AUTOMATIC CONTROL AND COMPUTER SCIENCES最新文献

The current road crack detection models have the issues of insufficient detection accuracy and ineffective detection of small cracks. To address these issues, this paper proposes an improved YOLOv5s road crack detection model. Firstly, the coordinate attention module was inserted after all the C3 modules in the backbone to promote the feature extraction capability. Secondly, we used C2f instead of C3 to strengthen feature fusion. Then the context augmentation module CAM was added before the last Concat to enhance the detection effect of small cracks. Finally, we replaced all but the first Conv module with a Ghost Module to minimize the quantity of parameters and calculations. For convenience, we call this improved model as YOLOv5s-CCCG. The experimental results show that compared with YOLOv5s, the improved model has an improvement of 4.7 and 9% in mAP@0.5 and mAP@0.5:0.95, respectively. The detection accuracy is higher than several other object detection algorithms.

Currently, there is a lack of regulatory safety in scenarios with haze or dense water mist. To address this issue, this study analyzes image dehazing technology and proposes an image dehazing model based on multiscale residual and mixed attention mechanism. This model improves image processing efficiency and image dehazing effect by combining multiscale residual networks with spatial attention, channel attention, and frequency attention. The model achieved peak signal-to-noise ratios of 35.76 and 34.39 dB, respectively, and structural similarity values of 0.9891 and 0.9870 in the indoor and outdoor test sets of the RESIDE dataset, which were significantly better than other comparison methods. In the NTIRE’18 test set, the model found the optimal peak signal-to-noise ratio of 12.26 dB in the 45th iteration, and the optimal similarity value of 0.684 in the 60th iteration. The application analysis in real-world task test sets showed that the research model had better visual effects and detail restoration ability. Time complexity analysis showed that the model had a lower runtime, indicating its efficient computational performance. The proposed model exhibits excellent dehazing performance and computational efficiency on multiple standard and real-world test sets, verifying the effectiveness of multiscale residual networks and mixed attention mechanisms in image dehazing tasks.

Automatic testing technology can reduce production costs and improve output rates, gradually replacing time-consuming and laborious manual testing. It is gradually replacing time-consuming and labor-intensive manual testing. However, there are currently significant development difficulties and high maintenance costs in automated testing systems. The testing work is too focused on software maintenance and case correction, resulting in a significant gap between the actual application results and expectations. In view of this, a lighter automated testing system is built based on the Vue.js framework. Considering the strong data dependency in existing detection algorithms, a reinforcement learning testing algorithm based on Sarsa is proposed to enhance the flexibility of testing. The results showed that the automated testing model had higher testing coverage, testing efficiency, and fault detection volume on the software program dataset F-Droid, with 87.5%, 90.1%, and 1003 respectively, all higher than the comparison algorithm. In robot motion control testing, the model had a lower root mean square error of 1.24%. The comparative model couldn’t converge or converge to over 5.0%. This indicates that the automated testing system improves testing efficiency and accuracy, help to reduce testing costs, and ensure system stability and operational quality.

Efficient and flexible autonomous obstacle avoidance motion capabilities have become an urgent and practical requirement for robots in the current production life. Aiming at the problem that the joint vibration caused by excessive impact affects the quality of work when the robotic arm is operating, an optimal spraying trajectory planning method for the robotic arm is proposed. In this paper, we take the four-degree-of-freedom robotic arm as the research object, use five times nonuniform B spline functions to construct the trajectory of the robotic arm, construct the multiobjective optimization function of time and impact, optimize the trajectory based on the multiobjective particle swarm optimization algorithm, and then get the required solution from the Pareto front-end through the normalization of the objective weighting function. Real robots are used for experimental verification, and the improved multiobjective optimization particle swarm algorithm, PAD-MOPSO is used to achieve the effect of multiobjective optimization of time and impact, and the displacement, velocity, acceleration, and torque during the motion process are within the constraint range.

With the continuous increase in people’s demand for electricity, to ensure the stable operation of the power system, accurate analysis of grid signals has become the primary task in improving power quality. This article uses a combination of FFT and wavelet transform to accurately analyze the steady-state and transient components in signals, improving detection accuracy. The specific plan is to use Newton interpolation to quasi synchronize the original sampling sequence, achieving consistency between the sampling period and the actual period; Using the modulemaximum method on the original signal to detect the presence of transient disturbances. The results showed that quasi synchronization can accurately correct the collected signals with high restoration and synchronization; The improved windowed interpolation FFT achieves a detection accuracy of 0.3% for steady-state harmonics; Wavelet divides the frequency band and uses module maxima to detect signal singularity, accurately extracting transient disturbance signals with an accuracy of up to 5%; Furthermore, we can apply this new combination algorithm to more complex power signal processing.

