Search

Since 2007, methods that utilize side-channel data to detect hardware Trojan (HT) problems have been widely studied. Machine learning methods are widely used for hardware Trojan detection, but with the development of integrated circuits (ICs), better results are usually obtained using deep learning methods. In this paper, we propose an architecture inspired by Residual-Block and Dense-Block and combine it with SE Attention Mechanism, which we named the Res-Dense-SE-Net network. By combining residual connectivity, dense connectivity, and attention mechanism, the Res-Dense-SE-Net network can enjoy the advantages of these three network architectures at the same time, which can improve the expressiveness and performance of the model. The Res-Dense-SE-Net network can capture the key features in the image better, and it can solve the problems of gradient vanishing and feature transfer efficiently, which can in turn improve the classification accuracy and the generalization ability of the model. Based on the publicly available AES series of hardware Trojans from TrustHub and the publicly available hardware Trojan-side channel data by Faezi et al., we evaluate the effectiveness of the method proposed in this paper. The experimental results show that when a single Trojan exists, the method proposed in this paper has a high accuracy rate; and when multiple types of hardware Trojans exist at the same time and need to be categorized, the categories of hardware Trojans can also be effectively identified, and the categorization accuracy is high compared with the existing deep learning methods. [ABSTRACT FROM AUTHOR]

In this paper, we propose a novel approach to solving theorem proving problems without relying on any deduction method. We transform logical formulas into numbers, vectors and matrices, and feed the corresponding data into various machine learning algorithms to predict their satisfiability. Here, we introduce ProverX, a novel theorem prover that utilizes various binary classification algorithms, ranging from traditional machine learning to deep learning, to tackle the satisfiability (SAT) problem of propositional logic. Empirical experiments were conducted to evaluate the performance of ProverX using datasets we generated. ProverX achieved accuracy rates ranging from 81.8% to 98.7%, demonstrating a remarkable speedup of almost 180 times compared to CTL-RP, a highly efficient prover. These results demonstrate the feasibility of replacing deduction with learning in theorem proving, opening promising avenues for further exploration in more complex logics, e.g., Modal Logic, Coalition Logic, Propositional Linear-Time Temporal Logic, Computation Tree Logic, etc., provided that their resolution methods exist. We also present a detailed analysis of the best-performing machine learning approaches for this task and introduce a new algorithm, IPDA, which has the potential to further enhance performance. [ABSTRACT FROM AUTHOR]

In numerous scientific disciplines and practical applications, addressing optimization challenges is a common imperative. Nature-inspired optimization algorithms represent a highly valuable and pragmatic approach to tackling these complexities. This paper introduces Dendritic Growth Optimization (DGO), a novel algorithm inspired by natural branching patterns. DGO offers a novel solution for intricate optimization problems and demonstrates its efficiency in exploring diverse solution spaces. The algorithm has been extensively tested with a suite of machine learning algorithms, deep learning algorithms, and metaheuristic algorithms, and the results, both before and after optimization, unequivocally support the proposed algorithm's feasibility, effectiveness, and generalizability. Through empirical validation using established datasets like diabetes and breast cancer, the algorithm consistently enhances model performance across various domains. Beyond its working and experimental analysis, DGO's wide-ranging applications in machine learning, logistics, and engineering for solving real-world problems have been highlighted. The study also considers the challenges and practical implications of implementing DGO in multiple scenarios. As optimization remains crucial in research and industry, DGO emerges as a promising avenue for innovation and problem solving. [ABSTRACT FROM AUTHOR]

As an important part of the bridge structure system, the underwater pile-pier structure usually occurs various defects on its surfaces due to its complex hydrological environment. The existing conventional defect detection approaches exist two aspects of problems: (1) insufficient definition and color distortion of the underwater images, and (2) low efficiency and error-prone. To solve these problems, this paper proposed the target defect detection model by integrating the image-fusion enhancement algorithm and the deep learning algorithm. Firstly, by analyzing the reasons for the degradation of the underwater images, the ACE (automatic color equalization) and CLAHE (contrast limited adaptive histogram equalization) algorithms are selected to enhance the image, respectively. Secondly, the two enhanced images are fused based on the point sharpness weight, and then the fusion results are further sharpened by the USM (unsharp mask) algorithm, thus obtaining the final fused images. Thirdly, 3,200 fused images are taken as the training set, by adopting the YOLOv3 algorithm to train the detection model, and then the training model is validated and tested by the other each 400 fused images, thus building up the target automatic detection model of underwater pile-pier surface defects. Finally, a series of comparison and discussion were conducted to validate the effectiveness of image-fusion and the robustness and effectiveness of the target detection model. The results found that the target detection model has excellent robustness against noise and effectiveness in the surface defect detection. This indicates that the image-fusion approach proposed in this paper can effectively enhance the image features, and the target detection model is feasible, robust, and effective in the automatic detection of surface defects on underwater pile-pier structures. [ABSTRACT FROM AUTHOR]

