Improve Detection Speed Research Articles

Neural Network Based on Multi-Scale Saliency Fusion for Traffic Signs Detection

Aiming at recognizing small-scale and complex traffic signs in the driving environment, a traffic sign detection algorithm YOLO-FAM based on YOLOv5 is proposed. Firstly, a new backbone network, ShuffleNet-v2, is used to reduce the algorithm’s parameters, realize lightweight detection, and improve detection speed. Secondly, the Bidirectional Feature Pyramid Network (BiFPN) structure is introduced to capture multi-scale context information, so as to obtain more feature information and improve detection accuracy. Finally, location information is added to the channel attention using the Coordinated Attention (CA) mechanism, thus enhancing the feature expression. The experimental results show that compared with YOLOv5, the mAP value of this method increased by 2.27%. Our approach can be effectively applied to recognizing traffic signs in complex scenes. At road intersections, traffic planners can better plan traffic and avoid traffic jams.

Real-Time Target Detection Method Based on Lightweight Convolutional Neural Network.

The continuous development of deep learning improves target detection technology day by day. The current research focuses on improving the accuracy of target detection technology, resulting in the target detection model being too large. The number of parameters and detection speed of the target detection model are very important for the practical application of target detection technology in embedded systems. This article proposed a real-time target detection method based on a lightweight convolutional neural network to reduce the number of model parameters and improve the detection speed. In this article, the depthwise separable residual module is constructed by combining depthwise separable convolution and non–bottleneck-free residual module, and the depthwise separable residual module and depthwise separable convolution structure are used to replace the VGG backbone network in the SSD network for feature extraction of the target detection model to reduce parameter quantity and improve detection speed. At the same time, the convolution kernels of 1 × 3 and 3 × 1 are used to replace the standard convolution of 3 × 3 by adding the convolution kernels of 1 × 3 and 3 × 1, respectively, to obtain multiple detection feature graphs corresponding to SSD, and the real-time target detection model based on a lightweight convolutional neural network is established by integrating the information of multiple detection feature graphs. This article used the self-built target detection dataset in complex scenes for comparative experiments; the experimental results verify the effectiveness and superiority of the proposed method. The model is tested on video to verify the real-time performance of the model, and the model is deployed on the Android platform to verify the scalability of the model.

Automated detection of knee cystic lesions on magnetic resonance imaging using deep learning.

BackgroundCystic lesions are frequently observed in knee joint diseases and are usually associated with joint pain, degenerative disorders, or acute injury. Magnetic resonance imaging-based, artificial intelligence-assisted cyst detection is an effective method to improve the whole knee joint analysis. However, few studies have investigated this method. This study is the first attempt at auto-detection of knee cysts based on deep learning methods.MethodsThis retrospective study collected data from 282 subjects with knee cysts confirmed at our institution from January to October 2021. A Squeeze-and-Excitation (SE) inception attention-based You only look once version 5 (SE-YOLOv5) model was developed based on a self-attention mechanism for knee cyst-like lesion detection and differentiation from knee effusions, both characterized by high T2-weighted signals in magnetic resonance imaging (MRI) scans. Model performance was evaluated via metrics including accuracy, precision, recall, mean average precision (mAP), F1 score, and frames per second (fps).ResultsThe deep learning model could accurately identify knee MRI scans and auto-detect both obvious cyst lesions and small ones with inconspicuous contrasts. The SE-YOLO V5 model constructed in this study yielded superior performance (F1 = 0.879, precision = 0.887, recall = 0.872, all class mAP0.5 = 0.944, effusion mAP = 0.945, cyst mAP = 0.942) and improved detection speed compared to a traditional YOLO model.ConclusionThis proof-of-concept study examined whether deep learning models could detect knee cysts and distinguish them from knee effusions. The results demonstrated that the classical Yolo V5 and proposed SE-Yolo V5 models could accurately identify cysts.

