Multiple Detection Techniques Research Articles

Enhancing human computer interaction with coot optimization and deep learning for multi language identification

Human-Computer Interaction (HCI) is a multidisciplinary field focused on designing and utilizing computer technology, underlining the interaction interface between computers and humans. HCI aims to generate systems that allow consumers to relate to computers effectively, efficiently, and pleasantly. Multiple Spoken Language Identification (SLI) for HCI (MSLI for HCI) denotes the ability of a computer system to recognize and distinguish various spoken languages to enable more complete and handy interactions among consumers and technology. SLI utilizing deep learning (DL) involves using artificial neural networks (ANNs), a subset of DL models, to automatically detect and recognize the language spoken in an audio signal. DL techniques, particularly neural networks (NNs), have succeeded in various pattern detection tasks, including speech and language processing. This paper develops a novel Coot Optimizer Algorithm with a DL-Driven Multiple SLI and Detection (COADL-MSLID) technique for HCI applications. The COADL-MSLID approach aims to detect multiple spoken languages from the input audio regardless of gender, speaking style, and age. In the COADL-MSLID technique, the audio files are transformed into spectrogram images as a primary step. Besides, the COADL-MSLID technique employs the SqueezeNet model to produce feature vectors, and the COA is applied to the hyperparameter range of the SqueezeNet method. The COADL-MSLID technique exploits the SLID process’s convolutional autoencoder (CAE) model. To underline the importance of the COADL-MSLID technique, a series of experiments were conducted on the benchmark dataset. The experimentation validation of the COADL-MSLID technique exhibits a greater accuracy result of 98.33% over other techniques.

Labeling of CC Chemokine Receptor 2 with a Versatile Intracellular Allosteric Probe.

Interest in affinity-based probes (AfBPs) as novel tools to interrogate G protein-coupled receptors (GPCRs) has gained traction in recent years. AfBPs represent an interesting and more versatile alternative to antibodies. In the present study, we report the development and validation of AfBPs that target the intracellular allosteric pocket of CCR2, a GPCR of interest for the development of therapies targeting autoimmune and inflammatory diseases and also cancer. Owing to the two-step labeling process of these CCR2 AfBPs through the incorporation of a click handle, we were successful in applying our most efficient probe in a variety of in vitro experiments and making use of multiple different detection techniques, such as SDS-PAGE and LC/MS-based proteomics. Collectively, this novel probe shows high selectivity, versatility, and applicability. Hence, this is a valuable alternative for CCR2-targeting antibodies and other traditional tool compounds and could aid in target validation and engagement in drug discovery.

Detection of Crabs and Lobsters Using a Benchmark Single-Stage Detector and Novel Fisheries Dataset

Crabs and lobsters are valuable crustaceans that contribute enormously to the seafood needs of the growing human population. This paper presents a comprehensive analysis of single- and multi-stage object detectors for the detection of crabs and lobsters using images captured onboard fishing boats. We investigate the speed and accuracy of multiple object detection techniques using a novel dataset, multiple backbone networks, various input sizes, and fine-tuned parameters. We extend our work to train lightweight models to accommodate the fishing boats equipped with low-power hardware systems. Firstly, we train Faster R-CNN, SSD, and YOLO with different backbones and tuning parameters. The models trained with higher input sizes resulted in lower frames per second (FPS) and vice versa. The base models were highly accurate but were compromised in computational and run-time costs. The lightweight models were adaptable to low-power hardware compared to the base models. Secondly, we improved the performance of YOLO (v3, v4, and tiny versions) using custom anchors generated by the k-means clustering approach using our novel dataset. The YOLO (v4 and it’s tiny version) achieved mean average precision (mAP) of 99.2% and 95.2%, respectively. The YOLOv4-tiny trained on the custom anchor-based dataset is capable of precisely detecting crabs and lobsters onboard fishing boats at 64 frames per second (FPS) on an NVidia GeForce RTX 3070 GPU. The Results obtained identified the strengths and weaknesses of each method towards a trade-off between speed and accuracy for detecting objects in input images.

