Search

Paper artifacts are contaminated by external factors in the process of preservation such as foxing. For the problem of backward technology of rapid detection of foxing on paper artifacts, a method based on hyperspectral imaging technology is proposed to detect foxing spots on paper artifacts. After selecting the region of interest and obtaining the corresponding average reflectance, the difference in the average reflectance is found after comparing the healthy regions with the diseased regions. Using band operation and minimum noise fraction to observe the characteristics of foxing image, although there is overlap in different parts, the distribution distinction between moldy and healthy regions is obvious; K-nearest neighbor method and BP neural network are applied to establish the spectral discrimination model of paper artifacts with foxing spots, and the overall discrimination rate of the two methods is 73.3% and 85%, respectively. The results show that hyperspectral imaging can be used for the identification of foxing spots, but the distinction between different parts is not good, and the discrimination effect still needs to be improved.

Light via our eyes influences visual performance, visual comfort and visual experience, but also affects several health related, non-image-forming (NIF) responses. New metrics have been developed to quantify the NIF effects of light. In order to incorporate these in lighting design practice, simulation tools are required that are able to process information about the spectral distribution of light sources and materials. However, most of the tools currently used for daylight and electric light simulations simplify the spectrum into RGB (Red, Green, Blue) colour values. This paper presents an overview of the currently used programs for simulating the NIF effects of light in building design and discusses the possibility of using existing spectral rendering software as an alternative. A review of literature shows that mostly Radiance or Radiance-based programs have been used so far, but new user-friendly tools could employ existing spectral rendering tools. As the NIF effects of light gain greater importance in lighting design, new simulation workflows are needed. This paper aims to support the development of future workflows by presenting the current state-of-the-art.

This paper deals with the new segmentation techniques for retinal blood vessels on fundus images. This technique aims at extracting thin vessels to reduce the intensity difference between thick and thin vessels. This paper proposes the modified UNet model by incorporating ResNet blocks into it which includes structured prediction. In this work we generate the visualization of blood vessels from retinal fundus image for two loss functions namely cross entropy loss and Dice loss where the network classifies several pixels simultaneously. The results shows higher accuracy by considering a much more expressive UNet algorithm and outperforms the past algorithms for Retinal Vessel Segmentation. The benefits of this approach will be demonstrated empirically.

Fire detection technology based on video images is an emerging technology that has its own unique advantages in many aspects. With the rapid development of deep learning technology, Convolutional Neural Networks based on deep learning theory show unique advantages in many image recognition fields. This paper uses Convolutional Neural Networks to try to identify fire in video surveillance images. This paper introduces the main processing flow of Convolutional Neural Networks when completing image recognition tasks, and elaborates the basic principles and ideas of each stage of image recognition in detail. The Pytorch deep learning framework is used to build a Convolutional Neural Network for training, verification and testing for fire recognition. In view of the lack of a standard and authoritative fire recognition training set, we have conducted experiments on fires with various interference sources under various environmental conditions using a variety of fuels in the laboratory, and recorded videos. Finally, the Convolutional Neural Network was trained, verified and tested by using experimental videos, fire videos on the Internet as well as other interference source videos that may be misjudged as fires.

Social Media is an arena in recent times for people to share their perspectives on a variety of topics. Most of the social interactions are through the Social Media. Though all the Online Social Networks allow users to express their views and opinions in many forms like audio, video, text etc, the most popular form of expression is text, Emoticons and Emojis. The work presented in this paper aims at detecting the sentiments expressed in the Social Media posts. The Machine Learning Models namely Bernoulli Bayes, Multinomial Bayes, Regression and SVM were implemented. All these models were trained and tested with Twitter Data sets. Users on Twitter express their opinions in the form of tweets with limited characters. Tweets also contain Emoticons and Emojis therefore Twitter data sets are best suited for the sentiment analysis. The effect of emoticons present in the tweet is also analyzed. The models are first trained only with the text and then they are trained with text and emoticon in the tweet. The performance of all the four models in both cases are tested and the results are presented in the paper.

Facial recognition is one of the main research directions in the field of artificial intelligence and image processing. It has been widely used in identity authentication, video surveillance and biological detection. Because it is non-contact, natural, convenient and reliable, facial recognition has become a popular choice for biometric systems. The accuracy of facial recognition still needs to be improved, the main goal of this paper is to improve the accuracy of face recognition. Based on the support vector machine method, the focus is on the feature extraction and feature matching of face images. In view of the particularity of face images, the pre-processing of face images is studied. In this paper, grayscale normalization and geometric normalization pre-processing methods are used. In order to reduce the interference factors of the image as much as possible, the features are high-lighted, and the non-featured parts are weakened, this paper adopts the Histogram of Oriented Gradient feature extraction method. Then we proposed a new method based on SVM, which uses a one-to-many method to construct multiple SVM classifiers, selects the optimal parameters through repeated experiments, and selects ORL face database for testing. The recognition rate can reach about 98.5%.

