Underground coal mine wireless sensor network is facing the problem of limited energy. It is urgent to study efficient data compression and collection methods to solve this problem. Distributed compressed sensing is an efficient data compression and collection method. An underground coal mine image compression and collection method based on distributed compressed sensing is proposed. The image is divided into multiple image blocks by column, and each image block is compressed and encoded separately. Then they are jointly recovered in the server by utilizing intra signal correlation and inter signal correlation. The experimental results show that, under the same compression rate, the method in this paper can obtain better image quality than compressed sensing, and it takes less time to restore the image and has stronger real time performance.

Aiming at solving the problem of wearing helmet detection of electric vehicle drivers and passengers, a wearing detection method of electric vehicle helmet based on improved YOLOv5s model is proposed. Based on YOLOv5s model, firstly, the network structure replaces convolution module in YOLOv5s with GhostBottleneck module to reduce the amount of parameters. Secondly, the GhostCSPBottleneck structure is designed to optimize feature extraction. Finally, attention mechanism is added in the backbone network to improve the detection accuracy. Experimental results show that the mean average precision of the improved model is 84.2%, which is 1.3 percent points higher than that of YOLOv5s. The parameters and weight of the model are compressed to 51.39% and 47.95%, which makes the model show good generalization in the scene of small targets and dense targets. The model is transplanted to the development board NVIDIA Jetson Xavier NX, and can achieve a detection speed of 28.2 FPS, which is able to meet the detection requirement of real time and accuracy.

The quality of intelligent assembly of complex products is an important problem in their assembly. It is necessary to study the designed digital twin assembly model of complex products for humancomputer interaction to improve the assembly quality of products. On the premise of ensuring the safety of complex product assembly models and effective controlling of data risks, the designed digital twin system assembles various digital assembly data based on physical product, builts a digital twin assembly model, and generates multiview dynamic of complex products based on digital twin AR technology to display product assembly. The assembly process quality optimization method based on Copula function is proposed to optimize the assembly quality of complex products. The test results show that the designed model can complete the AR multiview dynamic generation of complex product assembly, and the scoring results are above 0.90, which has a good humancomputer interaction effect. It can realize realtime management and control of ship product assembly, and the stability value of quality control is lower than 0.45.

The recommendation system is able to help users actively find personalized items that meet their preferences and recommend them to users. The matrix factorization method is a classic method in the recommendation system. In order to solve the problem of poor performance of matrix factorization due to sparse data in matrix factorization of recommendation systems, a method of improving matrix factorization is proposed. Firstly, the matrix factorization is improved, and then the potential additional information of the item itself is merged on the basis of the improved matrix factorization. Finally, the similarity between items is introduced for calculation, thereby improving the accuracy of the recommendation algorithm. The experimental results show that the improved method is significantly better than the existing matrix factorization recommendation model, and has significant application value.

In order to improve the image classification performance of convolutional neural network, through the data experiment on image data set CIFAR 10, the effects of optimization factors such as data set, number of hidden neurons and training method on the classification accuracy, running time and network parameters of convolutional neural network are verified. Thus, it provides a reference for effectively improving the image classification performance of convolutional neural network.

Aiming at the problems that the existing methods have weak visualization effect and difficult interaction in wind wave dynamic simulation, this paper presents a dynamic simulation method coupling wind wave model with WebGL. The threedimensional scene of wind wave is drawn based on WebGL technology. The dynamic simulation of wind wave in small wetlands is realized in different conditions with Three.js mainstream graphics library and particle system. The experimental results show that the dynamic simulation method is feasible and robust, provides users with more immersive interactive experience and realistic simulation scenes, and also provides new decision support for wetland management.

The engineering application of C band microstrip antenna array in radar is studied theoretically and experimentally. First, basic manufacturing factors of microstrip antenna array are designed according to the classical theory. Then, the microstrip antenna array is simulated and optimized through full wave simulation software to determine the parameters of patch unit and feed network. Finally, the microstrip antenna array is fabricated to verify the effectiveness of the design method. The experimental results show that the measured results of the microstrip antenna array are basically agreement with the simulation results, meet the engineering requirements, and obtain an antenna gain of up to 27.24 dB.

Training excellent talents in network engineering is the urgent need for China to implement the “Internet plus” strategy and to realize the “Networking” of important industries such as manufacturing in 14th Five Year.This paper expounds the development process of network engineering specialty combining with the school running orientation and school running characteristics of Anhui Jianzhu University. This paper constructs the excellent talent training mode of network engineering specialty based on the concept of engineering education from the aspects of following the development of the industry, formulating talent training scheme, reconstructing professional curriculum system, putting forward student centered teaching method, carrying out in depth integration of schools and enterprises, and improving students  innovative consciousness and ability.Finally, it briefly describes the effectiveness of talent training, and looks forward to the future talent training of network engineering specialty.

