惩罚的沃瑟斯坦生成对抗网络(Wasserstein Generative Adversarial Network with Gradient Penalty,WGAN-GP)来增强目标数据,从而在数量上平衡目标样本与海杂波样本。同时,对原始WGAN-GP网络的损失函数进行了改进,引入相位损失以确保生成数据能够反映真实数据的相位信息。基于这些数据,进一步提取了生成目标和海杂波的高维特征,并将其送入卷积神经网络(Convolutional Neural Network,CNN)进行训练。为了应对高维特征空间中虚警概率难以控制的问题,对CNN算法进行了改进,通过设置Softmax分类器的阈值,实现了虚警概率可控。最后,借助公开的IPIX雷达数据集进行实验验证,所提的WGAN-GP-CNN检测器在积累时间为1.024 s,虚警概率为0.001时,平均检测概率达到0.8683,具有良好的检测效果。

Using deep learning to detect small targets in infrared oversampling images

According to the oversampling imaging characteristics, an infrared small target detection method based on deep learning is proposed. A 7-layer deep convolutional neural network(CNN) is designed to automatically extract small target features and suppress clutters in an end-to-end manner. The input of CNN is an original oversampling image while the output is a cluttersuppressed feature map. The CNN contains only convolution and non-linear operations, and the resolution of the output feature map is the same as that of the input image. The L1-norm loss function is used, and a mass of training data is generated to train the network effectively. Results show that compared with several baseline methods, the proposed method improves the signal clutter ratio gain and background suppression factor by 3–4 orders of magnitude, and has more powerful target detection performance.

基于cosSTR-YOLOv7的多尺度遥感小目标检测

针对地理空间遥感图像存在目标分布密集、尺度变化范围较大及小目标特征信息过少等而造成目标检测精度不高的问题,提出了一种基于Swin Transformer(STR)和YOLOv7的多尺度遥感小目标检测算法cosSTR-YOLOv7。以YOLOv7作为基线网络,首先,使用STR模块替换主干网络中的E-ELAN模块,并利用余弦注意力机制和后正则化方法将其改进为cosSTR模块,以提升模型训练的稳定性;其次,在Neck部分构建新的特征融合层,以减少特征信息丢失;然后,在预测部分增加小目标预测层,以提升模型对小目标的检测能力;最后,采用新的SIoU损失函数计算定位损失,以加快模型收敛速度。利用遥感数据集DIOR进行实验,实验结果表明,所提算法平均精度均值(mAP)达到92.63%,对比原YOLOv7算法提高了3.73个百分点,对多尺度小目标的检测性能有显著提高。

Detection Algorithm of Surface Defect Word on Printed Circuit Board

For Printed Circuit Board(PCB)surface defect detection,traditional detection methods mostly focus on template matching-based reference method and manual detections,which have the disadvantages of low defect detection efficiency,large errors in defect identification and localization,and low versatility of detectionmethods.In order to furthermeet the requirements of high detection accuracy,real-time and interactivity required by the PCB industry in actual production life.In the current work,we improve the Youonly-look-once(YOLOv4)defect detection method to train and detect six types of PCB small target defects.Firstly,the original Cross Stage Partial Darknet53(CSPDarknet53)backbone network is preserved for PCB defect feature information extraction,and secondly,the original multi-layer cascade fusion method is changed to a single-layer feature layer structure to greatly avoid the problem of uneven distribution of priori anchor boxes size in PCB defect detection process.Then,the K-means++clustering method is used to accurately cluster the anchor boxes to obtain the required size requirements for the defect detection,which further improves the recognition and localization of small PCB defects.Finally,the improved YOLOv4 defect detection model is compared and analyzed on PCB dataset with multi-class algorithms.The experimental results show that the average detection accuracy value of the improved defect detection model reaches 99.34%,which has better detection capability,lower leakage rate and false detection rate for PCB defects in comparison with similar defect detection algorithms.

基于Cache-DCN YOLOX算法的交通标志检测方法研究

针对传统方式识别交通标志算法存在的检测精度较低的问题,提出了一种基于Cache-DCN YOLOX算法的交通标志识别方法;在该方法中,使用DCN可变形卷积替换backbone中的普通卷积,有效地增大了模型的感受野,提高了特征提取能力;使用EIoU损失函数代替YOLOX中的GIoU损失函数,优化了训练模型,提高了收敛的速度;优化设计了YOLOX算法中的强弱两阶段的训练过程,增强了模型的泛化性能,同时加入cache方案,进一步提高了检测精度;在交通标志数据集TT100K上进行了实验,提出方法的检测精度为67.2%,比原YOLOX算法的检测精度提升了6.4%,同时,在被遮挡的小目标等多种受干扰的环境下,提出的方法能够精确地检测出交通标志,并有着较好的置信度,满足实际需求。

基于多目标采样和改进Mask R-CNN的木瓜成熟度检测

目的:提高木瓜成熟度检测准确率及鲁棒性。方法:提出一种基于多目标采样和改进Mask R-CNN的木瓜成熟度检测方法。通过均值平均精度、准确率、精确率—召回率曲线和计算时间等指标,验证所提方法的有效性和鲁棒性,并将其检测效果与Faster R-CNN、RetinaNet和CenterMask等方法进行对比。结果:试验方法对木瓜成熟度检测的平均精度均值、50%平均精度均值、75%平均精度均值分别为98.43%,98.67%,98.68%,对未成熟、半成熟和成熟木瓜成熟度的平均检测精度为99.38%,98.81%,99.37%。结论:该方法可用于开发木瓜成熟度检测的电子系统,提升木瓜成熟度检测和木瓜分级的性能。

Fusion network for small target detection based on YOLO and attention mechanism

Target detection is an important task in computer vision research, and such an anomaly detection and the topic of small target detection task is more concerned. However, there are still some problems in this kind of researches, such as small target detection in complex environments is susceptible to background interference and poor detection results. To solve these issues, this study proposes a method which introduces the attention mechanism into the you only look once(YOLO) network. In addition, the amateur-produced mask dataset was created and experiments were conducted. The results showed that the detection effect of the proposed mothed is much better.