A defect detection method for topological phononic materials based on few-shot learning

Topological phononic materials have been widely used in many fields, such as topological antennas, asymmetric waveguides, and noise insulation. However, due to the limitations of the manufacturing process, topological protection is vulnerable to some severe defects that may affect the application effect. Therefore, the quality inspection of topological materials is essential to ensure reliable results. Due to the low contrast and irregularity of defects and the similarity of topological phononics, they are difficult to recognize by traditional image processing algorithms, so manual detection is still mainstream at present. But manual detection requires experienced inspectors, which is expensive and time-consuming. In addition, topological materials are expensive to produce, and there is no large publicly available dataset, but deep learning usually relies on large datasets for training. To solve the above problems, we propose an automatic deep learning topology structure defect detection method (ADLTSDM) in this work, which could classify not only the structure of topological materials but also detect the defects of topological phononics based on a small dataset. ADLTSDM exploits the prior knowledge of the topological material structure and achieves an augmentation factor of more than 100 times through the random and fixed interval screenshot algorithm, thus enabling the training of deep neural networks with only two raw data. For defect detection, ADLTSDM has an accuracy of more than 97% and improves detection speed by more than 38% compared with manual detection. For structure classification, ADLTSDM can achieve an accuracy of over 99% and seven times faster speed compared with manual classification. Besides, the detection standard of ADLTSDM is unified, so the accuracy will not be affected by the experience of the inspectors, which has more potential in high-throughput industrial applications.

Design and Acceleration of Field Programmable Gate Array-Based Deep Learning for Empty-Dish Recycling Robots

As the proportion of the working population decreases worldwide, robots with artificial intelligence have been a good choice to help humans. At the same time, field programmable gate array (FPGA) is generally used on edge devices including robots, and it greatly accelerates the inference process of deep learning tasks, including object detection tasks. In this paper, we build a unique object detection dataset of 16 common kinds of dishes and use this dataset for training a YOLOv3 object detection model. Then, we propose a formalized process of deploying a YOLOv3 model on the FPGA platform, which consists of training and pruning the model on a software platform, and deploying the pruned model on a hardware platform (such as FPGA) through Vitis AI. According to the experimental results, we successfully realize acceleration of the dish detection using a YOLOv3 model based on FPGA. By applying different sparse training and pruning methods, we test the pruned model in 18 different situations on the ZCU102 evaluation board. In order to improve detection speed as much as possible while ensuring detection accuracy, for the pruned model with the highest comprehensive performance, compared to the original model, the comparison results are as follows: the model size is reduced from 62 MB to 12 MB, which is only 19% of the origin; the number of parameters is reduced from 61,657,117 to 9,900,539, which is only 16% of the origin; the running time is reduced from 14.411 s to 6.828 s, which is only less than half of the origin, while the detection accuracy is decreased from 97% to 94.1%, which is only less than 3%.

Cloud-Based Architecture for YOLOv3 Object Detector using gRPC and Protobuf

The deep learning-based object detector accuracy has surpassed conventional detection methods. Although implementation is still limited to hardware capabilities, this problem can be overcome by combining edge devices with cloud computing. The recent study of cloud-based object detector architecture is generally based on representational state transfer (RESTful web services), which uses a pooling system method for data exchange. As a result, this system leads to a low detection speed and cannot support real-time data streaming. Therefore, this study aims to enhance the detection speed in cloud-based object recognition systems using gRPC and Protobuf to support real-time detection. The proposed architecture was deployed on the Virtual Machine Instance (VMI) equipped with a Graphics Processing Unit (GPU). The gRPC server and YOLOv3 deep learning object detector were executed on the cloud server to handle detection requests from edge devices. Furthermore, the captured images from the edge devices were encoded into Protobuf format to reduce the message size delivered to the cloud server. The results showed that the proposed architecture improved detection speed performance on the client-side in the range of 0.27 FPS to 1.72 FPS compared to the state-of-the-art method. It was also observed that it could support multiple edge devices connection with slight performance degradation in the range of 1.78 FPS to 1.83 FPS, depending on the network interface used.

