Workplace video surveillance and timely response to operational violations are critical to avoid operator injuries at power construction sites. Here, a system that combines remote substation construction management and artificial intelligence object detection techniques to intellectualize the power construction management process and identify violations during construction in real time is proposed. To improve the detection accuracy, a data augmentation method, including three operations: (1) object segmentation and background fusion; (2) partial erasing; and (3) other basic transformations, is also proposed. Six variants of the You Only Look Once (YOLO) model are trained as detectors for comparative experiments on a dataset collected at the practical construction site. The experimental results show that the detection precision and recall of the YOLOv5‐s model are 0.852 and 0.922, with high accuracy and low miss rate, which meet the requirements of robustness and accuracy in detecting realistic power construction violations.

The visible image contains many high-frequency components that provide texture details with high spatial resolution and definition consistent with human visual perception, but it is easily affected by external factors such as light, weather, and obstructions. On the other hand, the infrared image is a radiation image whose contrast is determined by the temperature difference between the target and the background, and is not easily affected by external conditions. Integrating complementary information from both image types into one image is therefore very useful. In our paper, we propose a dual-frequency cross-enhanced fusion network called DCFusion for infrared and visible image fusion. We design a frequency decomposition module and a frequency enhancement module based on Laplacian of Gaussian for feature decomposition and enhancement, respectively. We then build a dual-frequency cross-enhanced fusion generator network based on these two modules to achieve enhanced fusion. We also use the sum of visible and infrared discriminator and the visible discriminator to balance our fusion results, replacing the traditional single visible discriminator. Our method is an end-to-end model, avoiding the manual design of complex fusion rules like traditional methods. Compared with existing advanced fusion algorithms, our method outperforms most of them in qualitative comparison, quantitative comparison, and target detection accuracy. Finally, the experiment proves that our method can effectively enhance the fusion of the target scene even in harsh environments such as complex lighting, low illumination, and smoke scenes.

In order to separate the sub-signals and extract the feature frequency in the signal accurately, we proposed a parameter-adaptive time-varying filtering empirical mode decomposition (TVF-EMD) feature extraction method based on the improved grasshopper optimization algorithm (IGOA). The method not only improved the local optimal problem of GOA, but could also determine the bandwidth threshold and B-spline order of TVF-EMD adaptively. Firstly, a nonlinear decreasing strategy was introduced in this paper to adjust the decreasing coefficient of GOA dynamically. Then, energy entropy mutual information (EEMI) was introduced to comprehensively consider the energy distribution of the modes and the dependence between the modes and the original signal, and the EEMI was used as the objective function. In addition, TVF-EMD was optimized by IGOA and the optimal parameters matching the input signal were obtained. Finally, the feature frequency of the signal was extracted by analyzing the sensitive mode with larger kurtosis. The optimization experiments of 23 sets of benchmark functions showed that IGOA not only enhanced the balance between exploration and development, but also improved the global and local search ability and stability of the algorithm. The analysis of the simulation signal and bearing signal shows that the parameter-adaptive TVF-EMD method can separate the modes with specific physical meanings accurately. Compared with ensemble empirical mode decomposition (EEMD), variational mode decomposition (VMD), TVF-EMD with fixed parameters and GOA-TVF-EMD, the decomposition performance of the proposed method is better. The proposed method not only improved the under-decomposition, over-decomposition and modal aliasing problems of TVF-EMD, but could also accurately separate the frequency components of the signal and extract the included feature information, so it has practical significance in mechanical fault diagnosis.

Feature matching, referring to establishing high reliable correspondences between two or more scenes with overlapping regions, is of extremely significance to various remote sensing (RS) tasks, such as panorama mosaic and change detection. In this work, we propose an end-to-end deep network for mismatch removal, named hierarchical consensus attention network (HCA-Net), which is one of the critical steps in matching pipeline. Unlike existing practices, our HCA-Net does not rely on global geometric constraints and handcrafted structural representations. The key principle of the proposed HCA-Net is to adaptively enhance neighborhood consensus before evaluating correspondence. To this end, we design a consensus attention mechanism to regularize sparse matches directly. More specifically, consensus attention consists of two novel operations: an encoder–decoder module for calculating compatibility scores and a context-based density representation module. Such attention mechanism can be easily plugged into the existing inlier/outlier classification model in a stacked way to reject outliers. We also propose a hierarchical global-aware network for further improving the accuracy of outlier detection. We compare the proposed HCA-Net with seven state-of-the-art algorithms on several datasets (including various RS images), and the results reveal that our method significantly outperforms the other competitors.