As a value-added service that improves the efficiency of online customer service, customer service robots have been well received by sellers in recent years. Because the robot strives to free the customer service staff from the heavy consulting services in the past, thereby reducing the seller’s operating costs and improving the quality of online services. The purpose of this article is to study the intelligent customer service robot scene understanding technology based on deep learning. It mainly introduces some commonly used models and training methods of deep learning and the application fields of deep learning. Analyzed the problems of the traditional Encoder-Decoder framework, and introduced the chat model designed in this paper based on these problems, that is, the intelligent chat robot model (T-DLLModel) obtained by combining the neural network topic model and the deep learning language model. Conduct an independent question understanding experiment based on question retelling and a question understanding experiment combined with contextual information on the dialogue between online shopping customer service and customers. The experimental results show that when the similarity threshold is 0.4, the method achieves better results, and an F value of 0.5 is achieved. The semantic similarity calculation method proposed in this paper is better than the traditional method based on keywords and semantic information, especially when the similarity threshold increases, the recall rate of this paper is significantly better than the traditional method. The method in this article has a slightly better answer sorting effect on the real customer service dialogue data than the method based on LDA.

Infrared thermography technology is widely used in the thermal condition detection of insulators due to its advantages of non-contact, sensitive, online detection. To realize the automatic detection of the operating condition of insulators in complex environments, this paper proposes a method for the recognition and location of the insulator based on Region-based Fully Convolutional Networks (R-FCN). The model was trained and tested on the constructed insulator infrared data set, compared with the SSD model. The results showed that the R-FCN detecting insulators can not only accurately locate insulators, but have an AP (average precision) value as high as 89.2%. Therefore, the findings in this paper have verified that R-FCN has great advantages in the recognition and location of infrared images of insulators and has practical application value.

This paper takes the physical model of Radar Cat’s Eye as the research object, based on geomagicdesign software’s powerful reverse modeling function, the reverse modeling and design process of Radar Cat’s Eye is explained in detail. The detailed steps and main precautions of scanner calibration in the process of Radar Cat’s Eye reverse modeling are given. At the same time, the main contents of model scanning, data processing, reverse modeling, model comparative analysis and so on are also introduced and explained in detail. In short, the relevant contents of this paper provide reference and help for the reverse modeling and design of other models.

The development of music is a tortuous process, and the network relationship between each genre and each artist is intricate. In order to have a better understanding of the history of music, this paper tells the stories hidden in the history of music by means of data processing. Firstly, this paper establishes a model to evaluate the similarity between music by using ISOMAP algorithm. At the same time, the forest evolution model was established to mark the most revolutionary musical characters. Finally, using the Page-Rank algorithm, we get the founders of several music genres. It turns out that the figures who led the development of music don’t coincide with the figures who revolutionized music. Through the analysis of this paper, we can more clearly understand the development of music and the evolution of genres.

A signature is a mark or name that represents the identity of the people and the Signature Verification System (SVS) is used to validate the identity of people. The signature verification system is mostly used for bank cheques, vouchers, intelligence agencies and others. There are two types of SVS which are online and offline signature verification systems. The paper deals with an offline signature verification system. The proposed system consists of four main stages, (i) image acquisition, (ii) image pre-processing, (iii) feature extraction and (iv) classification. The image pre-processing steps involved binarization, noise removal using Gaussian filter and image resizing and thinning. In the feature extraction stage, Bag-of-Features with the Speeded Up Robust Features (SURF) extractor was utilized. In the third stage, the Support Vector Machine (SVM) classifier is used. Lastly, the confusion matrix and the verification rate were used to evaluate the performance of the classifier. In this paper, we implement and compare the performance of the signature verification system without entering the user ID and the signature verification system entering the user ID. For the ratio of 75% and 25% of the training and testing, respectively, the average accuracy for the signature verification system without entering the user ID is 71.36%, whereas the average accuracy for the signature verification system entering the user ID is 79.55%.

The traditional hepaticcell carcinoma (HCC) pathological grading depends on biopsy, which will cause damage to the patient's body and is not suitable for everyone's pathological grading diagnosis. The purpose of this paper is to study the pathological grading of liver tumors on MRI images by using deep learning algorithm, so as to further improve the accuracy of HCC pathological grading. An improved network model based on SE-DenseNet is proposed. The nonlinear mapping relationship between feature channels is modeled and recalibrated using attention mechanism, and rich deep-seated features are extracted, so as to improve the feature expression ability of the network. The method proposed in this paper is verified on the data set including 197 patients, including 130 training sets and 67 test sets. The experimental results are evaluated by receiver operating characteristic (ROC) and area under the ROC curve (AUC). The improved SE-Densenet network achieves good results, and AUC 0.802 is obtained on the test set. The experimental results show that the method proposed in this paper can well predict the pathological grade of HCC.

with the rise and rapid development of mobile communication, intelligent terminal and data system, we are entering the era of mobile Internet. In recent years, more and more data need to be processed and transmitted in daily life, and structured data is becoming more and more important. Among them, multi-mode matching technology can search data in a wider range. Matching for multiple patterns at a time avoids unnecessary matching, accelerates the matching process, and helps to find longer matching information and obtain higher accuracy. This paper mainly introduces the high camouflage intrusion detection method of structured database based on multi-mode matching. This paper uses the high disguised intrusion detection method of structured database based on multi-mode matching, collects sensitive information of wireless access points and stations through the communication of WLAN in multimodal matching, then intercepts and forges data packets to initiate replay attack. Replay attack is characterized by abnormal traffic in the network, which can be detected by statistical analysis. The experimental results show that the high camouflage intrusion detection method based on multi-mode matching makes the camouflage intrusion detection rate increase by 23%. The limitations of the design and research of camouflage intrusion detection are analyzed, discussed and summarized, so as to enrich the academic research results.