Lightweight Deeplearning Method for Multi-vehicle Object Recognition

The recognition method based on deep learning has a large amount of calculation for the changes of different traffic densities in the actual traffic environment. In this paper, an integrated recognition method YOLOv4-L is proposed for reducing computational complexity based on the YOLOv4. The characteristics of multi-lane traffic flow with different flow densities were analyzed for statistical data sets, and k-means++ clustering algorithm was used to optimize the prior frame parameters to improve the matching degree between the prior frame. GhostNet was used to replace CSPDarknet53 of original network structure of YOLOv4 as the feature extraction network. The depthwise separable convolution module was introduced to replace the original 3×3 common convolution in feature extraction network, reduce model parameters and improve detection speed. The network model is further improved both with accuracy and robustness with the help of comprehensive method of Mosaic data enhancement, learning rate cosine annealing and label smoothing. Experimental results show that, Recognition speed is greatly improved at the expense of minimal recognition accuracy reduction: the recognition speed improvement value is 47.81%, 49.15% , 56.06% in detection speed (FPS), respectively in free flow, synchronous flow and blocked flow, the reduction value of accuracy is 2.21%, 0.67%,, 0.05% mAP, respectively.

Application of optimal clustering and metric learning to patch-based anomaly detection

Anomaly detection in computer vision is a process of identifying observations or events deviated from the norm. Pixel- and patch-based approaches for locating abnormal regions suffer from a tremendously expensive computational complexity due to repeated calculations inherently required. The approach introduced in this letter aims to reducing such computational burden by employing an optimal clustering method for features extracted from a deep network-based encoder, and a metric learning which provides a way to bias the clusters found by the clustering algorithm. A set of loss functions is defined and used for training the encoder. After the training steps, the clustering result obtained from the training dataset is saved and used in the inference phase by measuring distances between the features of the input data and the clusters. The proposed approach has been applied to a patch-based anomaly detection, Patch-SVDD, to clarify the effectiveness of the idea. The experiment carried out with the implementation on MVTec-AD dataset results in improved detection speed by 10∼35% and better detection accuracy for many image classes, as well. In addition, the ablation study conducted proves the validity of the ideas introduced in the proposed approach.

Yarn Density Measurement for 3-D Braided Composite Preforms Based on Rotation Object Detection

This paper examines how to accurately measure yarn density for three-dimensional braided composite preforms in large images. However, due to the complex structure and dense yarns of the preform, there is a lack of effective methods to solve this problem in practical applications. To the end, the task of yarn density measurement is mainly performed by experienced operators. They usually use a ruler to measure the distance between a fixed number of fabrics to calculate density, which is time-consuming and subjective. Therefore, an effective and efficient method for yarn density measurement of 3D braided composite preforms is demanded. To address this problem, we propose a novel framework combining deep learning with traditional image processing techniques to measure yarn density automatically. First, to ensure the accuracy of yarn detection, we obtain a high resolution raw image of 2448×2048. The pre-processed images are then fed into our Yarn Detection Network (YDNet) based on rotation object detection method to detect per segmented yarn. To achieve this, we propose a Dilated Feature Network (DFN) that can obtain features at different scales and improve detection speed. Then, the Feature Alignment Module (FAM) is introduced to achieve higher detection accuracy. Moreover, we design a Yarn Loss (YLoss) to detect yarn boxes more accurately via considering the aspect ratio of the yarns as an important parameter. Then, we design a multi-level rotation mechanism based on the extracted yarns to correct the image and calculate the yarn density accurately. We conduct several experiments to verify the superiority of our proposed YDNet and the accuracy and effectiveness of the proposed framework for the automatic yarn density measurement. The experimental results show that our method is superior to other state-of-the-art methods.