Fast and accurate smoke detection is very important for reducing fire damage. Due to the complexity and changeable nature of smoke scenes, existing smoke detection technology has the problems of a low detection rate and a high false negative rate, and the robustness and generalization ability of the algorithms are not high. Therefore, this paper proposes a smoke detection model based on the improved YOLOv5. First, a large number of real smoke and synthetic smoke images were collected to form a dataset. Different loss functions (GIoU, DIoU, CIoU) were used on three different models of YOLOv5 (YOLOv5s, YOLOv5m, YOLOv5l), and YOLOv5m was used as the baseline model. Then, because of the problem of small numbers of smoke training samples, the mosaic enhancement method was used to randomly crop, scale and arrange nine images to form new images. To solve the problem of inaccurate anchor box prior information in YOLOv5, a dynamic anchor box mechanism is proposed. An anchor box was generated for the training dataset through the k-means++ clustering algorithm. The dynamic anchor box module was added to the model, and the size and position of the anchor box were dynamically updated in the network training process. Aiming at the problem of unbalanced feature maps in different scales of YOLOv5, an attention mechanism is proposed to improve the network detection performance by adding channel attention and spatial attention to the original network structure. Compared with the traditional deep learning algorithm, the detection performance of the improved algorithm in this paper was is 4.4% higher than the mAP of the baseline model, and the detection speed reached 85 FPS, which is obviously better and can meet engineering application requirements. [ABSTRACT FROM AUTHOR]

Semantic segmentation of remote sensing images plays a crucial role in urban planning and development. How to perform automatic, fast, and effective semantic segmentation of considerable size and high-resolution remote sensing images has become the key to research. However, the existing segmentation methods based on deep learning are complex and often difficult to apply practically due to the high computational cost of the excessive parameters. In this paper, we propose an end-to-end lightweight progressive attention semantic segmentation network (LPASS-Net), which aims to solve the problem of reducing computational costs without losing accuracy. Firstly, its backbone features are based on a lightweight network, MobileNetv3, and a feature fusion network composed of a reverse progressive attentional feature fusion network work. Additionally, a lightweight non-local convolutional attention network (LNCA-Net) is proposed to effectively integrate global information of attention mechanisms in the spatial dimension. Secondly, an edge padding cut prediction (EPCP) method is proposed to solve the problem of splicing traces in the prediction results. Finally, evaluated on the public datasets BDCI 2017 and ISPRS Potsdam, the mIoU reaches 83.17% and 88.86%, respectively, with an inference time of 0.0271 s. [ABSTRACT FROM AUTHOR]

Cultural and creative design is a new design mode based on the Internet platform, which gathers the design wisdom of the public and provides online solutions for design tasks. Evaluation indicators are used to score cultural and creative design schemes that require manual construction of an evaluation matrix, which makes the evaluation efficiency very low and increases the product design cycle. Therefore, based on the deep belief network of cultural creation, the method of constructing the evaluation model of the scheme is designed in this paper. According to the characteristics of the deep belief network, the k-medoids clustering method is adopted, and the clustering key value k is set. Meanwhile, the clustering result is calculated, and k central point elements are obtained. Moreover, project the central point elements back to the cultural and creative design plan to obtain the characteristic plan, and use manual evaluation to calculate the comprehensive score. Besides, the feature scheme data and the data simulation formula are used to construct the training set of the deep belief network, which are used to train the evaluation model of the deep belief network to solve the problem of the length of time that requires a large amount of manual evaluation to construct the output of the training set. In addition, according to the number of evaluation indicators, the number of nodes in the input layer of the deep belief network is determined to define the initialization method of network weights and biases, formulating experimental procedures to determine the optimal network structure. In the simulation experiment, through the test analysis of the comparison model, the function synchronization rate in this paper is 4.2% better than that of the comparison method. After multiple iterations, the test effect is 7 times higher than that of the comparison method, realizing the optimal evaluation of the design schemes that have not been manually evaluated in the same design scheme set. [ABSTRACT FROM AUTHOR]