With the advent of sensor technology, the exertion of multispectral image (MSI) is comely omnipresent. Denoising is an essential quest in multispectral image processing which further improves recital of unmixing, classification and supplementary ensuing praxis. Explication and ocular analysis are essential to extricate data from remote sensing images for broad realm of supplications. This paper describes curvelet transform based denoising of multispectral remote sensing images. The implementation of curvelet transform is done by using both wrapping function and unequally spaced fast Fourier transform (USFFT) and they diverge in selection of spatial grid which is used to construe curvelets at every orientation and scale. The coefficients of curvelets are docket by a scaling factor, angle and spatial location criterion. This paper crisps on denoising of Linear Imaging Self Scanning Sensor (LISS) III images. The proposed denoising approach has also been collated with some existing schemes for assessment. The efficacy of proposed approach is analyzed with calculation of facet matrices such as Peak signal to noise ratio and Structural similarity at distinct variance of noise..

With the rapid development of today’s technological society, recognition algorithms have received more and more attention. In addition, in recent years, deep learning algorithms have developed rapidly at the theoretical level, and related new technologies have also been applied to various industries. TensorFlow is a deep learning framework that performs well in all aspects. The purpose of this article is to study the realization of recognition algorithms based on TensorFlow’s deep learning mechanism and their optimization techniques. The target detection algorithm used in the system in this paper combines deep learning technology to replace the traditional method based on convolutional filtering. The paper is based on the TensorFlow deep learning framework. TensorFlow is an open source software library for machine intelligence. The learning software library of the network learning framework. This article uses a semi-automatic labeling method combined with an incremental learning algorithm to label the data set. After labeling the data, the parameters are set, the model is trained, and the model is finally trained and applied to the detection system. Studies have shown that: in the recognition algorithm, only the single sub-analysis stream is considered, and the short video sequence analysis stream can get the most excellent accuracy. Compared with the second best long video sequence analysis stream, it can also increase by about 3%.

The power consumption readings of sub meter and total meter of distribution transformer of low-voltage users follow the law of conservation of energy. The meter power loss rate of abnormal low-voltage users must also be abnormal. This paper studies the solution of the meter power loss rate under the four abnormal power consumption scenarios of single (multi) user and full (partial) period. The traditional linear solution method has accurate identification effect for the abnormal power consumption scenario of full period, but it cannot identify the abnormal power consumption scenario of partial period. In this paper, an improved artificial fish swarm algorithm is proposed. By adjusting the fixed step to the adaptive step, the power loss rate of each sub meter is obtained, and the abnormal power users are pinpointed. The research results are verified by simulation examples on IEEE European Low Voltage Test Feeder. The results show that the improved artificial fish swarm algorithm in this paper can identify abnormal power users for the above four abnormal electric field scenarios. The algorithm provides a new alternative for the identification of abnormal low voltage users.

Deep learning requires training on massive data to get the ability to deal with unfamiliar data in the future, but it is not as easy to get a good model from training on massive data. Because of the requirements of deep learning tasks, a deep learning framework has also emerged. This article mainly studies the efficient distributed image recognition algorithm of the deep learning framework TensorFlow. This paper studies the deep learning framework TensorFlow itself and the related theoretical knowledge of its parallel execution, which lays a theoretical foundation for the design and implementation of the TensorFlow distributed parallel optimization algorithm. This paper designs and implements a more efficient TensorFlow distributed parallel algorithm, and designs and implements different optimization algorithms from TensorFlow data parallelism and model parallelism. Through multiple sets of comparative experiments, this paper verifies the effectiveness of the two optimization algorithms implemented in this paper for improving the speed of TensorFlow distributed parallel iteration. The results of research experiments show that the 12 sets of experiments finally achieved a stable model accuracy rate, and the accuracy rate of each set of experiments is above 97%. It can be seen that the distributed algorithm of using a suitable deep learning framework TensorFlow can be implemented in the goal of effectively reducing model training time without reducing the accuracy of the final model.

With the improvement of people’s awareness of environmental protection in recent years, the related problems of water pollution treatment have gradually come into people’s view. As the source of life, water accounts for a huge proportion in our lives, but at the same time, water pollution is quietly spreading in places we don’t know. The continuous discharge of heavy industrial wastewater, agricultural wastewater and domestic sewage leads to increasingly serious water pollution. Sewage treatment(ST) is imperative, and its social benefits are huge, but the corresponding cost is high, and the return on investment is low. Traditional ST methods can not load large-scale ST. How to carry out ST based on artificial intelligence(AI), build ST plant control system, and make ST enter the era of automation is the problem to be solved. The purpose of this paper is to put forward the reform of control system for ST plant based on AI, apply AI into ST, and realize the automation and precision of ST plant. This paper mainly uses the fuzzy self-tuning PID control system algorithm, through the analysis of ST control object, analysis of fuzzy self-tuning PID controller design to complete the ST control system settings. In this paper, the literature review method and data analysis method are used. By collecting relevant data, the control system of ST plant is constructed to simulate ST, and the real-time data of ST is analyzed. The traditional PID control and fuzzy self-tuning PID control are compared. The experimental results show that the wastewater treatment plant system based on AI input, in the aspect of wastewater treatment, the concentration of COD and BOD in the treated wastewater are reduced by a certain proportion, the dissolved oxygen content in the wastewater reaches about 2.0mg/l, which meets the national discharge standard, and its rising time is reduced to 25 seconds, and the adjustment time is saved by 50 seconds.