Prediction of multiple transverse cracks in a composite beam using hybrid RNN-mPSO technique

The presence of faults is common in engineering composite structures. Precise detection of faults prior to failure has been the focal point for researchers working in this domain. Detection of damage from the modal characteristics is an age-old promising technique besides the artificial intelligence techniques such as neural network (NN), Artificial neural network (ANN), recurrent neural network (RNN), and particle swarm optimization (PSO) methods. However, multiple detection techniques combined can give better results. In this regard, recurrent neural networks (RNN) and modified particle swarm optimization (mPSO) are combined to detect damages in a glass fiber reinforced polymer (GFRP) composite cantilever beam. The proposed hybrid model consists of two stages; indices of relative natural frequencies applied to locate crack parameters (position and severity) by recurrent neural network technique and optimization of the results (position and severity) by mPSO. The results illustrate this method can be implemented to predict multiple transverse cracks and their parameters in a composite beam, manifested in the change of natural frequencies where the error is limited to 0.97% only.

Aerosol Vertical Structure and Optical Properties during Two Dust and Haze Episodes in a Typical Valley Basin City, Lanzhou of Northwest China

The vertical profiles of aerosol optical properties are vital to clarify their transboundary transport, climate forcing and environmental health influences. Based on synergistic measurements of multiple advanced detection techniques, this study investigated aerosol vertical structure and optical characteristics during two dust and haze events in Lanzhou of northwest China. Dust particles originated from remote deserts traveled eastward at different altitudes and reached Lanzhou on 10 April 2020. The trans-regional aloft (~4.0 km) dust particles were entrained into the ground, and significantly modified aerosol optical properties over Lanzhou. The maximum aerosol extinction coefficient (σ), volumetric depolarization ratio (VDR), optical depth at 500 nm (AOD500), and surface PM10 and PM2.5 concentrations were 0.4~1.5 km−1, 0.15~0.30, 0.5~3.0, 200~590 μg/m3 and 134 μg/m3, respectively, under the heavy dust event, which were 3 to 11 times greater than those at the background level. The corresponding Ångström exponent (AE440–870), fine-mode fraction (FMF) and PM2.5/PM10 values consistently persisted within the ranges of 0.10 to 0.50, 0.20 to 0.50, and 0.20 to 0.50, respectively. These findings implied a prevailing dominance of coarse-mode and irregular non-spherical particles. A severe haze episode stemming from local emissions appeared at Lanzhou from 30 December 2020 to 2 January 2021. The low-altitude transboundary transport aerosols seriously deteriorated the air quality level in Lanzhou, and aerosol loading, surface air pollutants and fine-mode particles strikingly increased during the gradual strengthening of haze process. The maximum AOD500, AE440–870nm, FMF, PM2.5 and PM10 concentrations, and PM2.5/PM10 were 0.65, 1.50, 0.85, 110 μg/m3, 180 μg/m3 and 0.68 on 2 January 2021, respectively, while the corresponding σ and VDR at 0.20–0.80 km height were maintained at 0.68 km−1 and 0.03~0.12, implying that fine-mode and spherical small particles were predominant. The profile of ozone concentration exhibited a prominent two-layer structure (0.60–1.40 km and 0.10–0.30 km), and both concentrations at two heights always remained at high levels (60~72 μg/m3) during the entire haze event. Conversely, surface ozone concentration showed a significant decrease during severe haze period, with the peak value of 20~30 μg/m3, which was much smaller than that before haze pollution (~80 μg/m3 on 30 December). Our results also highlighted that the vertical profile of aerosol extinction coefficient was a good proxy for evaluating mass concentrations of surface particulate matters under uniform mixing layers, which was of great scientific significance for retrieving surface air pollutants in remote desert or ocean regions. These statistics of the aerosol vertical profiles and optical properties under heavy dust and haze events in Lanzhou would contribute to investigate and validate the transboundary transport and radiative forcing of aloft aerosols in the application of climate models or satellite remote sensing.