In recent years, deep learning methods have been widely used for road extraction in remote sensing images. However, the existing deep learning semantic segmentation networks generally show poor continuity in road segmentation due to the high-class similarity between roads and buildings surrounding roads in remote sensing images, and the existence of shadows and occlusion. To deal with this problem, this paper proposes strip attention networks (SANet) for extracting roads in remote sensing images. Firstly, a strip attention module (SAM) is designed to extract the contextual information and spatial position information of the roads. Secondly, a channel attention fusion module (CAF) is designed to fuse low-level features and high-level features. The network is trained and tested using the CITY-OSM dataset, DeepGlobe road extraction dataset, and CHN6-CUG dataset. The test results indicate that SANet exhibits excellent road segmentation performance and can better solve the problem of poor road segmentation continuity compared with other networks. [ABSTRACT FROM AUTHOR]

To solve the problem of low accuracy in the traditional FunkSVD recommendation algorithm, an improved FunkSVD algorithm (RM-FS) is proposed. RM-FS is an improvement of the traditional FunkSVD algorithm, using RMSProp, a deep learning optimization algorithm. The RM-FS algorithm can not only solve the problem of reduced accuracy of the traditional FunkSVD algorithm because of iterative oscillations but also alleviate the impact of data sparseness on the accuracy of the algorithm, achieving the effect of improving the accuracy of the traditional algorithm. The experimental results show that the RM-FS algorithm proposed in this paper effectively improves the accuracy of the recommendation algorithm, which is better than the traditional FunkSVD recommendation algorithm and other improved FunkSVD algorithms. [ABSTRACT FROM AUTHOR]

The development of deep learning provides a new way for solving the colorization problem on the grayscale image. Excellent coding‐based methods appear in the automatic image colorization task, avoiding the unsaturated colour effect problem of previous methods based on the L2 loss function. Traditional neural networks come with high computational costs and a large number of parameters. Considering the limitation of memory and computing resources and aiming at lightweight, a novel grey image automatic colorization network is proposed. The basic idea of coding‐based methods is used, regarding the colorization task as a pixel‐level classification problem, meanwhile redesign and improve the colour encoding and decoding process. This network architecture leverages a lightweight convolution to reduce the computation and combines an efficient attention model to form a residual block as the kernel of the backbone network. Furthermore, an efficient image self‐attention mechanism placed at the end of the network is applied to enhance the ultimate colouring results. The method proposed in this paper can maintain the natural colouring effect and significantly reduce the computational amount and network model parameters. [ABSTRACT FROM AUTHOR]

Pixel-level image fusion is an effective way to fully exploit the rich texture information of visible images and the salient target characteristics of infrared images. With the development of deep learning technology in recent years, the image fusion algorithm based on this method has also achieved great success. However, owing to the lack of sufficient and reliable paired data and a nonexistent ideal fusion result as supervision, it is difficult to design a precise network training mode. Moreover, the manual fusion strategy has difficulty ensuring the full use of information, which easily causes redundancy and omittance. To solve the above problems, this paper proposes a multi-stage visible and infrared image fusion network based on an attention mechanism (MSFAM). Our method stabilizes the training process through multi-stage training and enhances features by the learning attention fusion block. To improve the network effect, we further design a Semantic Constraint module and Push–Pull loss function for the fusion task. Compared with several recently used methods, the qualitative comparison intuitively shows more beautiful and natural fusion results by our model with a stronger applicability. For quantitative experiments, MSFAM achieves the best results in three of the six frequently used metrics in fusion tasks, while other methods only obtain good scores on a single metric or a few metrics. Besides, a commonly used high-level semantic task, i.e., object detection, is used to prove its greater benefits for downstream tasks compared with single-light images and fusion results by existing methods. All these experiments prove the superiority and effectiveness of our algorithm. [ABSTRACT FROM AUTHOR]