The primary objective of this particular paper is to classify the health-related data without feature extraction in Machine Learning, which hinder the performance and reliability. The assumption of our work will be like, can we able to get better result for health-related data with the help of Tree based Machine Learning algorithms without extracting features like in Deep Learning. This study performs better classification with Tree based Machine Learning approach for the health-related medical data. After doing pre-processing, without feature extraction, i.e., from raw data signal with the help of Machine Learning algorithms we are able to get better results. The presented paper which has better result even when compared to some of the advanced Deep Learning architecture models. The results demonstrate that overall classification accuracy of Random Forest, XGBoost, LightGBM and CatBoost, Tree-based Machine Learning algorithms for normal and abnormal condition of the datasets was found to be 97.88%, 98.23%, 98.03% and 95.57% respectively.

Conveyor belt transfer is a widely used transportation means in industry and agriculture, with the help of the robot arms the workpiece on the belt can be picked and placed, replacing human sorters for production lines work. The position and orientation of the workpiece are important for grabbing by the robot arms. The goal of the paper was to investigate the acquisition of the position and orientation of the conveyor belt workpiece by means of the camera video overhead looking down the belt. The proposed method is the inter frame difference in nature, using the conveyor belt background as the first frame, but the other frames were not used wholly as usually, only an ROI all around the conveyor belt in the camera video was chosen, and the inter frame difference was carried out in the ROI. The ROI was of the same width as that of the belt in the video which was known in advance, while the length of the ROI was arbitrary, so one pixel in the frame was scaled to the actual length conveniently. Every read frame behind the background was computed the difference with the background in such ROI, and the four vertexes coordinates of the rectangle workpiece image on the belt were obtained when it passed the ROI, and then the distance apart from the right belt boundary was calculated due to the proportional relation between the width of workpiece and that of the ROI. Two kind workpiece orientation on the belt toward the left and right were judged using the same obtained four vertexes coordinates by means of Euclidian length, and the tilt angle was calculated by arc tangent function in favour of two narrow sides of rectangle workpiece grab. The actual test showed that the method of obtaining the position and orientation of workpiece on the belt proposed in the paper could be realized correctly.

Aiming at the problem that the current feature selection algorithm can not adapt to both supervised learning data and unsupervised learning data, and had poor feature interpretability, this paper proposed a hybrid feature selection model based on machine learning and knowledge graph. By the idea of hybridization, this model used supervised learning algorithms, unsupervised learning algorithms and knowledge graph technology to model from the perspective of data features and text features. Firstly, the data-based feature weights were obtained through the machine learning model, and then the text-based weights were obtained by using the knowledge graph technology, and the weight sets are combined to obtain a feature matrix with good explanatory properties that meets both the data and text features. Finally, the case analysis proves that the method proposed in this paper has good effects and interpretability.

In recent years, with the rapid development of artificial intelligence technology, people’s demand for wireless spectrum resources is increasing, which poses a huge challenge to the originally tight and limited wireless spectrum resources. On the other hand, the traditional fixed spectrum cooperative sensing and allocation algorithms result in extremely low spectrum utilization for a considerable part of the licensed spectrum. The purpose of this paper is to study the cooperative sensing and allocation algorithm of cognitive RS (radio spectrum) based on artificial intelligence. This dissertation focuses on cooperative perception and cognitive radio systems, respectively, from the aspects of cooperative perception of user fairness, maximization of system energy efficiency, and user detection when user access is busy. Firstly, a joint optimization model of fairness cooperative spectrum sensing and allocation is established to compensate the sensing overhead of cooperative users to ensure its fairness; then, define and analyze the energy efficiency of the cognitive system, and establish a joint optimization model of cooperative spectrum sensing and allocation based on artificial intelligence to maximize energy efficiency, and optimize wireless sensing and allocation parameters while ensuring maximum system energy efficiency. The experimental results show that when = 0.7, the algorithm proposed in this study has reached 100% of the RS perception performance, while the traditional algorithm only has 93%. The algorithm proposed in this paper has greater advantages in perception and distribution performance.

