Real-time Detection Performance Research Articles

A remote sensing image rotation object detection approach for real-time environmental monitoring

For the problems of resources and environment faced in sustainable development, the use of remote sensing image object detection technology is an effective means to detect and analyze this major problem. Aiming at the problems of low accuracy and slow detection speed of existing remote sensing image rotation object detection algorithms implemented based on Transformer, a remote sensing image rotation object detection method based on dynamic position information Transformer is proposed. Firstly, to improve the detection accuracy, the cross-attention operation of the decoder is improved, and the obtained results are iteratively updated with the position information of the object queries and used as the initial object queries for the next decoder; secondly, to improve the robustness of the network for remote sensing image object detection, the data processing method of image pyramid is designed; Finally, a rotating IoU matching loss function more suitable for oriented object detection is introduced to improve the accuracy of matching the predicted boxed to the true boxed. The detection algorithm proposed in this paper is experimentally verified on DOTA and SSDD datasets, and the average detection accuracy is 73.70% and 90.3%, respectively, which effectively improves the average detection accuracy of Transformer-based rotating object detection algorithm in aerial remote sensing images and has better real-time detection performance.

MagicCubePose, A more comprehensive 6D pose estimation network

Most of the current mainstream 6D pose estimation methods use template or voting-based methods. Such methods are usually multi-stage or have multiple assumptions and post-correction, which will cause a certain degree of information redundancy and increase the computational cost, their real-time detection performance is poor. We point out that traditional path aggregation networks introduce new errors, therefore, we propose a loss function: MagicCubeLoss, a portable module: MagicCubeNet, and the corresponding 6D pose estimation model: MagicCubePose. MagicCubePose has good expansion performance and can build more efficient models for different calculation power and scenarios. Experiments show that our model has good real-time detection performance and the highest ADD(-S) accuracy.

A Quantitative Logarithmic Transformation-Based Intrusion Detection System

Intrusion detection systems (IDS) play a vital role in protecting networks from malicious attacks. Modern IDS use machine-learning or deep-learning models to deal with the diversity of attacks that malicious users may employ. However, effective machine-learning methods incur a considerable cost in both the pretraining stage and the online detection process itself. Accordingly, this study proposes a quantitative logarithmic transformation-based intrusion detection system (QLT-IDS) that uses a straightforward statistical approach to analyze network behavior. Compared with machine-learning or deep-learning-based IDS methods, the proposed system requires neither a time-consuming and expensive data collection and training process, nor a GPU-included device to achieve a real-time detection performance. Furthermore, the system can deal not only with North-South attacks, but also East-West attacks, which pose a significant risk in real-world operations. The effectiveness of the proposed system is evaluated for both real-world campus network traffic and simulated traffic. The results confirm that QLT-IDS is able to detect a wide range of malicious attacks with a high precision, even under high down-sampling rate of the NetFlow records.

A Novel GB-SAR System Based on TD-MIMO for High-Precision Bridge Vibration Monitoring

Ground-based synthetic aperture radar (GB-SAR) is a highly effective technique that is widely used in landslide and bridge deformation monitoring. GB-SAR based on multiple input multiple output (MIMO) technology can achieve high accuracy and real-time detection performance. In this paper, a novel method is proposed to design transmitting and receiving array elements, which increases the minimum spacing of the antenna by sacrificing several equivalent phase centers. In MIMO arrays, the minimum antenna spacing in the azimuth direction is doubled, which increases the variety of antenna options for this design. To improve the accuracy of the system, a new method is proposed to estimate channel phase errors, amplitude errors, and position errors. The position error is decomposed into three directions with one compensated by the phase error and two estimated by the strong point. Finally, we validate the accuracy of the system and our error estimation method through simulations and experiments. The results prove that the GB-SAR system performs well in bridge deformation and vibration monitoring with the proposed method.