Finding a good compiler autotuning methodology, particularly for selecting the right set of optimisations and finding the best ordering of these optimisations for a given code fragment has been a long-standing problem. As the rapid development of machine learning techniques, tackling the problem of compiler autotuning using machine learning or deep learning has become increasingly common in recent years. There have been many deep learning models proposed to solve problems, such as predicting the optimal heterogeneous mapping or thread-coarsening factor; however, very few have revisited the problem of optimisation phase tuning. In this paper, we intend to revisit and tackle the problem using deep learning techniques. Unfortunately, the problem is too complex to be addressed in its full scope. We present a new problem, called reduced O3 subsequence labelling – a reduced O3 subsequence is defined as a subsequence of O3 optimisation passes that contains no useless passes, which is a simplified problem and expected to be a stepping stone towards solving the optimisation phase tuning problem. We formulated the problem and attempted to solve the problem by using genetic algorithms. We believe that with mature deep learning techniques, a machine learning model that predicts the reduced O3 subsequences or even the best O3 subsequence for a given code fragment could be developed and used to prune the search space of the problem of the optimisation phase tuning, thereby shortening the tuning process and also providing more effective tuning results. [ABSTRACT FROM AUTHOR]

The representation of consistent mixed reality (XR) environments requires adequate real and virtual illumination composition in real-time. Estimating the lighting of a real scenario is still a challenge. Due to the ill-posed nature of the problem, classical inverse-rendering techniques tackle the problem for simple lighting setups. However, those assumptions do not satisfy the current state-of-art in computer graphics and XR applications. While many recent works solve the problem using machine learning techniques to estimate the environment light and scene's materials, most of them are limited to geometry or previous knowledge. This paper presents a CNN-based model to estimate complex lighting for mixed reality environments with no previous information about the scene. We model the environment illumination using a set of spherical harmonics (SH) environment lighting, capable of efficiently represent area lighting. We propose a new CNN architecture that inputs an RGB image and recognizes, in real-time, the environment lighting. Unlike previous CNN-based lighting estimation methods, we propose using a highly optimized deep neural network architecture, with a reduced number of parameters, that can learn high complex lighting scenarios from real-world high-dynamic-range (HDR) environment images. We show in the experiments that the CNN architecture can predict the environment lighting with an average mean squared error (MSE) of 7.85 × 10−4 when comparing SH lighting coefficients. We validate our model in a variety of mixed reality scenarios. Furthermore, we present qualitative results comparing relights of real-world scenes. [Display omitted] • Automatic end-to-end method to estimate the environment lighting in XR applications. • A CNN architecture that learns a latent-space of the environment lighting. • A methodology to generate egocentric mixed-reality-views from HDR panoramas. • Real-time lighting estimation that does not make assumptions about the XR scene. [ABSTRACT FROM AUTHOR]

Specific emitter identification (SEI) can distinguish single-radio transmitters using the subtle features of the received waveform. Therefore, it is used extensively in both military and civilian fields. However, the traditional identification method requires extensive prior knowledge and is time-consuming. Furthermore, it imposes various effects associated with identifying the communication radiation source signal in complex environments. To solve the problem of the weak robustness of the hand-crafted feature method, many scholars at home and abroad have used deep learning for image identification in the field of radiation source identification. However, the classification method based on a real-numbered neural network cannot extract In-phase/Quadrature (I/Q)-related information from electromagnetic signals. To address these shortcomings, this paper proposes a new SEI framework for deep learning structures. In the proposed framework, a complex-valued residual network structure is first used to mine the relevant information between the in-phase and orthogonal components of the radio frequency baseband signal. Then, a one-dimensional convolution layer is used to a) directly extract the features of a specific one-dimensional time-domain signal sequence, b) use the attention mechanism unit to identify the extracted features, and c) weight them according to their importance. Experiments show that the proposed framework having complex-valued residual networks with attention mechanism has the advantages of high accuracy and superior performance in identifying communication radiation source signals. [ABSTRACT FROM AUTHOR]

Recently, deep learning has become a promising technique for constructing gesture recognition classifiers from surface electromyography (sEMG) signals in human–computer interaction. In this paper, we propose a gesture recognition method with sEMG signals based on a deep multi-parallel convolutional neural network (CNN), which solves the problem that traditional machine learning methods may lose too much useful information during feature extraction. CNNs provide an efficient way to constrain the complexity of feedforward neural networks by weight sharing and restriction to local connections. Sophisticated feature extraction is to be avoided and hand gestures are to be classified directly. A multi-parallel and multi-convolution layer convolution structure is proposed to classify hand gestures. Experiment results show that in comparison with five traditional machine learning methods, the proposed method could achieve higher accuracy. [ABSTRACT FROM AUTHOR]