As one of the core algorithms of machine vision, the mobile image multi-label recognition algorithm has received extensive attention from researchers in recent years and has been widely used in cutting-edge fields such as deep learning framework paddlepaddle platform, video surveillance, intelligent robots, and unmanned aerial vehicles. However, the existing recognition algorithms are not completely satisfied with the practical application in life and production. Due to the complexity of the platform environment, they can often only propose specific solutions based on existing problems, and there is no universal algorithm that is suitable for all kinds of Complex environment. The purpose of this paper is to study the multi-label recognition algorithm of moving images based on PaddlePaddle platform. This research mainly analyzes and researches the mobile image multi-tag space deployment plan and the multi-tag recognition algorithm, and further improves the tag reading rate and recognition reliability of the mobile image on the PaddlePaddle platform. This research first analyzes several key factors that affect the performance of UHF recognition system, considers the improvement plan of PaddlePaddle platform’s mobile image multi-tag recognition algorithm from the two aspects of space diversity and frequency diversity, and finally determines the multiple The label space diversity scheme, and the introduction of a multi-label optimization recognition algorithm to improve the recognition efficiency of the PaddlePaddle platform’s mobile image multi-label. Experimental data shows that the reading rate can reach 0.907 when identifying 300 tags in the experiment, and when the number of tags is greater than 300, the reading rate is close to 1, which verifies that the algorithm proposed in this paper is used in the multi-tag recognition of moving images on the PaddlePaddle platform.

This paper discusses the research work done for controlling the humanoid robot manually using deep learning. For teaching, personal assistance, search and rescue humanoid robot are used. Controlling manually makes it to do any task without any explicitly programming. Existing technique for manually controlling the humanoid are heavily dependent on hardware and they are not cost efficient. This paper proposes a novel method for controlling the humanoid using a 2D camera. The image from the 2D camera is processed and skeleton of the human body is captured using deep learning. Then the skeleton is used to control the actuators present in the humanoid robot using image classifier and ROS. As a proof of concept the upper body of the humanoid robot is controlled in real time using this method.

Osteogenesis Imperfecta (OI) is a bone disorder that causes bone to be brittle and easy to fracture. The patient suffered from this disease will have poor quality of life. Simulation on the bone fracture risk would help medical doctors to make decision in their diagnosis. Detection of edges from the OI images is very important as it helps radiologist to segmentize cortical and cancellous bone to make a good 3D bone model for analysis. The purpose of this paper is to review the fundamentals of fuzzy logic in edge detection of OI bone as it is yet to be implemented. Several fuzzy logic concepts are reviewed by previous studies which include fuzziness, membership functions and fuzzy sets regarding digital images. The OI images were produced by modalities such as Magnetic Resonance Imaging (MRI), Ultrasound, or Computed Tomography (CT). In summary, researchers from the reviewed papers concluded that fuzzy logic can be implemented to detect edges in noisy clinical images.

A formal-semantic normative system (FSNS) is presented in this paper. The FSNS allows us to describe the links and functional units of typical information processes. The possibility of using the FSNS to automate graphical-analytical modelling and machine learning is shown. Properties of the FSNS are described. On its basis, a method is developed for specifying a formal alphabet of system connection nodes that has a specific subject content. The rules for manipulating the received alphabetic characters are also presented. The possibility and expediency of building a normative system with increased expressive capabilities is substantiated. This paper provides a brief description of the «Unit-Function-Object» system-object approach. Considered the possibility of using DL for the FSNS in an unambiguous, formalized form. An example of using the FSNS is presented.

In This Paper Face Recognition has gain the more attention in the society of network multimedia for accessing information. Emerging Areas like network security, content indexing and extraction, and video compression benefits from face recognition technic because “people” are the center of attention in a lot of video footages. Face detection is advanced research area in the computer science vision. The research under this area is expanding in others areas like psychology. Localization of human face will be first and the initial stage as a study of face detection. In this paper the usage of Convolutional Neural Network (CNN) algorithm in detecting the criminal face and also the process of getting the location from image.

Motion blur distortion is the most common type of image distortion in daily life. the research on motion-blurred image restoration technology has developed more mature. Classical algorithms such as Wiener filter and Kalman filter and various improved algorithms can achieve better results, but they take a long time and have great limitations in actual image restoration application scenarios. To solve this problem, this paper proposes an algorithm for fast restoration of image motion blur, an improved algorithm based on Randon transform to judge the image motion blur angle, and studies the algorithm for estimating the blur length under the condition that the blur angle is determined, and obtains two motion blur parameters of blurred image. Experimental results show that the fast image restoration method proposed in this paper has shorter time consumption, stronger anti-noise interference ability and better practicability.

The rapid development of deep learning has accelerated the progress of related technologies in the computer vision field and it has broad application prospects. Due to flower inter-class similarity and intra-class differences, flower image classification has essential research value. To achieve flower image classification, this paper proposes a deep learning method using the current powerful object detection algorithm YOLOv5 to achieve fine-grained image classification of flowers. Overlap and occluded objects often appear in the images of the flowers, so the DIoU_NMS algorithm is used to select the target box to enhance the detection of the blocked objects. The experimental dataset comes from the Kaggle platform, and experimental results show that the proposed model in this paper can effectively identify five types of flowers contained in the dataset, Precision reaching 0.942, Recall reaching 0.933, and mAP reaching 0.959. Compared with YOLOv3 and Faster-RCNN, this model has high recognition accuracy, real-time performance, and good robustness. The mAP of this model is 0.051 higher than the mAP of YOLOv3 and 0.102 higher than the mAP of Raster-RCNN.

