The fractional PID controller, having been designed, effectively improves upon the outcomes of the standard PID controller.
Convolutional neural networks have garnered extensive use in hyperspectral image classification recently, exhibiting exceptional performance. However, the pre-determined convolution kernel's receptive field frequently results in insufficient feature extraction, and the high redundancy in spectral information complicates the process of extracting spectral features. For these problems, we propose a novel solution: a 2D-3D hybrid convolutional neural network (2-3D-NL CNN) that includes a nonlocal attention mechanism and both an inception block and a nonlocal attention module. The network's multiscale receptive fields, essential for extracting multiscale spatial features of ground objects, are provided by the inception block using convolution kernels of varying sizes. The nonlocal attention module broadens the network's understanding of spatial and spectral contexts, while decreasing spectral redundancy, leading to improved spectral feature extraction. The Pavia University and Salins hyperspectral datasets served as a testing ground for evaluating the efficacy of the inception block and nonlocal attention module in experiments. Our model's classification accuracy, across both datasets, stands at 99.81% and 99.42%, respectively, exceeding the performance of existing models.
From active seismic sources in the external environment, we precisely measure vibrations using fiber Bragg grating (FBG) cantilever beam-based accelerometers, which are designed, optimized, fabricated, and tested. Several advantages are inherent in FBG accelerometers, including their ability for multiplexing, their immunity to electromagnetic disturbances, and their high sensitivity. Calibration, fabrication, and packaging of a simple PLA cantilever beam accelerometer, complemented by FEM simulations, are discussed. A finite element simulation, coupled with laboratory calibrations using a vibration exciter, examines the relationship between cantilever beam parameters and their influence on natural frequency and sensitivity. The optimized system, based on the test results, exhibits a resonance frequency of 75 Hz, functioning within the 5-55 Hz range, while maintaining a high sensitivity of 4337 pm/g. Microbiology inhibitor In the final phase of testing, a field comparison is conducted between the packaged FBG accelerometer and standard 45-Hz vertical electro-mechanical geophones. Seismic sledgehammer shots were acquired consecutively along the test line, and a comparative analysis was carried out on the experimental results from both systems. Suitability of the designed FBG accelerometers for the task of recording seismic traces and identifying the initial arrival times is unequivocally demonstrated. The promising potential of seismic acquisitions is evident in the system optimization and subsequent implementation.
Radar-based human activity recognition (HAR) offers a non-invasive approach for various applications, including human-computer interfaces, intelligent security systems, and sophisticated surveillance, while prioritizing privacy. Inputting radar-preprocessed micro-Doppler signals into a deep learning network represents a promising strategy for classifying human activities. Despite the impressive accuracy achievable with conventional deep learning algorithms, the complexity of their network structures hinders their deployment in real-time embedded applications. In this investigation, a highly efficient network with an attention mechanism is put forward. According to the time-frequency representation of human activity, this network disconnects the Doppler and temporal features of the radar preprocessed signals. The Doppler feature representation is derived sequentially by the one-dimensional convolutional neural network (1D CNN) with the application of a sliding window. HAR is executed through the application of an attention-mechanism-based long short-term memory (LSTM) to the time-ordered Doppler features. The activity's features are effectively strengthened using an average cancellation method, yielding improved clutter reduction within the context of micro-motion. The new system boasts a 37% improvement in recognition accuracy, significantly surpassing the accuracy of the traditional moving target indicator (MTI). The superior expressiveness and computational efficiency of our method, confirmed by two human activity datasets, distinguishes it from traditional methods. A key characteristic of our approach is the achievement of recognition accuracy near 969% on both datasets, combined with a network structure significantly lighter than those of algorithms exhibiting similar recognition accuracy. A substantial potential exists for the application of the method detailed in this article to real-time HAR embedded systems.
