A Remote Calibration Device Using Edge Intelligence

Power system facility calibration is a compulsory task that requires in-site operations. In this work, we propose a remote calibration device that incorporates edge intelligence so that the required calibration can be accomplished with little human intervention. Our device entails a wireless serial port module, a Bluetooth module, a video acquisition module, a text recognition module, and a message transmission module. First, the wireless serial port is used to communicate with edge node, the Bluetooth is used to search for nearby Bluetooth devices to obtain their state information and the video is used to monitor the calibration process in the calibration lab. Second, to improve the intelligence, we propose a smart meter reading method in our device that is based on artificial intelligence to obtain information about calibration meters. We use a mini camera to capture images of calibration meters, then we adopt the Efficient and Accurate Scene Text Detector (EAST) to complete text detection, finally we built the Convolutional Recurrent Neural Network (CRNN) to complete the recognition of the meter data. Finally, the message transmission module is used to transmit the recognized data to the database through Extensible Messaging and Presence Protocol (XMPP). Our device solves the problem that some calibration meters cannot return information, thereby improving the remote calibration intelligence.

Camera-based OCR scene text detection issues: A review

Camera-based scene text detection and recognition is a research area that has attracted countless attention and had made noticeable progress in the area of deep learning technology, computer vision, and pattern recognition. They are highly recommended for capturing text on-scene images (signboards), documents with a multipart and complex background, images on thick books and documents that are highly fragile. This technology encourages real-time processing since handheld cameras are built with very high processing speed and internal memory, are quite easy and flexible to use than the traditional scanner whose usability is limited as they are not portable in size and cannot be used on images captured by cameras. However, characters captured by traditional scanners pose fewer computational difficulties as compared to camera captured images that are associated with divers’ challenges with consequences of high computational complexity and recognition difficulties. This paper, therefore, reviews the various factors that increase the computational difficulties of Camera-Based OCR, and made some recommendations as per the best practices for Camera-Based OCR systems.

Reduced Annotation Based on Deep Active Learning for Arabic Text Detection in Natural Scene Images

<div>- A novel approach is presented to reduced annotation based on Deep Active Learning for Arabic text detection in Natural Scene Images.</div><div>- A new Arabic text images dataset (7k images) using the Google Street View service named TSVD.</div><div>- A new semi-automatic method for generating natural scene text images from the streets.</div><div>- Training samples is reduced to 1/5 of the original training size on average.</div><div>- Much less training data to achieve better dice index : 0.84</div>

Reduced Annotation Based on Deep Active Learning for Arabic Text Detection in Natural Scene Images

<div>- A novel approach is presented to reduced annotation based on Deep Active Learning for Arabic text detection in Natural Scene Images.</div><div>- A new Arabic text images dataset (7k images) using the Google Street View service named TSVD.</div><div>- A new semi-automatic method for generating natural scene text images from the streets.</div><div>- Training samples is reduced to 1/5 of the original training size on average.</div><div>- Much less training data to achieve better dice index : 0.84</div>

CSFF-Net: Scene Text Detection Based on Cross-Scale Feature Fusion

In the last years, methods for detecting text in real scenes have made significant progress with an increase in neural networks. However, due to the limitation of the receptive field of the central nervous system and the simple representation of text by using rectangular bounding boxes, the previous methods may be insufficient for working with more challenging instances of text. To solve this problem, this paper proposes a scene text detection network based on cross-scale feature fusion (CSFF-Net). The framework is based on the lightweight backbone network Resnet, and the feature learning is enhanced by embedding the depth weighted convolution module (DWCM) while retaining the original feature information extracted by CNN. At the same time, the 3D-Attention module is also introduced to merge the context information of adjacent areas, so as to refine the features in each spatial size. In addition, because the Feature Pyramid Network (FPN) cannot completely solve the interdependence problem by simple element-wise addition to process cross-layer information flow, this paper introduces a Cross-Level Feature Fusion Module (CLFFM) based on FPN, which is called Cross-Level Feature Pyramid Network (Cross-Level FPN). The proposed CLFFM can better handle cross-layer information flow and output detailed feature information, thus improving the accuracy of text region detection. Compared to the original network framework, the framework provides a more advanced performance in detecting text images of complex scenes, and extensive experiments on three challenging datasets validate the realizability of our approach.

Scene text removal via cascaded text stroke detection and erasing

Recent learning-based approaches show promising performance improvement for the scene text removal task but usually leave several remnants of text and provide visually unpleasant results. In this work, a novel end-to-end framework is proposed based on accurate text stroke detection. Specifically, the text removal problem is decoupled into text stroke detection and stroke removal; we design separate networks to solve these two subproblems, the latter being a generative network. These two networks are combined as a processing unit, which is cascaded to obtain our final model for text removal. Experimental results demonstrate that the proposed method substantially outperforms the state-of-the-art for locating and erasing scene text. A new large-scale real-world dataset with 12,120 images has been constructed and is being made available to facilitate research, as current publicly available datasets are mainly synthetic so cannot properly measure the performance of different methods.