Ensemble Learning-based Multiple Sclerosis Detection Technique Using Magnetic Resonance Imaging

Multiple sclerosis (MS) is a complicated neurological disorder that leads to demyelination of nerve fibers in the central nervous system, causing severe symptoms and gradual impairment. Prompt and precise diagnosis of MS is essential for prompt intervention and individualized treatment planning. This research presents a new method for detecting MS: magnetic resonance imaging (MRI) data. Utilizing current progress in deep learning and ensemble learning methodologies, we use SWIN transformer and MobileNetV3-small for extracting features from MRI images. These features are then used for classification using CatBoost, XGBoost, and random forest algorithms. The suggested framework is tested and confirmed effective using the Kaggle MS database, which consists of various MRI images. The experimental findings show a remarkable average accuracy of 99.8% and a little loss of 0.07, highlighting the effectiveness of the suggested strategy in discriminating between aberrant and normal MRI pictures that indicate MS. This study enhances the field of medical image analysis by providing a precise and effective framework for automated diagnosis of MS. This framework has the potential to enhance diagnostic efficiency and improve patient outcomes. Combining deep learning feature extraction with ensemble classifiers offers a robust and easily understandable approach for diagnosing MS and has the potential to be used in clinical settings. Future research should prioritize validating the suggested technique on more extensive datasets and incorporating it into clinical practice to enhance early identification of MS and provide individualized patient treatment.

Establishing an integrated, four-dimensional quality assessment system for traditional Chinese medicine: A case study of Shuanghuanglian oral liquid

The quality of traditional Chinese medicine (TCM) is the premise to ensure its safety and effectiveness in clinical application. In this study, a complete quality control system for four-dimensional fingerprinting of TCM was innovatively constructed based on multiple detection techniques, and the quality of Shuanghuanglian oral liquid (SHL) was evaluated. Electrochemical fingerprinting (ECFP) as an emerging method without pretreatment provides rich and quantifiable information for SHL samples. The first quantitative ECFP of SHL was developed by the B-Z oscillation system. Eight characteristic parameters were analyzed and a good linear relationship was found between the oscillation lifetime and sample volume, by which the calculated values of the added sample volume (VL) showed different fluctuations between samples. What is more, high-performance liquid chromatography five-wavelength fusion fingerprint (HPLC-FWFP), GC fingerprint (GC-FP), and UV quantum fingerprint (UV-QFP) was established. Meanwhile, the purity of the peaks of the HPLC-FWFP was verified by the dual-wavelength absorption coefficient ratio spectrum (DWAR). Equal weighted ratio quantitative fingerprinting method (EWRQFM) was successfully proposed to extract all potential features for the overall quality assessment of the samples. Finally, a comprehensive evaluation strategy was proposed, namely the variation coefficient weighting algorithm (VCWA). The results of qualitative and quantitative evaluation of HPLC-FWFP, GC-FP, electrochemical quantum fingerprints (EC-QFP), and UV-QFP were integrated by this method. The established evaluation system is also a suitable strategy to control the quality of other TCM preparations.

The Possibility of Detecting our Solar System through Astrometry

Searching for exoplanets with different methods has always been the focus of astronomers over the past few years. Among multiple planet detection techniques, astrometry stands out for its capability to accurately determine the orbital parameters of exoplanets. In this study, we examine the likelihood of extraterrestrial intelligent civilizations detecting planets in our solar system using the astrometry method. By conducting injection-recovery simulations, we investigate the detectability of the four giant planets in our solar system under different observing baselines and observational errors. Our findings indicate that extraterrestrial intelligence could detect and characterize all four giant planets, provided they are observed for a minimum of 90 yr with signal-noise ratios exceeding 1. For individual planets such as Jupiter, Saturn, and Neptune, a baseline that surpasses half of their orbital periods is necessary for detection. However, Uranus requires longer observing baselines since its orbital period is roughly half of that of Neptune. If the astrometry precision is equal to or better than 10 μas, all 8707 stars located within 30 pc of our solar system possess the potential to detect the four giant planets within 100 yr. Additionally, our prediction suggests that over 300 stars positioned within 10 pc from our solar system could detect our Earth if they achieve an astrometry precision of 0.3 μas.

Multi-Object Detection and Tracking Using Reptile Search Optimization Algorithm with Deep Learning