Small-Object Detection for UAV-Based Images Using a Distance Metric Method

Object detection is important in unmanned aerial vehicle (UAV) reconnaissance missions. However, since a UAV flies at a high altitude to gain a large reconnaissance view, the captured objects often have small pixel sizes and their categories have high uncertainty. Given the limited computing capability on UAVs, large detectors based on convolutional neural networks (CNNs) have difficulty obtaining real-time detection performance. To address these problems, we designed a small-object detector for UAV-based images in this paper. We modified the backbone of YOLOv4 according to the characteristics of small-object detection. We improved the performance of small-object positioning by modifying the positioning loss function. Using the distance metric method, the proposed detector can classify trained and untrained objects through object features. Furthermore, we designed two data augmentation strategies to enhance the diversity of the training set. We evaluated our method on a collected small-object dataset; the proposed method obtained 61.00% mAP50 on trained objects and 41.00% mAP50 on untrained objects with 77 frames per second (FPS). Flight experiments confirmed the utility of our approach on small UAVs, with satisfying detection performance and real-time inference speed.

Tree Trunk and Obstacle Detection in Apple Orchard Based on Improved YOLOv5s Model

In this paper, we propose a tree trunk and obstacle detection method in a semistructured apple orchard environment based on improved YOLOv5s, with an aim to improve the real-time detection performance. The improvement includes using the K-means clustering algorithm to calculate anchor frame and adding the Squeeze-and-Excitation module and 10% pruning operation to ensure both detection accuracy and speed. Images of apple orchards in different seasons and under different light conditions are collected to better simulate the actual operating environment. The Gradient-weighted Class Activation Map technology is used to visualize the performance of YOLOv5s network with and without improvement to increase interpretability of improved network on detection accuracy. The detected tree trunk can then be used to calculate the traveling route of an orchard carrier platform, where the centroid coordinates of the identified trunk anchor are fitted by the least square method to obtain the endpoint of the next time traveling rout. The mean average precision values of the proposed model in spring, summer, autumn, and winter were 95.61%, 98.37%, 96.53%, and 89.61%, respectively. The model size of the improved model is reduced by 13.6 MB, and the accuracy and average accuracy on the test set are increased by 5.60% and 1.30%, respectively. The average detection time is 33 ms, which meets the requirements of real-time detection of an orchard carrier platform.

Bidirectional Statistical Feature Extraction Based on Time Window for Tor Flow Classification

The anonymous system Tor uses an asymmetric algorithm to protect the content of communications, allowing criminals to conceal their identities and hide their tracks. This malicious usage brings serious security threats to public security and social stability. Statistical analysis of traffic flows can effectively identify and classify Tor flow. However, few features can be extracted from Tor traffic, which have a weak representational ability, making it challenging to combat cybercrime in real-time effectively. Extracting and utilizing more accurate features is the key point to improving the real-time detection performance of Tor traffic. In this paper, we design an efficient and real-time identification scheme for Tor traffic based on the time window method and bidirectional statistical characteristics. In this paper, we divide the network traffic by sliding the time window and then calculate the relative entropy of the flows in the time window to identify Tor traffic. We adopt a sequential pattern mining method to extract bidirectional statistical features and classify the application types in the Tor traffic. Finally, extensive experiments are carried out on the UNB public dataset (ISCXTor2016) to validate our proposal’s effectiveness and real-time property. The experiment results show that the proposed method can detect Tor flow and classify Tor flow types with an accuracy of 93.5% and 91%, respectively, and the speed of processing and classifying a single flow is 0.05 s, which is superior to the state-of-the-art methods.

YOLO-Rip: A modified lightweight network for Rip currents detection

Rip currents form on beaches worldwide and pose a potential safety hazard for beach visitors. Therefore, effectively identifying rip currents from beach scenes and providing real-time alerts to beach managers and beachgoers is crucial. In this study, the YOLO-Rip model was proposed to detect rip current targets based on current popular deep learning techniques. First, based on the characteristics of a large target size in rip current images, the neck region in the YOLOv5s model was streamlined. The 80 × 80 feature map branches suitable for detecting small targets were removed to reduce the number of parameters, decrease the complexity of the model, and improve the real-time detection performance. Subsequently, we proposed adding a joint dilated convolutional (JDC) module to the lateral connection of the feature pyramid network (FPN) to expand the perceptual field, improve feature information utilization, and reduce the number of parameters, while keeping the model compact. Finally, the SimAM module, which is a parametric-free attention mechanism, was added to optimize the target detection accuracy. Several mainstream neural network models have been used to train self-built rip current image datasets. The experimental results show that (i) the detection results from different models using the same dataset vary greatly and (ii) compared with YOLOv5s, YOLO-Rip increased the mAP value by approximately 4% (to 92.15%), frame rate by 2.18 frames per second, and the model size by only 0.46 MB. The modified model improved the detection accuracy while keeping the model streamlined, indicating its efficiency and accuracy in the detection of rip currents.