Addressing issues such as high parameter count and low detection accuracy in weed detection algorithms, this paper presents a study on a weed detection algorithm based on an improved YOLOv5s model. Firstly, the backbone network was enhanced using the ParC modul4e to reduce computational demands and increase model detection speed; secondly, the C3BRA module, designed based on the BRA attention mechanism, replaced the original C3 module to focus on the extraction and reinforcement of key feature information; finally, the SIoU loss function replaced the CIoU loss function, accelerating network convergence and improving model detection accuracy. Experimental validation on the test dataset compared the improved model with the original YOLOv5s model, showing that the modified model increased the P value by 2.8%, the mAP value by 1.7%, and reduced model parameters by 10.7%, better meeting the requirements for weed detection.

To tackle the issues of incomplete feature reconstruction and insufficient feature representation in the traditional extreme learning machine autoencoder (ELM-AE), this paper proposes a multilayer GEELM-AE architecture based on cyclic structure (GEELM-AE-MCS). First, we embed the weights into the reconstruction error function to enhance local feature clustering by integrating graph embedding theory. Second, we incorporate the graph embedding matrix into the ELM feature space to preserve both global structural information and similarity of the feature data, thereby enabling the algorithm to establish a more effective boundary for feature discrimination. Finally, we propose GEELM-AE-MCS, which leverages each self-encoder’s dimensionality reduction capability to further enhance algorithm performance. Experimental results demonstrate that GEELM-AE-MCS exhibits superior feature representation and classification capabilities compared to state-of-the-art algorithms.

This study presents a mathematical modeling and numerical performance evaluation of a robust adaptive control strategy for the stabilization and trajectory tracking of a remotely operated underwater vehicle (ROV). Through the precise design and control of ROVs for seabed and dam inspections, these systems can efficiently substitute for human intervention, thereby eliminating the necessity to dewater structures during maintenance operations. The developed adaptive tracking controller leverages radial basis function neural networks (RBF NNs) to estimate the unknown nonlinear dynamics of the system. To further enhance robustness, The controller integrates sophisticated robust control strategies to correct modeling inaccuracies in the neural network and manage bounded external disturbances. The stability and performance of the system are rigorously validated through Lyapunov-based stability analysis. The effectiveness and dependability of the suggested method are demonstrated by means of extensive numerical simulations.

The fingerprint-based biometric technology is vulnerable to several kinds of attacks. One of the simplest attacks to carry out on the fingerprint sensor is the presentation attack. Numerous fingerprint presentation attack detection (FPAD) strategies have been put out in recent years. These FPAD techniques have yielded promising results on cross-material datasets. However, when training and testing datasets come from different domains (sensors), the performance of the FPAD approach can degrade by up to 30%. Therefore, to achieve a consistent performance, a robust FPAD approach must learn domain-independent features. We have developed an unsupervised divergence-based domain adaptation (UDDA) method with an adaptive loss function (ALF) to minimize the domain shift in FPAD. The ALF integrates domain divergence loss (DDL) and classification loss. In a cross-sensor scenario, the ALF helps learn domain-invariant features and provides reliable classification of real and fraudulent fingerprints. Experimental results demonstrate that the proposed UDDA approach reduces the cross-sensor average classification error (ACE) by 19.94% on LivDet 2015 and 19.23% on LivDet 2017.

In order to ensure the security of identity authentication of electric power, financial and other terminal products, this paper proposes a security operation method based on security chip identity authentication. There are four main methods of this method, which need to be completed with corresponding instructions: The first is the authentication identity method. The security chip needs to send the “verify identity authentication pin” command to the interface device. After the verification is passed, the interface device can legally read and write to the security chip. The second method is to unlock the identity, need to execute the “unlock identity authentication pin” command, this method can make the locked security chip restore normal operation. The third method is to reset the identity, which needs to execute the “reload identity authentication pin” command, which can set the user’s identity authentication pin to the newly entered identity authentication pin, so as to ensure the restorability of the authentication. The fourth is to change the identity method, the security chip needs to execute the “change identity authentication pin” command, this method can update the original identity authentication pin to the newly entered identity authentication pin. Based on actual application security requirements, users can use the four types of commands to reasonably combine applications, select different security operation methods, and perform corresponding application functions such as verify, unlock, reload, and change identity authentication pin, so as to achieve secure operation of the security chip authentication pin. Identify whether the identity authentication pin is in a locked state in response to commands such as verify identity authentication pin, unlock identity authentication pin, reload identity authentication pin, or change identity authentication pin. Recognizing that the identity authentication pin is in a locked state, it is prohibited to execute the verify identity authentication pin command and change identity authentication pin command; it is allowed to execute the unlock identity authentication pin command and reload identity authentication pin command to unlock identity authentication pin. Therefore, after the identity authentication pin is locked, the identity authentication pin can be unlocked based on the unlock identity authentication pin command or reload identity authentication pin command, so that the security chip can continue to be used while meeting security requirements.