유해 아크 정량화를 위한 스마트 진단 프로그램 구현

Hazardous arc generated by electrical faults are detected for safety accidents in power facilities. Arc detection has a different measurement method depending on the faults. the arc flash from short circuit or overload is analyzed by voltage and current for diagnosis. However, there is a disadvantage in detecting the arc because of sensitivity to noise before it occur. In addition, detection method based on the photosensor may cause an malfunction by the light around environment. The photosensor with 220 nm can prevent malfunction because it not work in sun of the light. The arc size according to the detection speed is shown as the arc duration. This method reduces error according to the detection speed. However, it is necessary to implement a program capable of determining detection and accuracy determination for quantitative measurement. In this paper, we propose a detection method and program with improved detection speed to detect the arc duration of 10 μs. the error of program is compared with the arc duration generated in real time. In the future, we propose that customized application will be possible according to the factors of arc generation in the electric power field.

Extraction-free rapid cycle RT-qPCR and extreme RT-PCR for SARS-CoV-2 virus detection

Introduction/ObjectiveSince the start of the coronavirus disease 2019 (COVID-19) pandemic, molecular diagnostic testing for detection of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) has faced substantial supply chain shortages and noteworthy delays in result reporting after sample collection. Supply chain shortages have been most evident in reagents for RNA extraction and rapid diagnostic testing. In this study, we explored the kinetic limitations of extraction-free rapid cycle RT-qPCR for SARS-CoV-2 virus detection using the commercially available capillary based LightCycler.Methods/Case ReportWe optimized reverse transcription and PCR under extraction-free and rapid thermocycling conditions utilizing hydrolysis probe-based detection methods using a Roche LightCycler.Results (if a Case Study enter NA)This protocol improves detection speed while maintaining the sensitivity and specificity of hydrolysis probe-based detection. Percentage agreement between the developed assay and previously tested positive patient samples was 97.6% (n= 40/41) and negative patient samples was 100% (40/40). We further demonstrate that using purified RNA, SARS-CoV-2 testing using extreme RT-PCR and product verification by melting can be completed in less than 3 minutes.ConclusionWe developed a protocol for sensitive and specific RT-qPCR of SARS-CoV-2 RNA from nasopharyngeal swabs in less than 20 minutes, with minimal hands-on time requirements. Overall, these studies provide a framework for increasing the speed of SARS-CoV-2 and other infectious disease testing.

Crater Detection and Recognition Method for Pose Estimation

A crater detection and recognition algorithm is the key to pose estimation based on craters. Due to the changing viewing angle and varying height, the crater is imaged as an ellipse and the scale changes in the landing camera. In this paper, a robust and efficient crater detection and recognition algorithm for fusing the information of sequence images for pose estimation is designed, which can be used in both flying in orbit around and landing phases. Our method consists of two stages: stage 1 for crater detection and stage 2 for crater recognition. In stage 1, a single-stage network with dense anchor points (dense point crater detection network, DPCDN) is conducive to dealing with multi-scale craters, especially small and dense crater scenes. The fast feature-extraction layer (FEL) of the network improves detection speed and reduces network parameters without losing accuracy. We comprehensively evaluate this method and present state-of-art detection performance on a Mars crater dataset. In stage 2, taking the encoded features and intersection over union (IOU) of craters as weights, we solve the weighted bipartite graph matching problem, which is matching craters in the image with the previously identified craters and the pre-established craters database. The former is called “frame-frame match”, or FFM, and the latter is called “frame-database match”, or FDM. Combining the FFM with FDM, the recognition speed is enabled to achieve real-time on the CPU (25 FPS) and the average recognition precision is 98.5%. Finally, the recognition result is used to estimate the pose using the perspective-n-point (PnP) algorithm and results show that the root mean square error (RMSE) of trajectories is less than 10 m and the angle error is less than 1.5 degrees.