Damage detection of grotto murals based on lightweight neural network

Non-contact damage detection of grotto murals is a challenging task, due to their minor cracks and subtle defects. In this paper, a lightweight neural network-based grotto mural damage detection algorithm, named Ghost-C3SE YOLOv5, is proposed, which can effectively reduce the network size while obtaining reasonable damage detection results. First, based on the YOLOv5 detection algorithm, the dimension of the convolution layer is reduced, by adjusting the network structure and integrating a Ghost module. Second, the channel attention mechanism is introduced into the feature extraction backbone network, to adjust the weight of each feature according to its importance, which can accelerate the convergence speed of the loss function during the model training process. Finally, the experiment results have shown that the lightweight model Ghost-C3SE YOLOv5, applied on the Pascal Voc dataset, reduces the number of parameters by about 22.55 million while ensuring the detection precision and recall rate. Also, the training time and the model size are reduced by 36.21% and 46.04%, respectively. Precision is increased by 1.29% and recall remains comparative, while the utilization rate of the GPU is improved by 41.55%. This has addressed the shortcomings of laborious process and low accuracy in manual detection of the grotto murals. Furthermore, a real-time detection performance of 44.86 FPS is achieved here for damage detection of murals in the YunGang Grottoes, which has overcome the drawbacks of high model complexity, high computational cost, slow detection speed and low memory utilization of various existing deep learning algorithms.

Highly stretchable, self-healable and self-adhesive polyzwitterion ionogels enabled with binary noncovalent interactions

Developing polyzwitterion ionogels with large stretchability, self-healability, and self-adhesion is crucial for ionic conductors in human-motion detection, while it is rarely realized in a single material. Here, a poly(zwitterion-acrylic acid) copolymer ionogel (PCIG) is fabricated by photopolymerization, among which solvated ionic liquids are tightly locked within a polyzwitterion matrix through binary hydrogen-bonded and electrostatic interactions. The PCIG exhibits a considerable stretchability of >850%, excellent fatigue resistance for 4000 cycles, and self-healability with a healing efficiency of >78%. The PCIG also shows a >80% high transparency, wide temperature tolerance, and excellent self-adhesion with various substrates. As a result, the PCIG can be used as a stretchable ionic conductor in wearable strain sensors, showing excellent sensing performance in real-time detection of human physiological movements. The study indicates that the developed polyzwitterion ionogels with attractive characteristics are an ideal candidate for wearable strain sensors that could tolerate wide temperatures and harsh environments.

DetectFormer: Category-Assisted Transformer for Traffic Scene Object Detection.

Object detection plays a vital role in autonomous driving systems, and the accurate detection of surrounding objects can ensure the safe driving of vehicles. This paper proposes a category-assisted transformer object detector called DetectFormer for autonomous driving. The proposed object detector can achieve better accuracy compared with the baseline. Specifically, ClassDecoder is assisted by proposal categories and global information from the Global Extract Encoder (GEE) to improve the category sensitivity and detection performance. This fits the distribution of object categories in specific scene backgrounds and the connection between objects and the image context. Data augmentation is used to improve robustness and attention mechanism added in backbone network to extract channel-wise spatial features and direction information. The results obtained by benchmark experiment reveal that the proposed method can achieve higher real-time detection performance in traffic scenes compared with RetinaNet and FCOS. The proposed method achieved a detection performance of 97.6% and 91.4% in AP50 and AP75 on the BCTSDB dataset, respectively.

A detection method for impact point water columns based on improved YOLO X

This paper proposes an improved method to accurately and expediently detect water columns at the shells’ impact point. The suggested method combines a lightweight depthwise convolutional neural network (MobileNet v3) with the You Only Look Once X (YOLO X) algorithm, namely, YOLO X-m (MobileNet v3) that aims to simplify the network’s structure. Specifically, we used a weighted average pooling network and a spatial pyramid pooling network comprising multiple convolutional layers to retain as many features as possible. Moreover, we improve the activation and loss functions to reduce network calculations and afford better precision as well as fast and accurate water column detection. The experimental results reveal that YOLO X-m (MobileNet v3) ensures a good detection performance and adaptability to various light intensities, distances, and multiple water columns. Compared with the original YOLO X-m model, the improved network model achieves a 75.76% frames per second improvement and a 71.11% capacity reduction, while its AP50 decreases by only 1.29%. The proposed method is challenged against the single shot multibox detector and various YOLO variants, revealing its appealing accuracy, real-time detection performance, and suitability for practical applications and projects.