Multiple-Object Tracking (MOT) has become more popular because of its commercial and academic potential. Though various techniques were devised for managing this issue, it becomes a challenge because of factors such as severe object occlusions and abrupt appearance changes. Tracking presents the optimal outcomes whenever the object moves uniformly without occlusion and in the same direction. However, this is generally not a real scenario, particularly in complicated scenes such as dance events or sporting where a greater number of players are tracked, moving quickly, varying their speed and direction, along with distance and position from the camera and activity they are executing. In dynamic scenes, MOT remains the main difficulty due to the symmetrical shape, structure, and size of the objects. Therefore, this study develops a new reptile search optimization algorithm with deep learning-based multiple object detection and tracking (RSOADL–MODT) techniques. The presented RSOADL–MODT model intends to recognize and track the objects that exist with position estimation, tracking, and action recognition. It follows a series of processes, namely object detection, object classification, and object tracking. At the initial stage, the presented RSOADL–MODT technique applies a path-augmented RetinaNet-based (PA–RetinaNet) object detection module, which improves the feature extraction process. To improvise the network potentiality of the PA–RetinaNet method, the RSOA is utilized as a hyperparameter optimizer. Finally, the quasi-recurrent neural network (QRNN) classifier is exploited for classification procedures. A wide-ranging experimental validation process takes place on DanceTrack and MOT17 datasets for examining the effectual object detection outcomes of the RSOADL–MODT algorithm. The simulation values confirmed the enhancements of the RSOADL–MODT method over other DL approaches.

Establishment of a multiplex polymerase chain reaction detection assay for three common harmful microalgae in the East China Sea.

It is urgent to develop techniques that can simultaneously detect multiple microalgae, due to the diversity of harmful algal blooms (HABs)-forming algal species. The target algae species in this study are Heterosigma akashiwo, Prorocentrum donghaiense and Karenia mikimotoi. These algae are the dominant species that cause HABs in the East China Sea, and the multiple detection technique focusing on these three algae is not common. Therefore, this study established a multiplex polymerase chain reaction(mPCR) to diagnose the three algae, which is simple and low cost. First, the corresponding specific primers were designed based on the D1-D2 region of the large subunit (LSU) ribosomal DNA sequence. Then, mPCR was established and the reaction conditions were optimized. And the specificity, sensitivity, and stability of mPCR were evaluated. The result of specificity test showed that the established mPCR had good specificity for the target microalgae and did not cross-react with eighteen non-target microalgae. The sensitivity of experiment was 3.3 × 10-1ng μL-1, and the established mPCR was not affected by the interfering microalgae. Moreover, the practicability evaluation of mPCR by using the simulated natural water samples showed that the detection limit of target microalgae was 100 cells mL-1, which could meet the demand for early warning of HABs. In summary, the established mPCR is characterized by strong specificity, good stability, and multiple analysis to detect H. akashiwo, P. donghaiense, and K. mikimotoi.

Class-associative multiple target recognition for highly compressed color images in a joint transform correlator

We propose a practical near real-time compression-based class-associative multiple target detection technique for color images. The size of the color images is reduced to its utmost ratio to speed up the processing time and to increase the storage capacity of the recognition system. A fringe-adjusted joint transform correlation technique is employed to successfully detect the compression-based multiple targets in colored images. In addition, to eliminate the false alarms and zero-order terms due to multiple desired and undesired objects in an input scene, we have used the shifted phase-encoded and the reference phase-encoded techniques. The performance of detecting the class-associative multiple targets for large compression ratios (up to 94%) and under strong noisy conditions (Gaussian and salt and pepper noises) is assessed through many computer simulations experiments. Furthermore, detecting small targets or even a small part of a target is presented for images occluded up to 90%.

Quantitative Assessment Methods of Early Enamel Caries with Optical Coherence Tomography: A Review

Early detection of caries is an urgent problem in the dental clinic. Current caries detection methods do not detect early enamel caries accurately, and do not show microstructural changes in the teeth. Optical coherence tomography (OCT) can provide imaging of tiny, demineralized regions of teeth in real time and noninvasively detect dynamic changes in lesions with high resolution and high sensitivity. Over the last 20 years, researchers have investigated different methods for quantitative assessment of early caries using OCT. This review provides an overview of the principles of enamel caries detection with OCT, the methods of characterizing caries lesion severity, and correlations between OCT results and measurements from multiple histological detection techniques. Studies have shown the feasibility of OCT in quantitative assessment of early enamel lesions but they vary widely in approaches. Only integrated reflectivity and refractive index measured by OCT have proven to have strong correlations with mineral loss calculated by digital microradiography or transverse microradiography. OCT has great potential to be a standard inspection method for enamel lesions, but a consensus on quantitative methods and indicators is an important prerequisite. Our review provides a basis for future discussions.