The reinforcement learning algorithms based on policy gradient may fall into local optimal due to gradient disappearance during the update process, which in turn affects the exploration ability of the reinforcement learning agent. In order to solve the above problem, in this paper, the cross-entropy method (CEM) in evolution policy, maximum mean difference (MMD), and twin delayed deep deterministic policy gradient algorithm (TD3) are combined to propose a diversity evolutionary policy deep reinforcement learning (DEPRL) algorithm. By using the maximum mean discrepancy as a measure of the distance between different policies, some of the policies in the population maximize the distance between them and the previous generation of policies while maximizing the cumulative return during the gradient update. Furthermore, combining the cumulative returns and the distance between policies as the fitness of the population encourages more diversity in the offspring policies, which in turn can reduce the risk of falling into local optimal due to the disappearance of the gradient. The results in the MuJoCo test environment show that DEPRL has achieved excellent performance on continuous control tasks; especially in the Ant-v2 environment, the return of DEPRL ultimately achieved a nearly 20% improvement compared to TD3. [ABSTRACT FROM AUTHOR]

For decades, machine learning techniques have been used to process Amharic texts. The potential application of deep learning on Amharic document classification has not been exploited due to a lack of language resources. In this paper, we present a deep learning model for Amharic news document classification. The proposed model uses fastText to generate text vectors to represent semantic meaning of texts and solve the problem of traditional methods. The text vectors matrix is then fed into the embedding layer of a convolutional neural network (CNN), which automatically extracts features. We conduct experiments on a data set with six news categories, and our approach produced a classification accuracy of 93.79%. We compared our method to well-known machine learning algorithms such as support vector machine (SVM), multilayer perceptron (MLP), decision tree (DT), XGBoost (XGB), and random forest (RF) and achieved good results. [ABSTRACT FROM AUTHOR]

Advances in mathematical models and optimization introduce powerful tools for solving the visual object tracking problem, but this fundamental problem remains unsolved due to various challenges such as illumination variation, deformation and occlusion. Recent progress in correlation filter based models has provided an effective and efficient tracking solution. This type of trackers exploits the property of convolution theorem to significantly speed up their computational processes. However, their scheme of training correlation filters restricts them to use only a few negative examples to train their model, consequently lowers their performance. Therefore, in this paper, we propose to combine the fast calculation advantage of correlation filter with an online discriminative learning model, which can fully exploit the negative examples in the context around the target. The proposed tracker proceeds in a coarse-to-fine scheme: the proposed tracker will first employ a correlation filter to generate a coarse estimation of the location of the target, and then employ a translating model to refine its estimation and calculate the scale variation of the target via a scaling model. Both the translating model and the scaling model are formulated in the framework of our online discriminative model. Besides, we also propose an effective offline representation learning method to generate robust image feature for our correlation filter. Extensive experiments on the online Object Tracking Benchmark against the state-of-the-art methods validate the effectiveness of the proposed tracker. [ABSTRACT FROM AUTHOR]

To solve the problems of poor exploration ability and convergence speed of traditional deep reinforcement learning in the navigation task of the patrol robot under indoor specified routes, an improved deep reinforcement learning algorithm based on Pan/Tilt/Zoom(PTZ) image information was proposed in this paper. The obtained symmetric image information and target position information are taken as the input of the network, the speed of the robot is taken as the output of the next action, and the circular route with boundary is taken as the test. The improved reward and punishment function is designed to improve the convergence speed of the algorithm and optimize the path so that the robot can plan a safer path while avoiding obstacles first. Compared with Deep Q Network(DQN) algorithm, the convergence speed after improvement is shortened by about 40%, and the loss function is more stable. [ABSTRACT FROM AUTHOR]

Deep learning-based target detection techniques have already made a wide-range impact on our daily life. Currently, a feature pyramid is a widely utilized technique for multiscale target detection, the effectiveness of the technique has already been proved. Nevertheless, in the pyramid structure, problems, such as multiscale feature alignment, model turmoil after fusion, feature redundancy, and no-local feature fusion, exist. In this paper, we propose a novel progressive structure network to solve the aforementioned problems. The proposed structure contains three modules: multiscale feature alignment fusion, different scale channels & spatial location adaptive weighted fusion, and multiscale global and local feature fusion. The proposed structure is capable of fusing information from different feature layers more effectively. Subsequently, the semantic gaps among different scales can be reduced. Furthermore, the proposed structure can maintain the stability of the detection network and its performance has been proved by comparing with other state-of-art feature fusion method. The proposed progressive network structure has also been applied to actual target detection tasks and the practical application effectiveness of our method has been verified. [ABSTRACT FROM AUTHOR]