Infrared Small Target Detection via Interpatch Correlation Enhancement and Joint Local Visual Saliency Prior

Small target detection is the primary technology for infrared search and tracking (IRST) systems and plays a vital role in practical applications. Existing algorithms have the following challenges: 1) insufficient local and nonlocal feature extraction and 2) imbalance between accuracy and real-time detection performance. In this study, a novel model for fast detection based on interpatch correlation enhancement (IPCE) and joint local visual saliency prior is proposed to overcome such issues. Regarding the correlation in interpatch dimension, the improved tensor nuclear norm is used to further extract the low-rank structure of the background tensor, which fully exploits the low-rank component and reduces the iteration times. Furthermore, with the hypothesis that the target is locally saliency, a prior model based on the visual saliency mechanism is proposed as the constraint of the target tensor. It effectively reduces the false detection of the sparse edge structure. In general, the proposed IPCE jointly exploits both local and nonlocal correlation of the original image, achieving robustness in different scenarios. Finally, the proposed model is solved by the alternating direction method of multipliers (ADMM). Experiments on seven datasets demonstrate that IPCE outperforms the state of the arts in terms of the balance between detection efficiency and accuracy.

Surface Defect Detection of Industrial Parts Based on YOLOv5

Industrial product quality inspection, a crucial procedure in industrial production, is crucial in assuring product yield. Product safety and quality inspections on industrial assembly lines are predominantly manual, and there is currently a dearth of safe and dependable inspection techniques. An improved surface defect detection approach based on YOLOv5 is proposed for the problem of surface flaws in industrial components in order to improve the quality detection effect of industrial production parts. To improve the effect of dense object detection, the image features are extracted by the convolutional network and enhanced by coordinate attention. BiFPN is utilized to fuse multi-scale features in order to lower the rate of missed detection and false detection for small target samples. The detectors from the Transformer structure are added to the complex problem of fine-grained detection to improve the predictability of challenging occurrences. According to the experimental findings, on the dataset for industrial parts defects, the proposed network increases the recall of the original algorithm in abnormal classes by 5.3%, reaching 91.6%. Its inference speed can approach 95FPS, indicating an improved real-time detection performance.

Deep Learning for the Improvement of Object Detection in Augmented Reality

In this article, we give a comprehensive overview of recent methods in object detection using deep learning and their uses in augmented reality. The objective is to present a complete understanding of these algorithms and how augmented reality functions and services can be improved by integrating these methods. We discuss in detail the different characteristics of each approach and their influence on real-time detection performance. Experimental analyses are provided to compare the performance of each method and make meaningful conclusions for their use in augmented reality. Two-stage detectors generally provide better detection performance, while single-stage detectors are significantly more time efficient and more applicable to real-time object detection. Finally, we discuss several future directions to facilitate and stimulate future research on object detection in augmented reality. Keywords: object detection, deep learning, convolutional neural network, augmented reality.

Avoiding transduction-induced heating in suspended microchannel resonators using piezoelectricity

Calorimetry of single biological entities remains elusive. Suspended microchannel resonators (SMRs) offer excellent performance for real-time detection of various analytes and could hold the key to unlocking pico-calorimetry experiments. However, the typical readout techniques for SMRs are optical-based, and significant heat is dissipated in the sensor, altering the measurement and worsening the frequency noise. In this manuscript, we demonstrate for the first time full on-chip piezoelectric transduction of SMRs on which we focus a laser Doppler vibrometer to analyze its effect. We demonstrate that suddenly applying the laser to a water-filled SMR causes a resonance frequency shift, which we attribute to a local increase in temperature. When the procedure is repeated at increasing flow rates, the resonance frequency shift diminishes, indicating that convection plays an important role in cooling down the device and dissipating the heat induced by the laser. We also show that the frequency stability of the device is degraded by the laser source. In comparison to an optical readout scheme, a low-dissipative transduction method such as piezoelectricity shows greater potential to capture the thermal properties of single entities.