In order to put an end to the product quality accidents caused by cutting tool breakage or severe wear in machining process, the paper explored an ELM model detection method based on voice recognition. In this paper, firstly it analysed the features of cutting sound signal in time-frequency domain, then discussed the extraction method of tool working state sensitive spectral energy statistical feature based on wavelet packet decomposition, and finally constructed an ELM fast detection model based on sound feature recognition. The experimental results demonstrated that the ELM detection model could achieve higher detection accuracy and faster response time. The simulation results show that the ELM model is effective and applicable in detecting tool wear with the help of sound recognition.

New technologies such as artificial intelligence have developed at a rapid pace in recent years and are increasingly being used in the process of managing news in bulk. The development of deep learning has facilitated unprecedented progress in the field of computing and has opened our eyes to the possibility of using AI for news text classification. In this paper, based on the system requirements analysis, we describe the process of functional modules arising from the requirements analysis, design the internal details of functional modules, including algorithms and detailed principles, and finally obtain a prototype of news text classification software, which results in the pre-design expectations. The research in this paper makes the system development work more concrete, while providing software users, software developers, and analysts and testers with a unified and comprehensive understanding of the system’s functional implementation.

Traditional action recognition algorithms often only pay attention to video RGB features or optical flow features. These methods do not make good use of the audio information in the video. Based on RGB and optical flow characteristics, this paper introduces the processing of audio information, and classifies videos based on element-level fine-grained multi-modal fusion. Through experimental comparison, the accuracy of the multi-modal fusion algorithm proposed in this paper is improved by 7.38% on the HMDB51 dataset and 3.18% on the UCF101 dataset compared to the simple modal splicing. At the same time, it is proved that the introduction of audio modes can effectively improve the performance of the model.

Semantic segmentation of parking lot scenes is the prerequisite of environment perception for automatic parking technology. It provides environmental semantic information for automatic parking of vehicles. However, due to the dim light of the parking lot environment, unclear ground signs, road reflections and other factors, semantic segmentation such as FCN The method is not yet able to segment ground signs such as background and lane lines to meet the perception needs of automatic parking. This paper proposes a parking lot scene semantic segmentation method based on a surround view system. The surround view system consists of 4 fisheye cameras. The images acquired by each camera are subjected to distortion correction, inverse perspective transformation and image stitching fusion to obtain a ring view. Based on the ring view, a semantic segmentation algorithm based on attention and feature fusion is proposed. Experiments are carried out on a self-made parking lot dataset with a size of 1280x960 pixels. The results show that the method proposed in this paper improves the mIoU of the FCN model by 12.3%.

Traffic data encryption has been a trend in most Internet applications, and traditional protocol filtering based on fixed port and traffic classification based on Deep Packet Inspection are unable to Identify encrypted traffic. Recently, the traffic classification method based on the statistical characteristics of network traffic, which can solve the problem of encrypting data or user privacy protection, has been widely discussed. However, the traditional supervised learning method requires manual marking of a large amount of network traffic data, which is tedious and time-consuming. In this paper, a improved semi-supervised traffic classification framework based on BIRCH clustering method is proposed, and through experiments, the proposed algorithm, supervised learning algorithm and classical semi-supervised traffic classification algorithm are analyzed and compared. The results show that the algorithm proposed in this paper has higher overall accuracy and classification accuracy, and the algorithm can increase the accuracy on traffic classification.

In order to improve the accuracy of the super-resolution network and reduce the number of model parameters, this paper improves its RCAB module on the basis of RCAN, and builds a reconstruction network RSAN that can improve the quality and efficiency of image super-resolution reconstruction. By replacing the original channel attention module with a more efficient and lightweight shuffle attention, it is mainly used to reduce the number of parameters, supplemented by improving the accuracy; and replacing part of the ordinary convolution in RCAN with split convolution is mainly used to improve accuracy, supplemented by reducing feature redundancy and parameters. The experimental results show that RSAN in this paper can not only obtain better subjective visual evaluation and objective quantitative evaluation, but also reduce the number of network model parameters and improve the efficiency of the network to a certain extent.

Aiming at the congestion control problem of SDN, this paper designs a network congestion control scheme based on Q-learning algorithm of reinforcement Learning. According to the usage of link bandwidth in the network, the algorithm adjusts the transmission rate of the source data stream continuously, and finally obtains the optimal data stream allocation scheme. Experimental simulation evaluates the performance of the algorithm, the proposed control algorithm based on reinforcement learning congestion, can effectively improve the link utilization, reduce network congestion. The experimental results show that the congestion control algorithm proposed in this paper can allocate the optimal rate for each data stream, reduce network congestion, and has better performance than the traditional SDN algorithm.

Handwritten digit recognition refers to the recognition of handwritten digits written and printed on paper, and the recognition results are stored in the computing machine in the form of text. According to the characteristics of a handwritten digital image, based on the multi-layer perceptron method, this paper introduces the source and principle of handwritten digital image recognition in detail, and analyzes the steps, related functions and theories of handwritten digital image recognition in sequence. The proposed algorithm was validated on the handwritten dataset MNIST, and the model was evaluated and adjusted. Eventually, the trained model was migrated to Android.