Cyber security is used to protect and safeguard computers and various networks from ill-intended digital threats and attacks. It is getting more difficult in the information age due to the explosion of data and technology. There is a drastic rise in the new types of attacks where the conventional signature-based systems cannot keep up with these attacks. Machine learning seems to be a solution to solve many problems, including problems in cyber security. It is proven to be a very useful tool in the evolution of malware detection systems. However, the security of AI-based malware detection models is fragile. With advancements in machine learning, attackers have found a way to work around such detection systems using an adversarial attack technique. Such attacks are targeted at the data level, at classifier models, and during the testing phase. These attacks tend to cause the classifier to misclassify the given input, which can be very harmful in real-time AI-based malware detection. This paper proposes a framework for generating the adversarial malware images and retraining the classification models to improve malware detection robustness. Different classification models were implemented for malware detection, and attacks were established using adversarial images to analyze the model's behavior. The robustness of the models was improved by means of adversarial training, and better attack resistance is observed. [ABSTRACT FROM AUTHOR]

Precise navigation is essential for autonomous underwater vehicles (AUVs). The measurement deviation of the navigation sensors, especially the microelectromechanical systems (MEMS) sensors, is a crucial factor that affects the localization accuracy. Deep learning is a novel method to solve this problem. However, the calculation cycle and robustness of the deep learning method may be insufficient in practical application. This paper proposes an adaptive navigation algorithm with deep learning to address these questions and realize accurate navigation. Firstly, this algorithm uses deep learning to generate low-frequency position information to correct the error accumulation of the navigation system. Secondly, the χ2 rule is selected to judge if the Doppler velocity log (DVL) measurement fails, which could avoid interference from DVL outliers. Thirdly, the adaptive filter, based on the variational Bayesian (VB) method, is employed to estimate the navigation information simultaneous with the measurement covariance, improving navigation accuracy even more. The experimental results, based on AUV field data, show that the proposed algorithm could realize robust navigation performance and significantly improve position accuracy. [ABSTRACT FROM AUTHOR]

With improvements in human living standard, people's demands on food quality are getting higher and higher. Effective food recognition algorithms are needed to obtain more useful food information. In order to solve the problem of low accuracy and slow speed of food recognition algorithms, a new food recognition algorithm based on CNN algorithm is proposed. First, the proposed algorithm preprocess the food images which are collected from the internet. And then use the traditional convolution extract the features from food images. The jumping convolution which is designed in this paper to extract food features jumping and combines the features from traditional convolutions. This algorithm cannot only solve the food recognition problem effectively, but also reduce the calculation parameters. Compared with the experimental results of other deep learning networks, the proposed algorithm has a good effect, and can recognize the food quickly and reduce the training time. [ABSTRACT FROM AUTHOR]

In this paper, a proposed algorithm that dynamically changes the neural network structure is presented. The structure is changed based on some features in the cascade correlation algorithm. Cascade correlation is an important algorithm that is used to solve the actual problem by artificial neural networks as a new architecture and supervised learning algorithm. This process optimizes the architectures of the network which intends to accelerate the learning process and produce better performance in generalization. Many researchers have to date proposed several growing algorithms to optimize the feedforward neural network architectures. The proposed algorithm has been tested on various medical data sets. The results prove that the proposed algorithm is a better method to evaluate the accuracy and flexibility resulting from it. [ABSTRACT FROM AUTHOR]

To solve the problems of low accuracy, low real-time performance, poor robustness and others caused by the complex environment, this paper proposes a face mask recognition and standard wear detection algorithm based on the improved YOLO-v4. Firstly, an improved CSPDarkNet53 is introduced into the trunk feature extraction network, which reduces the computing cost of the network and improves the learning ability of the model. Secondly, the adaptive image scaling algorithm can reduce computation and redundancy effectively. Thirdly, the improved PANet structure is introduced so that the network has more semantic information in the feature layer. At last, a face mask detection data set is made according to the standard wearing of masks. Based on the object detection algorithm of deep learning, a variety of evaluation indexes are compared to evaluate the effectiveness of the model. The results of the comparations show that the mAP of face mask recognition can reach 98.3% and the frame rate is high at 54.57 FPS, which are more accurate compared with the exiting algorithm. [ABSTRACT FROM AUTHOR]