A Performance-Sensitive Malware Detection System Using Deep Learning on Mobile Devices

Currently, Android malware detection is mostly performed on server side against the increasing number of malware. Powerful computing resource provides more exhaustive protection for app markets than maintaining detection by a single user. However, apart from the applications provided by the official market, apps from unofficial markets and third-party resources are always causing serious security threats to end-users. Meanwhile, it is a time-consuming task if the app is downloaded first and then uploaded to the server side for detection, because the network transmission has a lot of overhead. In addition, the uploading process also suffers from the security threats of attackers. Consequently, a last line of defense on mobile devices is necessary and much-needed. In this paper, we propose an effective Android malware detection system, MobiTive, leveraging customized deep neural networks to provide a real-time and responsive detection environment on mobile devices. MobiTive is a preinstalled solution rather than an app scanning and monitoring engine using after installation, which is more practical and secure. Original deep learning models cannot be directly deployed and executed on mobile devices due to various performance limitations, such as computation power, memory size, and energy. Therefore, we evaluate and investigate the following key points:(1) the performance of different feature extraction methods based on source code or binary code;(2) the performance of different feature type selections for deep learning on mobile devices;(3) the detection accuracy of different deep neural networks on mobile devices;(4) the real-time detection performance and accuracy on different mobile devices;(5) the potential based on the evolution trend of mobile devices' specifications; and finally we further propose a practical solution (MobiTive) to detect Android malware on mobile devices.

A modified YOLOv3 detection method for vision-based water surface garbage capture robot

To tackle the water surface pollution problem, a vision-based water surface garbage capture robot has been developed in our lab. In this article, we present a modified you only look once v3-based garbage detection method, allowing real-time and high-precision object detection in dynamic aquatic environments. More specifically, to improve the real-time detection performance, the detection scales of you only look once v3 are simplified from 3 to 2. Besides, to guarantee the accuracy of detection, the anchor boxes of our training data set are reclustered for replacing some of the original you only look once v3 prior anchor boxes that are not appropriate to our data set. By virtue of the proposed detection method, the capture robot has the capability of cleaning floating garbage in the field. Experimental results demonstrate that both detection speed and accuracy of the modified you only look once v3 are better than those of other object detection algorithms. The obtained results provide valuable insight into the high-speed detection and grasping of dynamic objects in complex aquatic environments autonomously and intelligently.

An Approach to Improve SSD through Skip Connection of Multiscale Feature Maps.

SSD (Single Shot MultiBox Detector) is one of the best object detection algorithms and is able to provide high accurate object detection performance in real time. However, SSD shows relatively poor performance on small object detection because its shallow prediction layer, which is responsible for detecting small objects, lacks enough semantic information. To overcome this problem, SKIPSSD, an improved SSD with a novel skip connection of multiscale feature maps, is proposed in this paper to enhance the semantic information and the details of the prediction layers through skippingly fusing high-level and low-level feature maps. For the detail of the fusion methods, we design two feature fusion modules and multiple fusion strategies to improve the SSD detector's sensitivity and perception ability. Experimental results on the PASCAL VOC2007 test set demonstrate that SKIPSSD significantly improves the detection performance and outperforms lots of state-of-the-art object detectors. With an input size of 300 × 300, SKIPSSD achieves 79.0% mAP (mean average precision) at 38.7 FPS (frame per second) on a single 1080 GPU, 1.8% higher than the mAP of SSD while still keeping the real-time detection speed.

Real-Time Detection for Cache Side Channel Attack using Performance Counter Monitor

Cache side channel attacks extract secret information by monitoring the cache behavior of a victim. Normally, this attack targets an L3 cache, which is shared between a spy and a victim. Hence, a spy can obtain secret information without alerting the victim. To resist this attack, many detection techniques have been proposed. However, these approaches have limitations as they do not operate in real time. This article proposes a real-time detection method against cache side channel attacks. The proposed technique performs the detection of cache side channel attacks immediately after observing a variation of the CPU counters. For this, Intel PCM (Performance Counter Monitor) and machine learning algorithms are used to measure the value of the CPU counters. Throughout the experiment, several PCM counters recorded changes during the attack. From these observations, a detecting program was implemented by using these counters. The experimental results show that the proposed detection technique displays good performance for real-time detection in various environments.