基于RFID和LSTM的固定资产智能感知方法

针对传统基于射频识别的固定资产感知方法存在射频信号易受环境影响、多个标签互干扰、感知识别精度低等问题，提出一种基于射频识别（RFID）和长短期记忆（LSTM）神经网络的固定资产状态智能感知方法。为克服RFID信号接收时间非连续性导致的识别精度低问题，引入序列分析思想，使用LSTM对一段时间内接收到的RFID信号序列进行模式学习，构建基于RFID和LSTM的固定资产感知模型，实现固定资产位置的二分类辨识。所提方法在高校实验室条件下开展实测实验进行性能验证。结果表明，使用序列分析的固定资产状态感知模型辨识准确率可达 99.26%，比传统基于离散时刻点的 RFID 感知模型辨识准确率高 11.05%，且对多标签的识别准确率可达93.0%。

An Intelligent Perceptual Method of Fixed Assets Based on RFID and LSTM

To solve the disadvantages of traditional RFID-based fixed asset perception methods, such as vulnerability of radio signalsto environmental interference, multiple tag interferences with each other, and low recognition accuracy, an intelligent perceptual method offixed assets based on long and short-term memory neural networks (LSTM) and Radio Frequency Identification (RFID) is proposed.The sequence analysis is introduced to deal with the low recognition accuracy caused by the discontinuity of RFID signal reception, anda LSTM-based model is developed to learn the pattern of RFID signal sequences received over a period of time, which can realize thetwo-category identification of the location of fixed assets. The proposed model has been validated by real experiments carried out in auniversity laboratory. The results show that the recognition accuracy of the fixed asset state using time sequence analysis can reach99.26%, which is 11.05% higher than that of the traditional RFID perception model based on discrete moments, and the recognitionaccuracy of multiple tags can reach 93.0%.