In the massive mechanical fault data, the value density of fault information is low, and the data quality is uneven. What’s more, the multi-source signals collected by different sampling methods are different. At present, the expert system or shallow neural network model with weak self-learning ability cannot meet the requirements. Therefore, aiming at the characteristics of coupling, uncertainty and concurrency of mechanical faults, this paper constructs two kinds of 2D-CNN fault feature data sets, and uses Convolutional Neural Network (CNN) with strong self-learning ability to build two kinds of fault diagnosis models: CNN-Z and CNN-F. With CNN-Z and CNN-F models as base learners, this paper utilizes the ensemble algorithm Gradient Boosting Decision Tree (GBDT) to combine multiple bases. Compared with the results of the single base learner, the outcomes have higher accuracy and lower generalization error. Through the analysis and comparison of the performance indicators of the algorithm, this paper concludes that the diagnosis error of the fault diagnosis algorithm based on CNN and GBDT is the lowest with 1.79%, and the effectiveness, reliability and accuracy of the proposed algorithm in mining hidden fault state information are verified.

Nowadays, investigators of phrases and text sentimental orientation identification mainly divided into two ways: semantic comprehension and machine learning. Machine learning needs a lot of training corpus and also cannot handle general field words effectively, meanwhile, semantic comprehension cannot get high scores at precision and recall. Therefore, in this paper, we propose a new fusion method based on our semantic comprehension for judging phrases’ polarity. In this paper, firstly, we modify the knowledge of Hownet, on the basis of its four primitives, the fifth primitive—sentimental primitive was proposed by our research, and annotated it into Hownet manually; secondly, we propose a new phrase sentimental similarity calculation method to compute word’s sentimental value; at last, we integrate transductive learning into this method to identify phrases’ sentimental orientation. The performance of experiment show that, compared with SVM or traditional semantic comprehension, it can get better results.

Aiming at covering the shortage of single source sensor imaging and improving the contrast between the target and the background in image, this paper proposes an adaptive weight fusion algorithm of infrared and visible image based on a High-frequency Domain Convolutional Neural Network (HDCNN). Firstly, the high and low frequency components of the original image are obtained by using the Daubechies wavelet transform, and then a high-frequency domain convolutional neural network which can detect the frequency information ratio of infrared and visible light in the high-frequency subband is trained. Secondly, the network is used to perform adaptive weight fusion for the high frequency components and regional energy is used for fusion of the low frequency components. Finally, the fusion image is obtained by inverse wavelet transform. A large number of experiments have proved that the algorithm in this paper has a greater improvement over similar comparison algorithms in objective evaluation metrics such as standard deviation, spatial frequency and average gradient. The algorithm enhances the contrast between the target and the background in the fusion image, and enriches the characteristic information of the target itself.

In the fingertip intelligent blood collection robot, we need to move the blood collection needle to the designated position quickly and accurately, but it is difficult to meet the two indicators of speed and accuracy at the same time. This paper proposes a fast navigation method based on binocular vision. This method obtains the coordinates of the fingertip blood sampling point through three-dimensional reconstruction and contour extraction, and guides the robotic arm to quickly move the blood sampling needle to a position near the blood sampling point according to the coordinate position. The method in this paper is used for the rough navigation of blood collection robots, and it is verified by experiments that this method can meet the needs of blood collection robots.

According to the relevant provisions of the "Election Law of the People's Republic of China", the engineering software required for the implementation of large-scale conference elections can set voting regulations for "differential elections", "equal elections", and "alternative elections". This paper proposes an automatic ballot recognition system based on image processing, which is suitable for all levels of people’s congresses, party and government meetings or other meetings. It has complete functions and stable recognition. It can automatically count and summarize the voting results and display the voting results on the big screen. Show on the display system. Firstly, the paper ballot is scanned by a scanner to obtain the image of the ballot. Secondly, the system will preprocess the image of the ballot and segment the image into regions.Thirdly, according to the handwritten character recognition algorithm based on image processing and Shen filter proposed in this paper, the handwritten characters "√" and "×" in the subregion are recognized.The final experimental results show that the recognition accuracy of the system is as high as 99.82%, which has an absolute advantage over the traditional manual voting, and can meet the performance requirements of high efficiency, accuracy, objectivity and security of large-scale conference election.

This paper constructs a deep learning method for detecting network traffic anomalies to enhance the secure transmission of data in networks due to the complex, diverse and numerous types of anomalous traffic in current networks. The method combines multiple convolutional auto-encoders (Multi-CAE) with a long short-term memory network. The convolutional auto-encoders are obtained by combining stacked auto-encoders with convolutional layers, which can not only reduce feature loss but also effectively extract the spatial structure of samples. The use of Multi-CAE greatly improves the feature extraction capability, and combined with the long short-term memory network to extract temporal features, the effective features extracted in this paper are more comprehensive and less losses compared to the models used in other researches. A comparison of the loss values in the training of CAE (Convolutional Auto-Encoders) and SAE (Stacked Auto-Encoders) in the experiments shows that the loss values of CAE are about one-tenth lower than those of SAE, and the method consisting of Multi-CAE and LSTM for the USTC- TFC2016 dataset was trained with accuracy values up to 99.98%, and the precision, recall and f1-score parameters were also above 99%, outperforming other studies.

