25 July 2022 in IEEE Transactions on Instrumentation and Measurement
Reading power equipment meters often requires loads of manpower, which is a trivial, repetitive, and error-prone task. While conventional automated recognition methods using computer vision (CV) techniques are inflexible under diverse scenarios, in this article, we propose a lightweight meter recognition method that combines deep learning and traditional CV techniques for automated meter reading. For meter detection, an adaptive anchor and global context (GC) module are deployed to improve the feature extraction ability of lightweight backbone without increasing computational cost. Then, an feature pyramid network (FPN) and a path aggregation network (PANet) are developed to realize the information interaction between different feature layers and achieve multiscale prediction. Our method also includes a multitask segmented network to read the detected meters, accelerating the detection speed. Experiments demonstrate that our proposed method can achieve a detection speed of 123 frame per second (FPS) in GeForce GTX 1080 and can obtain an accuracy of 88.2% mean average precision (mAP)50:95. In the case of insufficient training samples, the method can still achieve an accuracy of 80.9% mAP50:95. In addition, we build a power meter images (PMIs) dataset, which contains 1800 images in real scene. The dataset and method we proposed can help with further upgrades of traditional substations. In the future, we also hope to extend the algorithm to edge computing cameras for substations. The newly developed dataset and code are available at https://github.com/zzfan3/electric_meter_detect_recognize .
17 July 2023 in IEEE Transactions on Artificial Intelligence
The K-Nearest Neighbor (KNN) classifies unlabeled samples according to the parameter $k$ , which is a user-defined constant and usually depends on prior knowledge. The selection of $k$ is crucial, as the size of the sample neighborhood affects the classification accuracy. To tackle this issue, we introduce the adaptive KNN (AKNN), which constructs a decision tree to assign different numbers of $k$ -values to different samples. In AKNN, we use the sample label information to calculate the weight between samples. Furthermore, to extend AKNN to a multi-view scenario, we propose a method namely multi-view adaptive KNN (MVAKNN), which integrates information from every single view by using the Dempster-Shafer (D-S) theory. We conduct experiments on three benchmark multi-view image datasets and the results show that MVAKNN exhibits desirable classification accuracy, outperforming some single-view and multi-view methods. Experiments with Gaussian noises show the robustness of the proposed method. The code of our paper is available at https://github.com/zzfan3/Multi-view-Adaptive-KNN .
15 April 2024 in Information Sciences
28 February 2024 in Pattern Analysis and Applications
3 April 2025 in IEEE Transactions on Fuzzy Systems
With the rapid advancement of data acquisition technologies, multiview data have been widely applied in fields such as social networks, computer vision, and natural language processing. Multiview data typically contain information arising from different views or sensors, offering more perspectives for observation. The multiview nature also brings challenges, such as high dimensionality, noise, heterogeneity, and redundancy. Particularly, in scenarios with limited labeled data, traditional single-view learning methods often struggle to handle these complex issues. To address this, this article proposes an unsupervised multiview learning framework that integrates Takagi–Sugeno–Kang fuzzy systems and graph-based adaptive regularization (MvTSK-GAR) to handle the heterogeneity and redundancy in multiview data effectively. Specifically, this article first captures the uncertainty in multiview data through fuzzy rules and models the structural relationships between the data using graph-based adaptive regularization. The framework does not rely on a large amount of labeled data. Instead, it integrates complementary information coming from different views to automatically mine latent patterns, thus generating more accurate and stable data representations. Experimental results demonstrate that the proposed framework performs well in various real-world applications, particularly excelling in high-dimensional data processing and noise reduction. Good performance in multiple publicly available datasets validates the effectiveness of our approach.
11 December 2025 in Applied Soft Computing