Extraction-Free Rapid Cycle Quantitative RT-PCR and Extreme RT-PCR for SARS-CoV-2 Virus Detection

Since the start of the coronavirus disease 2019 (COVID-19) pandemic, molecular diagnostic testing for detection of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) has faced substantial supply chain shortages and noteworthy delays in result reporting after sample collection. Supply chain shortages have been most evident in reagents for RNA extraction and rapid diagnostic testing. This study explored the kinetic limitations of extraction-free rapid cycle quantitative real-time RT-PCR for SARS-CoV-2 virus detection using the commercially available capillary-based LightCycler. After optimizing for time and reaction conditions, a protocol for sensitive and specific quantitative RT-PCR of SARS-CoV-2 RNA from nasopharyngeal swabs in <20 minutes was developed, with minimal hands-on time requirements. This protocol improves detection speed while maintaining the sensitivity and specificity of hydrolysis probe-based detection. Percentage agreement between the developed assay and previously tested positive patient samples was 97.6% (n = 40/41), and negative patient samples was 100% (40/40). The study further demonstrates that using purified RNA, SARS-CoV-2 testing using extreme RT-PCR, and product verification by melting can be completed in <3 minutes. Overall, these studies provide a framework for increasing the speed of SARS-CoV-2 and other infectious disease testing.

Digital Image Forensic based on Machine Learning approach for Forgery Detection and Localization

Machine learning for multimedia forensic is a new way of image forgery detection due to its amazing features of fast forgery detection. Compared with existing techniques of Deep Learning and Convolution Neural Network (“CNN”), machine learning improves security in the specific forged region under various test conditions. Some researchers use Support Vector Machine (“SVM”) and k-nearest neighbors (k-NN) algorithms to detect forgeries and another category uses unsupervised classification, including self-organization feature map (SOFM) and fuzzy c-means. But there occurs a need to address the detection speed improvement under the present scenario. The proposed algorithm has been developed using a machine learning approach to improve detection speed by pre-processing of feature extraction and feature reduction using “DWT” and “PCA” where data is trained by support vector machine (“SVM”) to provide quick results under various test conditions. This work specifies different image attacks like all types of geometric transformation, post-processing operations, etc., and presents efficiency in forgery detection and localization in case of multiple forgeries.

Ghost Imaging with Deep Learning for Position Mapping of Weakly Scattered Light Source

We propose ghost imaging (GI) with deep learning to improve detection speed. GI, which uses an illumination light with random patterns and a single-pixel detector, is correlation-based and thus suitable for detecting weak light. However, its detection time is too long for practical inspection. To overcome this problem, we applied a convolutional neural network that was constructed based on a classification of the causes of ghost image degradation. A feasibility experiment showed that when using a digital mirror device projector and a photodiode, the proposed method improved the quality of ghost images.

Real-time kiwifruit detection in orchard using deep learning on Android™ smartphones for yield estimation

Fast and accurate detection of kiwifruit in orchard under natural environment is the primary technology for yield estimation. Deep learning has become a prevalent way of fruit detection and achieved outstanding results. Besides, easy-carry smartphones are getting popular and powerful. In this paper, single shot multibox detector (SSD) with two lightweight backbones MobileNetV2 and InceptionV3 were employed to develop an Android APP named KiwiDetector for field kiwifruit detection. An 8-bit quantization method was used to compress model size and improve detection speed by quantizing weight tensor and activation function data of convolutional neural networks from 32-bit floating point to 8-bit integer. Detection test was performed on 100 selected kiwifruit field images with resolution of 3,968 × 2,976 pixels using the four models on a HUAWEI P20 smartphone. Results showed that MobileNetV2, quantized MobileNetV2, InceptionV3, and quantized InceptionV3 obtained true detected rate (TDR) of 90.8%, 89.7%, 87.6%, and 72.8%, respectively. The TDR of MobileNetV2 and quantized MobileNetV2 was generally consistent and higher than InceptionV3 and quantized InceptionV3. For processing an image on the smartphone, MobileNetV2, quantized MobileNetV2, InceptionV3, and quantized InceptionV3 took about 163 ms, 103 ms, 1085 ms, and 685 ms with model sizes of 17.5 MB, 4.5 MB, 96.1 MB, and 24.1 MB, respectively. Quantized MobileNetV2 reached a significant TDR with the fastest detection speed and the smallest model size. The results indicated that the proposed Android APP is promising for yield estimation.