KRYSTAL: Knowledge graph-based framework for tactical attack discovery in audit data

Attack graph-based methods are a promising approach towards discovering attacks and various techniques have been proposed recently. A key limitation, however, is that approaches developed so far are monolithic in their architecture and heterogeneous in their internal models. The inflexible custom data models of existing prototypes and the implementation of rules in code rather than declarative languages on the one hand make it difficult to combine, extend, and reuse techniques, and on the other hand hinder reuse of security knowledge – including detection rules and threat intelligence. KRYSTAL tackles these challenges by providing a knowledge graph-based, modular framework for threat detection, attack graph and scenario reconstruction, and analysis based on RDF as a standard model for knowledge representation. This approach provides query options that facilitate contextualization over internal and external background knowledge, as well as the integration of multiple detection techniques, including tag propagation, attack signatures, and graph queries. We implemented our framework in an openly available prototype and demonstrate its applicability on multiple scenarios of the DARPA Transparent Computing dataset. Our evaluation shows that the combination of different threat detection techniques within our framework improved detection capabilities. Furthermore, we find that RDF provenance graphs are scalable and can efficiently support a variety of threat detection techniques.

On optimal segmentation and parameter tuning for multiple change-point detection and inference

Change-point analysis is the task of finding abrupt (and significant) changes in the underlying model of a signal or time series. Change-point detection methods typically involve specifying the maximum number of segments to search for and the minimum segment length. However, there is no objective way to pre-specify these two parameters, and it mostly depends upon the particular application. Within this framework, a recursive optimization algorithm is developed that is capable of exploring and fine tuning these two input parameters, and optimally segmenting a time series. This multiple change-point detection technique therefore addresses a wide class of real-life contexts and problems where the identification of optimal level shifts in a time series is the main goal. Extensive simulation results are presented and a real-life example is given to illustrate the implementation of the developed scheme in practice and to unfold its capabilities. Concluding remarks and suggestions for future research are also provided.

Dual-detection approach for a charge variant analysis of monoclonal antibody combination products using imaged capillary isoelectric focusing.

The clinical benefits of treatments with a combination of two or more therapeutic monoclonal antibodies (mAbs) have emerged in recent years. Imaged capillary isoelectric focusing is a frequently used technology in the biopharmaceutical industry for charge variant analysis of protein therapeutics. However, with the wide concentration ranges of combination products, one component may fall within the linear detection range, whereas the other does not. Here, we report a novel methodology to explore charge variants of mAb mixtures using multiple detection techniques simultaneously. We use ultraviolet absorbance to monitor the charge variants of the high-concentration component and native fluorescence (FL) to monitor the variants of the low-concentration one. Charge variants of mixtures that span 40-fold in ratio differences can be accurately quantified with this approach. In contrast to the conventional methods, it is not necessary to prepare and analyze two samples at different concentrations and combine the results for combination product testing. Additionally, the use of FL detection enables the charge variant analysis of highly potent/low abundant mAbs in a mixture. This methodology is more quality-control friendly and efficient for the charge variant analysis of combination products with wide ratios.

A Comparative Study of Various Edge Detection Techniques for Underwater Images

Nowadays, underwater image identification is a challenging task for many researchers focusing on various applications, such as tracking fish species, monitoring coral reef species, and counting marine species. Because underwater images frequently suffer from distortion and light attenuation, pre-processing steps are required in order to enhance their quality. In this paper, we used multiple edge detection techniques to determine the edges of the underwater images. The pictures were pre-processed with the use of specific techniques, such as enhancement processing, Wiener filtering, median filtering and thresholding. Coral reef pictures were used as a dataset of underwater images to test the efficiency of each edge detection method used in the experiment. All coral reef image datasets were captured using an underwater GoPro camera. The performance of each edge detection technique was evaluated using mean square error (MSE) and peak signal to noise ratio (PSNR). The lowest MSE value and the highest PSNR value represent the best quality of underwater images. The results of the experiment showed that the Canny edge detection technique outperformed other approaches used in the course of the project.

