Type-Balanced Federated Learning for Visual Analog Meter Reading

Weida Zhao; Logan Bellamy; Yazhou Tu; Jiaqi Wang

cs.CVSep 25, 2026

Type-Balanced Federated Learning for Visual Analog Meter Reading

Authors: Weida Zhao, Logan Bellamy, Yazhou Tu, Jiaqi Wang

Organizations: Department of Computer Science and Software Engineering, Auburn University, Auburn, USA

Abstract

Analog dial meters are widely deployed in industrial applications and utility sites, where environments and meter types vary and inspection data may be sensitive. Currently, automatic meter readers must be individually developed and deployed for each environment and meter type in practice. Deep learning could handle this variability but requires diverse labeled data that are costly to collect and update. In practice, meter images are distributed across independent sites, each with limited labels, while raw images often cannot be pooled because of ownership, governance, or privacy constraints. To address these challenges, we present a federated framework for visual analog meter reading that enables multiple sites to collaboratively train a reading model without sharing their raw images. Our framework consists of a four-stage pipeline: (1) dial localization, (2) thin-structure segmentation trained federatively across clients, (3) polar unwrapping, and (4) tick-counting decoding for final reading. To enable systematic evaluation of this setting, we release MeterFL, a 1,382-image mask-annotated dataset organized into deployment-motivated pseudo-clients derived from visual attributes via deterministic rules, with dHash near-duplicate control between the segmentation train and test splits. We evaluate both segmentation quality and end-to-end reading accuracy. MeterFL is publicly available at https://github.com/weidazhaoooo/Meter-FL.

Figures & tables

Figure 1: Federated meter-reading framework. (a) Clients exchange only segmentation parameters with a type-balanced server: each instrument type receives equal total weight, with sample-size weighting within types. (b) The shared local reading pipeline, with schematic stage illustrations and solid inference arrows. The gray dotted guide identifies stage 2, the only federatively trained component. Images and masks remain local.

Figure 2: Metadata-driven construction of MeterFL (left) and the resulting pseudo-clients with train/test image counts (right). VLM descriptions are mapped to pseudo-clients by deterministic rules. † Below the minimum-size gate (gray).

		Segmentation		Meter reading
Training	Method	Avg IoU	IoU STD	Med. %FS	$\leq$ 5%FS	Acc
Centralized	Frontier VLM (zero-shot)	–	–	18.0	7.4	49.1
	Qwen2.5-VL-7B (zero-shot)	–	–	18.4	9.7	35.3
	Qwen2.5-VL-32B (zero-shot)	–	–	17.2	20.1	21.9
	Qwen2.5-VL-72B (zero-shot)	–	–	13.4	13.4	45.4
	InternVL3.5-8B (zero-shot)	–	–	13.5	17.1	51.7

Table 1: Centralized versus federated training on MeterFL: average and population STD of per-client pointer IoU (eight clients, 272 test images; mean

\pm

std over 3 seeds) and median %FS, share within 5%FS, and Acc (269 images; means over 3 seeds). VLMs have no segmentation columns. Bold: best per regime.

Figure 3: Four-stage reading of a real meter image: (1) the orange ROI is cropped and resized, shown enlarged relative to the close-up input; (2) tick and pointer segmentation; (3) circular sampling and polar unwrapping; (4) tick-count decoding. The pointer is 10 of the 20 tick intervals from zero ( $\hat{f}=0.5$ ). Each interval represents 0.005 MPa, giving a reading of $10\times 0.005=0.05$ MPa, which matches the ground truth.

Figure 4: Effect of initialization on average pointer IoU over the 272 test images of the eight clients. Bars show means over three seeds; error bars show seed standard deviations.

Figure 5: Pointer-IoU controls. (a) Full-data training; Pool/Bal.: centralized training without/with type-balanced sampling. (b) Local-model transfer. (c,d)

K{=}20

TB controls with synthetic and ImageNet initialization, respectively. Error bars: reported seed STD.

Federated Learning

Federated learning trains shared models across data holders without requiring them to pool their raw data.

See all 961 papers

The Impact of Federated Learning on Distributed Remote Sensing Archives
Apr 13, 2026Anand Umashankar, Karam Tomotaki-Dawoud, Nicolai SchneiderFederated Learning Non-IID Federated Learning
Technische Universität Berlin Berlin, Germany
FedTR: Federated Learning Framework with Transfer Learning for Industrial Visual Inspection
Jul 9, 2026Vikash Sathiamoorthy, Shuo Huai, Hao Kong +7Industrial Inspection Optical Character Recognition
HP-NTU Digital Manufacturing Corporate Lab, Nanyang Technological University, Singapore · School of Computer Science and Engineering, Nanyang Technological University, Singapore · Department of Computer Science, Norwegian University of Science and Technology, Norway +2
DeepAFL: Deep Analytic Federated Learning
Feb 28, 2026Jianheng Tang, Yajiang Huang, Kejia Fan +8Gradient-Free Optimization Federated Learning
Peking University, PRC · Key Lab of High Confidence Software Technologies (PKU), Ministry of Education, PRC · Central South University, PRC +4

Apr 13, 2026J/K move · Enter open · S save

The Impact of Federated Learning on Distributed Remote Sensing Archives

Anand Umashankar, Karam Tomotaki-Dawoud, Nicolai Schneider

Technische Universität Berlin Berlin, Germany

Federated Learning Non-IID Federated Learning Federated Learning Aggregation Remote Sensing Image Understanding

Open paper AskPDF arXiv

Remote sensing archives are inherently distributed: Earth observation missions such as Sentinel-1, Sentinel-2, and Sentinel-3 have collectively accumulated more than 5 petabytes of imagery, stored and processed across many geographically dispersed platforms. Training machine learning models on such data in a centralized fashion is impractical due to data volume, sovereignty constraints, and geographic distribution. Federated learning (FL) addresses this by keeping data local and exchanging only model updates. A central challenge for remote sensing is the non-IID nature of Earth observation data: label distributions vary strongly by geographic region, degrading the convergence of standard FL algorithms. In this paper, we conduct a systematic empirical study of three FL strategies -- FedAvg, FedProx, and bulk synchronous parallel (BSP) -- applied to multi-label remote sensing image classification under controlled non-IID label-skew conditions. We evaluate three convolutional neural network (CNN) architectures of increasing depth (LeNet, AlexNet, and ResNet-34) and analyze the joint effect of algorithm choice, model capacity, client fraction, client count, batch size, and communication cost. Experiments on the UC Merced multi-label dataset show that FedProx outperforms FedAvg for deeper architectures under data heterogeneity, that BSP approaches centralized accuracy at the cost of high sequential communication, and that LeNet provides the best accuracy-communication trade-off for the dataset scale considered.

Abstract

Federated Learning

Explore similar work

The Impact of Federated Learning on Distributed Remote Sensing Archives

FedTR: Federated Learning Framework with Transfer Learning for Industrial Visual Inspection

DeepAFL: Deep Analytic Federated Learning