【摘要】 噪声环境下的入侵事件识别中，噪声干扰导致传统方法特征提取困难，进而影响事件识别准确率。为此，提出一种集成扩张卷积降噪自编码器（DCDAE）和残差神经网络的相位敏感光时域反射计（φ-OTDR）事件识别方法。首先采用DCDAE中的编码器部分，将高维入侵事件信号降维至关键特征表示，再通过解码器重建高质量信号并提取其特征表示，最后将所得特征表示输入残差神经网络中进一步提取特征与分类。研究结果表明，相较于传统识别方法，DCDAE结合残差神经网络的方法对噪声的容忍度更高，识别性能更优。更多还原

【Abstract】 In the recognition of intrusion events in noisy environment, noise interference makes it difficult to extract features by traditional methods, which affects the accuracy of event recognition. To address this problem, a phase-sensitive optical time domain reflectometer(φ-OTDR) event recognition method integrating dilated convolutional denoising autoencoder(DCDAE) and residual neural network is proposed. The proposed method utilizes the encoder part of the DCDAE to reduce the dimensionality from high-dimensional signal to key feature representations. Subsequently, the decoder reconstructs a highquality signal and extracts its feature representation. Finally, these features are input to the residual network for further processing and classification. Compared with traditional recognition methods, the proposed event recognition method based on DCDAE and residual neural network exhibits higher tolerance to noise and superior recognition performance.更多还原