Detection of Multiple Drones in a Time-Varying Scenario Using Acoustic Signals

Detection of unauthorized drones is mandatory for defense organizations and also for human life protection. Currently, detection methods based on thermal, video, radio frequency (RF) and acoustic signals exist. In previous research, we presented an acoustic signals-based multiple drones detection technique utilizing independent component analysis (ICA) in the presence of interfering sources. In this paper, a method is proposed in which the mixed signals are first separated taking the ICA technique into account. After extracting the features, the support vector machines (SVM) and the k-nearest neighbors (KNN) are used to identify multiple drones in the field. This technique can detect multiple drones in static and quasi-static mixing scenarios, while failing in time-varying scenarios. In this paper, a time-varying drone detection technique (TVDDT) is proposed that first stores a data set of the mixed signals in a time-varying scenario, where time variations occur within the processing data blocks. After estimating the mixing matrices, we developed a technique to track variations in the channel. This technique is based on variations in the mixing coefficients. The proposed channel tracking technique performs classification and detection based on minimum variation criteria in the channel. The proposed TVDDT technique is evaluated through simulations and its superior performance is observed.

A scoping review of autoantibodies as biomarkers for canine autoimmune disease.

BackgroundAutoantibody biomarkers are valuable tools used to diagnose and manage autoimmune diseases in dogs. However, prior publications have raised concerns over a lack of standardization and sufficient validation for the use of biomarkers in veterinary medicine.ObjectivesSystematically compile primary research on autoantibody biomarkers for autoimmune disease in dogs, summarize their methodological features, and evaluate their quality; synthesize data supporting their use into a resource for veterinarians and researchers.AnimalsNot used.MethodsFive indices were searched to identify studies for evaluation: PubMed, CAB s, Web of Science, Agricola, and SCOPUS. Two independent reviewers (AET and ELC) screened titles and abstracts for exclusion criteria followed by full‐text review of remaining articles. Relevant studies were classified based on study objectives (biomarker, epitope, technique). Data on study characteristics and outcomes were synthesized in independent data tables for each classification.ResultsNinety‐two studies qualified for final analysis (n = 49 biomarker, n = 9 epitope, and n = 34 technique studies). A high degree of heterogeneity in study characteristics and outcomes reporting was observed. Opportunities to strengthen future studies could include: (1) routine use of negative controls, (2) power analyses to inform sample sizes, (3) statistical analyses when appropriate, and (4) multiple detection techniques to confirm results.ConclusionsThese findings provide a resource that will allow veterinary clinicians to efficiently evaluate the evidence supporting the use of autoantibody biomarkers, along with the varied methodological approaches used in their development.

Recognition Model of Counterfeiting Digital Records of Biometric Photographic Image

Biometric portrait as one of the most important means of identifying requirements through strict definition of dimensional relationships, preservation of realistic information about all technical characteristics of the photographic image, so that all biometric values can be digitized and used in recognition. The great variety and accessibility of applications for digital processing of digital record of a photographic image has enabled a visually convincing display of a forged photograph that leaves a different impression on the viewer and transmits a different, that is, a forged message. Due to the need to prove the authenticity of the digital record of the photographic image, methods have been developed for the analysis of the record that can detect deviations from the real record even when there are no visual signs of processing the photographic image. Not all analysis techniques can detect certain methods of photo manipulation, so multiple digital photography detection and analysis techniques need to be applied. In order to prove its authenticity, the scientific paper deals with methods for analysis and detection of forgery of digital photography with respect to the digital record and the structure of JPEG format.

Multiple Diseases and Pests Detection Based on Federated Learning and Improved Faster R-CNN

Traditional disease and pest detection technology employ cloud based deep learning, which facing the pressures such as high data storage and communication costs, unbalanced and insufficient data from orchards, diversity of pests and diseases, and complex detection environments. In this paper, we propose a multiple pest detection technique based on Federated Learning (FL) and improved Faster Region Convolutional Neural Network (R-CNN). As the new distributed computing model, FL can derive a shared model integrating the advantages of data from all parties without uploading local data, and also reduces the communication cost. A restriction M is added to the FL algorithm to ensure the convergence of the model and improve the training speed. According to the original Faster R-CNN network, ResNet-101 is used instead of VGG-16 to construct the base convolutional layer to maintain the original structure of small-sized targets and improve the detection speed. Then, the multi-size fusion of feature maps from different convolutional layers is performed to improve the detection accuracy of multi-size multiple pests and diseases. Finally, a Soft-NMS algorithm is proposed to solve the apple obscured problem after the RPN network. The experimental results show that the improved Faster R-CNN can achieve an average accuracy of 90.27% on multiple pest detection, and the detection time is only 0.05 seconds per image. After using FL, the mAP of the model reached 89.34% and the model training speed was improved by 59%.