基于深度学习和图像识别的电力配件智能出入库

Intelligent Storage and Retrieval of Power Accessories Based on Deep Learning and Image Recognition

针对电力配件种类繁多、型号各异,依靠射频识别(radio frequency identification,RFID)技术开展电力配件出入库管理,不能覆盖所有电力配件,容易出现出入库、退库不准确、效率低,以及出入库电力配件质量不满足生产要求的问题,开展基于机器学习和图像识别的电力配件智能识别研究。首先采用灰度处理、二值化等方法对原始图像进行处理,之后通过最小外接矩形校正原始图像,然后以卷积神经网络(convolutional neural networks,CNN)、卷积递归神经网络(convolutional recurrent neural network,CRNN)等深度神经网络为核心,结合CTC损失函数,构建适用于识别电力配件的深度学习模型,并依据图像识别吻合度,同步推荐疑似配件设备。通过智能设备采集电力配件图像,采用上述方法实时识别配件名称、型号,提示外形尺寸、适用范围、产品用途。实验结果表明,基于机器学习和图像识别的电力配件智能识别结果准确率达到95%,显著提升了仓储出入库管理的智能化水平。

The power accessories have various types and models,and the storage and retrieval management with RFID technology cannot cover them all,which often leads to the inaccuracy and low efficiency of power accessories storage and retrieval,and the management quality not to meet the production requirements.In view of these problems,we carry out a research on intelligent recognition of power accessories based on machine learning and image recognition to correct the deficiency of RFID technology for storage and retrieval management of power accessories.Firstly,the gray-scale processing and binarization methods are used to process the original images,and the minimum circumscribed rectangle is used to correct the original images.Secondly,a deep learning model suitable for identifying power accessories is constructed using CNN and CRNN deep neural networks with combination of CTC loss function,and the suspected accessories are recommended synchronously according to the image recognition coincidence.The images of power accessories are acquired by intelligent equipment,and their name and model are identified in real time using the proposed methods,with prompt of their overall dimension,application scope and product use.The experimental results show that the accuracy of the intelligent recognition of power accessories based on machine learning and image recognition reaches 95%,which significantly improves the intelligent level of warehousing management.