Anchor-free Convolutional Network with Dense Attention Feature Aggregation for Ship Detection in SAR Images

In recent years, with the improvement of synthetic aperture radar (SAR) imaging resolution, it is urgent to develop methods with higher accuracy and faster speed for ship detection in high-resolution SAR images. Among all kinds of methods, deep-learning-based algorithms bring promising performance due to end-to-end detection and automated feature extraction. However, several challenges still exist: (1) standard deep learning detectors based on anchors have certain unsolved problems, such as tuning of anchor-related parameters, scale-variation and high computational costs. (2) SAR data is huge but the labeled data is relatively small, which may lead to overfitting in training. (3) To improve detection speed, deep learning detectors generally detect targets based on low-resolution features, which may cause missed detections for small targets. In order to address the above problems, an anchor-free convolutional network with dense attention feature aggregation is proposed in this paper. Firstly, we use a lightweight feature extractor to extract multiscale ship features. The inverted residual blocks with depth-wise separable convolution reduce the network parameters and improve the detection speed. Secondly, a novel feature aggregation scheme called dense attention feature aggregation (DAFA) is proposed to obtain a high-resolution feature map with multiscale information. By combining the multiscale features through dense connections and iterative fusions, DAFA improves the generalization performance of the network. In addition, an attention block, namely spatial and channel squeeze and excitation (SCSE) block is embedded in the upsampling process of DAFA to enhance the salient features of the target and suppress the background clutters. Third, an anchor-free detector, which is a center-point-based ship predictor (CSP), is adopted in this paper. CSP regresses the ship centers and ship sizes simultaneously on the high-resolution feature map to implement anchor-free and nonmaximum suppression (NMS)-free ship detection. The experiments on the AirSARShip-1.0 dataset demonstrate the effectiveness of our method. The results show that the proposed method outperforms several mainstream detection algorithms in both accuracy and speed.

Improving equivalence in fibrinogen evaluation between the prothrombin time-derived fibrinogen assay and Clauss method using a pooled plasma calibrator

AimsBoth the Clauss and prothrombin time-derived fibrinogen (PT-Fib) methods have widely been used in fibrinogen (Fib) evaluation. We aimed to improve the interchangeability between these two methods.MethodsThirty fresh plasmas of...

Real-time Traffic Sign Text Detection Based on Deep Learning

Aiming at the task of traffic sign text detection in natural scenes, a two-stage cascade detection model based on deep learning is proposed. The proposed model first locates the regions of interest (RoI) of text-based traffic sign by applying improved SSD (Single Shot MultiBox Detector) network. Then a rotation-based text detection network is used to detect text strings in the located RoI. Moreover, the lightweight convolutional neural network MobileNetV2 is combined with the deep learning component in two stages, which reduces network parameters and improves detection speed of the model. On the one hand, the proposed approach takes full advantage of the information of traffic signs. By this way the search area of text detection can be reduced, which makes it possible to simplify the text detector. On the other hand, it can obtain more complete text detection results by using rotation-based text detector. The experimental results show that the proposed method can perform well on different data sets, it can not only keep a high accuracy of detection, but also meets the realtime requirements.

Gaussian Scale-Space Enhanced Local Contrast Measure for Small Infrared Target Detection

Robust small-target detection plays an important role in the infrared (IR) search and track system, but it is still a challenge to detect small IR target under complex background. In this letter, an effective method inspired by the scale-space theory and the contrast mechanism of the human vision system is proposed. First, Gaussian scale-space (GSS) of an IR image is constructed by the convolution of a variable-scale Gaussian function. Second, the gray features of the local image can be directly represented by downsampling in a scale image, and enhanced local contrast measure (ELCM) is defined to enhance small target and suppress complex background. Then, the saliency map is obtained by using max-pooling operation, and an adaptive threshold is adapted to segment real targets. Experimental results on a test set with three real IR sequences demonstrate that the proposed method has a good performance in target enhancement and background suppression, and shows strong robustness under complex background. Especially, the proposed method has high computational efficiency, which can improve detection speed.