Fully Convolutional Sequence Recognition Network for Water Meter Number Reading

Abstract

One of the most widely used frameworks for image-based sequence recognition is the convolutional recurrent neural network, which uses a convolutional neural network (CNN) for feature extraction and a recurrent neural network (RNN) for sequence modeling. However, the RNN is computationally expensive in both training and inference, which limits its application in time-constrained systems. Some models replace the RNN with an attention mechanism for sequence modeling but still, require expensive iterative computations. In this paper, we argue that a CNN with sufficient depth can capture contextual information, eliminate the need for recurrent operations and thus be fully parallelized. We focus on the problem of water meter number reading (WNR), which is a typical sequence recognition task but has rarely been investigated. We propose a fully convolutional sequence recognition network (FCSRN) for the fast and accurate reading of water meter numbers. Furthermore, we design an augmented loss (AugLoss) function to manage the intermediate states of the digits and effectively improve performance. The experimental results demonstrate that the FCSRN has the ability to capture contextual information and eliminate the need for recurrent layers, and simultaneously requires fewer parameters and less computation. The FCSRN with AugLoss outperforms RNN-based and attention-based models. In addition, AugLoss can effectively improve the performance for RNN-based and attention-based models. Moreover, we constructed and released a dataset that contains 6000 water meter images with labels

PGC-Net: A Light Weight Convolutional Sequence Network for Digital Pressure Gauge Calibration

“…Existing automatic meter reading (AMR) methods usually borrow ideas from scene text recognition and can be divided into character based and whole word based. Current AMR methods usually attempt to recognize text images of a specified meter device, e.g., water meter [6], [7], utility meter [8], [9], gas meter [10], [11] or energy meter [12]. Among them, the methods in [9]- [12] are character based.…”

“…We compared PGC-Net with CRNN [26], FCN [31] and FCSRN [7]. CRNN adopted a tweaked version of VGG network architecture for scene text recognition.…”

“…The associate editor coordinating the review of this article and approving it for publication was Mehul S. Raval. not applicable in AGC due to the following reasons: (1) existing methods usually adopted a deep convolutional network as the image extractor [7]. It is because AMR is not time sensitive and the images utilized in AMR merely need to be periodically recognized.…”

Light-Weight Spliced Convolution Network-Based Automatic Water Meter Reading in Smart City

“…Early DNN based technologies called segmentation-based methods [7], [8], which always include two steps: character segmentation and recognition, e.g., Bissacco et al [8] first use a bounding box to fix the position of number in image and then employ a convolutional neural network (CNN) to recognition objective. Recent DNN based methods are called the segmentation-free methods [9], [10], which utilize recurrent neural network (RNN) to recognizes the number in an image as a whole without fixing the position of objects. Although the DNN based methods achieve good performance in a number recognition task, these still exist several challenges to hinder their applications in industrial environments.…”

Detail-attention convolutional neural network for meter recognition

“…近些年来, 卷积神经网络开始被应用于仪表的读 数识别中. 目前卷积神经网络在天然气表 [8~10] 和水 表 [11,12] 的自动化识别中已经取得了不错的识别效果, 但是水表和天然气表的样式单一, 且工作环境相对简 单. Dai等人 [13] 提出了将卷积神经网络用于温度表的 实时识别, 但仍然局限于特定种类的温度仪表. 贺嘉 琪 [14] 提出了利用神经网络对仪表区域进行检测, 后处 理还是基于经典的阈值分割法和霍夫直线法, 对光照 和噪声等十分敏感.…”