Rack Scale Area Detection Based on the Improved Efficient and Accuracy Scene Text

In the collection of equipment information in the machine room, the location information of the equipment on the rack is often taken by manually observing the scale and estimating it, which is inefficient and highly error-prone. To reduce the misrecognition rate, this paper uses a deep learning method to detect the rack-scale area and provide management personnel with guidance for reading. The current network that can better complete the task of scene character detection has the EAST (Efficient and Accuracy Scene Text). The rack-scale area in this application scenario belongs to small targets, so this network is difficult to meet the detection task of small targets. In this paper, the structure of the EAST is adjusted to address the problem. The ResNet_50 is used to replace the PVANet as the feature extraction network, which expands the number of feature map’s channels and improves the detection ability of small targets. Meanwhile, BLSTM was added to the original network to improve the network's ability to process serialized data. The F-Score value obtained in the experimental test is 85.33%, which meets the task requirements well.