In order to quantitatively study the accuracy of measuring length, angle, straightness and parallelism of mechanical parts with monocular machine vision, a reliability modeling and analysis method for accuracy of machine vision has been proposed. The model comprehensively considers the influences of external illumination, random noise, the angle between the object to be measured and the observation station, camera distortion and edge detection algorithm on the measurements accuracy of monocular machine vision. Monte Carlo method and common edge detection algorithm are used to simulate the monocular machine vision measurement of the length, angle, straightness and parallelism of mechanical parts under different working conditions. Comparing the experimental results with the simulation results, the length error, angle error, straightness error and parallelism error differ by 0.22 μm, 0.007 3°, 0.002 2 mm and 0.031 9 mm, respectively, which verify the effectiveness of the proposed method.

Aiming at the problem of electromagnetic interference in the electromagnetic driver for optical fiber alignment, based on analysing coupling mechanism of anomalous photovoltaic effect, pyroelectric effect, thermal expansion effect and piezoelectric effect of lead lanthanum zirconate titanate(PLZT) ceramics, the photoinduced deformation formula of PLZT ceramic multi energy field coupling in the illumination stage is deduced, and the deflection output mathematical model of PLZT ceramic laminated cantilever optical control driver under light conditions is constructed. The driving displacement of the optical control driver is simulated and analyzed. The results show that the driving displacement and response speed of the optical control driver increase with the increase of light intensity, and gradually reach a saturation value. When the light intensity is 200 mW·cm-2, the maximum output displacement of the optical control driver for optical fiber alignment can reach 226.7 μm.

Aiming at the problems of low position control accuracy and high control cost of ordinary cable drive parallel robot(CDPR), a CDPR with force position hybrid control system is designed based on the built in sensor of the motor. The model of planar four cable parallel robot is constructed, the kinematics inverse equation and static equilibrium equation are obtained, and the force position hybrid control system is designed. A precise rope retracting device is designed, with Arduino as the control board and bus servo as the driving device, to realize the robot position and trajectory control. Simulation and prototype experiments show that the system has high accuracy and feasibility, and it is of certain significance for realizing highprecision positioning of CDPR.

The old houses built in rural areas are mainly masonry structures without ring beams and structural columns, and improving their seismic performance has become an urgent engineering problem to be solved. The reinforcement technology of externally attached prefabricated ring beams and structural columns is proposed, so as to improve the seismic performance. The numerical simulation and experimental research on the damage characteristics of the reinforced wall are carried out.The results show that the seismic performance of masonry structures can be significantly improved by the reinforcement measures of ring beams and structural columns.The hysteretic deformation amplitudes of prefabricated and castinplace reinforcement methods are different. The deformation amplitudes of walls reinforced with prefabricated ring beams and structural columns are relatively small, and the deformation hysteresis effect is more obvious. The seismic effect of the reinforcement measures of prefabricated ring beams and structural columns exceed the reinforcement methods of cast in place.

At present, blast furnace slag(BFS) mixed concrete is widely used in infrastructure. This paper studies the relationship between the pore structure and chloride penetration resistance of BFS mixed concrete. The test results show that:1)BFS can optimize the pore structure of BFS mixed concrete by reducing the connectivity of pores and the proportion of harmful pores, and then affect the chloride penetration resistance of BFS mixed concrete. 2)Water cement ratio (W/C) and curing age can affect the chloride penetration resistance of BFS mixed concrete by affecting the most probable pore diameter and the proportion of harmful pores. 3)The rapid chloride ion migration coefficient of BFS mixed concrete has a high correlation with the proportion of harmful pores with diameter greater than 50 nm, but has a weak correlation with the capillary porosity and the most probable pore diameter.

Using the method of bioinformatics, the core promoter and transcription binding site are predicted in the candidate promoter region with a length of 2 000 bp upstream of the 5′ end of the pig hypoxia inducible factor HIF 1α published in NCBI. The physicochemical properties, subcellular localization, transmembrane domain, signal peptide, advanced structure and evolutionary relationship of the protein encoded by HIF 1α are analyzed. The results show that there is a potential promoter region at the upstream of 5′ end of pigs HIF1α. There are CpG islands and multiple transcription factor binding sites. The protein encoded by HIF 1α is a hydrophilic protein with an isoelectric point of 5.11. It is composed of 824 amino acids, and the highest content is leucine. HIF 1α protein is mainly located in the nucleus, and there is no transmembrane domain, so it is a non transmembrane protein. There is no signal peptide in the HIF 1α protein, and its secondary structure is mainly α spiral and random curl. In addition, through the comparison of HIF 1α sequence homology, it is found that sheep has the closest homology with pig s HIF 1α gene and the original chicken has the furthest homology. In conclusion, the study provides a theoretical basis for the exploration of structure and function of pig s HIF 1α　gene.