In response to the demands of high-performance line-of-sight (LOS) stabilization of the optronic mast, especially in high oceanic conditions and substantial platform swaying, a combined approach employing adaptive radial basis function neural networks (RBFNNs) and sliding mode control (SMC) is formulated. To address the uncertainties within the optronic mast system, an adaptive RBFNN approximates the nonlinear and parameter-varying ideal model, thus reducing the big-amplitude chattering associated with high switching gains in SMC. Online construction and optimization of the adaptive RBFNN relies on the current state error information, thereby avoiding the need for any preliminary training data. In order to alleviate the system's chattering, a saturation function is applied to the time-varying hydrodynamic and friction disturbance torques, rather than the sign function. The asymptotic stability of the proposed control method is explicitly proven using the Lyapunov stability framework. The validity of the proposed control method is ascertained through a comprehensive series of simulations and practical experiments.
For the last of this three-paper set, we employ photonic technologies to monitor the environment. After a review of configurations optimal for high-precision farming, we now analyze the obstacles to accurately measuring soil water content and effectively forecasting landslides. Moving forward, we concentrate our efforts on a next-generation of seismic sensors capable of functioning in both terrestrial and underwater contexts. Lastly, we delve into the application of optical fiber sensors within the context of radiation exposure.
Extensive structures, exhibiting thin walls similar to aircraft skins and ship shells, frequently measure several meters but maintain a thickness of only a few millimeters. By means of the laser ultrasonic Lamb wave detection method (LU-LDM), signals can be identified over extensive distances, excluding the need for physical contact. tendon biology Moreover, this technology exhibits remarkable flexibility in the design of measurement point arrangements. The review begins by examining the laser ultrasound and hardware configurations of LU-LDM, a key aspect of its characteristics. The subsequent categorization of the methods relies on three factors: the amount of wavefield data gathered, the spectral characteristics, and the arrangement of measurement points. The positive and negative aspects of different methodologies are compared, and the optimal scenarios for implementing each are articulated. From the third perspective, we consolidate four methods that guarantee a judicious balance between detection efficacy and accuracy. In summary, anticipated future trends are suggested, and the present shortcomings and gaps within the LU-LDM model are showcased. This review details a complete LU-LDM framework, anticipated to serve as a crucial technical reference for employing this technology in extensive, thin-walled structures.
Dietary salt (sodium chloride) can have its salty character intensified through the addition of particular substances. Food manufacturers have used this effect in salt-reduced foods to inspire healthier eating behaviors. Therefore, a neutral evaluation of the salt level in food, derived from this consequence, is indispensable. immune stimulation Prior research has explored sensor electrodes incorporating lipid/polymer membranes and sodium ionophores for assessing the enhanced saltiness stemming from branched-chain amino acids (BCAAs), citric acid, and tartaric acid. This study details the development of a novel saltiness sensor, based on a lipid/polymer membrane, to quantify the enhancement of saltiness perception by quinine. A different lipid, replacing a previously used lipid which unexpectedly reduced initial readings, was crucial to achieving reliable results. Subsequently, the lipid and ionophore concentrations were adjusted to achieve the desired outcome. Both NaCl samples and those augmented with quinine displayed logarithmic reactions. The application of lipid/polymer membranes to novel taste sensors, as indicated by the findings, allows for an accurate assessment of the saltiness enhancement.
In agricultural contexts, soil color is a substantial factor in evaluating soil health and recognizing its properties. Within the respective fields of archaeology, science, and agriculture, Munsell soil color charts are broadly employed. The reliability of soil color determination using the chart is challenged by subjective interpretation and the possibility of mistakes. This study employed popular smartphones to digitally determine soil colors, drawing upon images from the Munsell Soil Colour Book (MSCB). The captured soil color data is then compared to the true color, determined via a commonly employed sensor, the Nix Pro-2. Our study has shown that there are variations in the color readings produced by smartphones and the Nix Pro. To tackle this problem, we explored diverse color models and, in the end, established a color-intensity relationship between the Nix Pro and smartphone imagery, examining various distance metrics. Consequently, this investigation seeks to precisely ascertain Munsell soil color from the MSCB by fine-tuning the pixel intensity values in smartphone-captured images.