Search

Nonwoven fabric defect detection is an essential part of the production process, and it is important to realize fast and accurate nonwoven fabric defect detection to improve production efficiency. Aiming at the problems of nonwoven fabric defect detection in which most defect targets are small and traditional industrial cameras are unable to recognize foreign impurities of the same color, a nonwoven fabric defect detection method based on hyperspectral imaging and improved YOLOv5 is proposed. Specifically, hyperspectral imaging technology is used to replace traditional vision and solve the problem of homochromatic foreign matter impurities from the dimension of spectrum. In addition, the LSK attention module is introduced into the YOLOv5 backbone network, which enables the network to learn the key feature information and enhance the detection of small targets. Finally, an improved light repetitive group frequency permutation network (Light-RepGFPN) is proposed in the neck structure, by which the feature fusion capability of the model is enhanced, so that the high-level semantic information and the low-level spatial information are fully fused to improve the detection accuracy. Combined with the LSK and Light-RepGFPN modules, we propose an improved YOLOv5 (LL-YOLOv5) defect detection network. It is experimentally proved that LL-YOLOv5 can achieve an average accuracy mean 90.3% in the nonwoven fabric hyperspectral image defects dataset, which is 2.2% higher than that of the original model, respectively.

Electromagnetic inverse scattering (EMIS) is uniquely positioned among many inversion methods because it enables to image the scene in a contactless, quantitative and super-resolution way. Although many EMIS approaches have been proposed to date, they usually suffer from two important challenges, i.e., time-consuming data acquisition and computationally -prohibitive data post processing, especially for large-scale objects with high and even moderate contrasts. To tackle the challenges, we here propose a framework of intelligent EMIS with the aid of deep learning techniques and information metasurfaces, enabling to the efficient data acquisition and instant data processing in a smart way. Towards this goal, as illustrative examples, we considerably extend the canonical contrast source inversion (CSI) algorithm, a canonical EMIS method by updating the contrast via the generative adversarial network (GAN), an unsupervised deep learning approach, leading to a novel physics-informed unsupervised deep learning method for EMIS, referred to as CSI-GAN in short. Compared with existing deep learning solutions for EMIS, our method relies on the supervision of physical law instead of the labeled training dataset, beating the bottleneck arising from the collection of labeled training datasets. Furthermore, we propose a scheme of adaptive data acquisition with the use of information metasurface in a cost-efficiency way, remarkably reducing the number of measurements and thus speeding up the data acquisition but maintaining the reconstruction's quality. Illustrative examples are provided to demonstrate the performance gain in terms of reconstruction quality, showing the promising potentials for providing the intelligent scheme for the EMIS problems.

This paper proposed an improved full-order sliding-mode observer (IFSMO), applied to a sensorless control system of surface-mounted permanent magnet synchronous motor (SPMSM), to obtain a high-precision rotor speed and position. First, the IFSMO method, which combines the new sliding-mode function with the variable boundary layer function, is presented to suppress the chattering and accelerate the convergence speed of the system. Meanwhile, since the observer has the characteristics of a second-order low-pass filter, the high-frequency noise contained in the estimated back EMF signal can be effectively filtered out without an additional low-pass filter. Then, the proposed PLL is designed to obtain rotor position and speed. Finally, simulations and experiments of the SPMSM system based on the IFSMO method under different working conditions are implemented, which demonstrates the effectiveness of the proposed algorithm.

Moving shadow elimination plays an important role in the field of moving object detection. However, the accuracy of shadow elimination is an open question, due to illumination and complex texture. Furthermore, the problem of misclassification of moving object caused by shadow has also become increasingly serious. To address this problem, this paper presents a moving shadow elimination algorithm based on the fusion of multi-feature pattern, which can enhance the accuracy of moving object detection system. In this approach, a dual-channel HSV color space feature and a uniform extended scale invariant local ternary pattern (UESILTP) texture feature are synthesized to elimination shadow. It greatly overcomes the misjudgment of dark object by color feature and the false detection caused by inconspicuous texture characteristics of moving object. Meantime, a method of region growth is adopted to fill the existing cavities in the color space. Finally, qualitative and quantitative comparisons with state-of-the-art methods show that the algorithm is effective.