Applying deep learning methods to achieve synthetic aperture radar (SAR) image ship target detection has become a current hot research topic. However, most of the existing detection networks are designed based on optical images, so when they are directly applied to SAR image target detection, there will be the following two problems: 1) There are big differences between SAR image and optical image in imaging mechanism, geometric characteristics and radiation characteristics, etc, which makes the detection network have more redundant imformation; 2) Real-time performance and accuracy are both very important in SAR image target detection task, and applying a network designed based on optical image cannot balance the performance of the two well. In response to the above problems, this paper proposes a lightweight network LSSNet specially designed for SAR image ship detection task. In earlier layers, we use depthwise separable convolution instead of conventional convolution to design dense block; in deeper layers, stacked dense blocks with shortcuts are used to enhance deep supervision, and finally achieve high-speed and high-accuracy detection. This paper uses SSDD dataset as baseline for expeiment and results show that LSSNet have higher accurracy and detecition speed. Its lightweight structure will also help transplant to hardware devices in the future.

The traditional swarm intelligence optimization K-means algorithm has some problems, such as poor global search ability and blind area selection of initial center points, which leads to the reduction of clustering availability. In order to avoid the above limitations, this paper proposed IABC k-means algorithm. Firstly, the employed bees stage in the traditional artificial bee colony ABC algorithm uses the current colony optimal solution information to guide its optimization search. Secondly, this paper wanted to solve the search ability of the employed bees and expand the information sharing range between various individuals, a random guidance mechanism is proposed in the onlookers bees stage. Finally, chaotic sequence is introduced in the scout bee stage to accelerate the convergence speed of the algorithm. IABC algorithm is proposed and applied to K-means clustering algorithm to improve the poor global search ability of K-means algorithm and the random selection of initial center points. Experiments show that the IABC and IABCK-means proposed in this paper effectively improves the clustering availability.

For the purpose of faster and more accurate anomalous traffic detection with increasing classes of data traffic in the network, this paper proposes a new anomalous traffic detection method based on stacked auto-encoders and a long short-term memory network model. The method uses Multi-SAE to extract the effective features of sequential traffic, which is obtained by concatenating multiple stacked auto-encoders, and a long short-term memory network to extract the temporal structure of the effective features, with the Multi-SAE and the long short-term memory network in a back-and-forth tandem structure. To further improve the efficiency of the detection, redundant MAC addresses are also removed in the pre-processing. From the experimental results, this paper achieves effective detection of twenty types of data traffic with an accuracy rate of 98.25%, which is higher than that of the same category of research by nearly 2 percentage points, and the parameters of precision, recall and f1-score also reach over 96%, improving the detection results.

In this paper, based on the research focus of the current logistics field, a deep learning based express single-side bar code extraction algorithm is designed. This algorithm can recognize the original image and read out the information of the package. The algorithm is divided into two parts: bar code localization algorithm and bar code recognition algorithm. To avoid high delay, we use the YOLOV5S network to complete the barcode localization algorithm. First, the YOLOV5S algorithm is used to frame out the barcode, and the bounding Box is cut off. Then, OpenCV's method is used to obtain the deflection Angle of the image. Finally, affine transformation is used to correct the images with deflection. The bar code recognition algorithm is to use ZBAR algorithm to decode the bar code and output the decoded content in the form of string, and finally realize the location and recognition of the bar code on the single side of the express. Finally, through the test of the data set, it can be seen from the experimental results that this paper has accomplished the goal well.

Recover the high-resolution image from a single low-resolution image is known as single image super-resolution reconstruction (SISR). SSIR is currently a research hotspot in computer vision. In this paper, we study the feature learning module of the magnification-arbitrary network for Super-Resolution(Meta-SR), and propose an improved method of image feature learning. Its idea is multi-scale residual dense block (MRDB). We integrate it into the existing structure of the Meta-SR network. As a result, the same network has increased the receptive field, and it can obtain more detailed features of the image. This structure can be extended to other existing SISR network models. In the published and standard test set Set14, compared with the interpolation method, the average PSNR value of the algorithm in this paper increases by 3.79dB, 3.01dB and 2.85dB at amplification factors 2, 3 and 4, respectively. Through the analysis of SISR results, it can be obtained that the improved method proposed in this paper has a significant improvement in all indicators, and can more fully extract the image detail features, thereby improving the quality of the reconstructed image.

Communication of the human brain with the surroundings became reality by using Brain- Computer Interface (BCI) based mechanism. Electroencephalography (EEG) being the non-invasive method has become popular for interaction with the brain. Traditionally, the devices were used for clinical applications to detect various brain diseases but with the advancement in technologies, companies like Emotiv, NeuoSky are coming up with low cost, easily portable EEG based consumer graded devices that can be used in various application domains like gaming, education etc as these devices are comfortable to wear also. This paper reviews the fields where the EEG has shown its impact and the way it has proved useful for individuals with severe motor disorder, rehabilitation and has become a means of communication to the real world. This paper investigates the use of Cubic SVM algorithm In the EEG classification. EEG feature extraction is Implemented by maximum overlap discrete wavelet transform (MODWT) to reduce the dimensionality of data. The Sliding Window Technique is used to calculate the mean within each window samples. The feature vectors are loaded into the support vector machine (SVM) and optimize tree.