A multi-invasion event recognition method based on M-SSCNN-CAResNet

Focusing on the problem of low classification accuracy of existing perimeter intrusion event recognition methods based on the principle of optical fiber sensing, this study introduces an intrusion recognition method founded on a multiscale sliding convolution residual network model (M-SSCNN-CAResNet). Firstly, the one-dimensional vibration signal is converted into a Gramian Angular Difference Field (GADF) image and processed into a grayscale image. This data processing method is used to highlight the characteristics of the signal over time. Secondly, the processed image is input into the M-SSCNN-CAResNet network for training, which uses multiscale sliding convolution with residual structure to extract deeper features of the image. The Coordinated Attention (CA) module used simultaneously can preserve the positional information of the elements in the feature map. The findings indicate that the approach presented in this paper is capable of achieving a recognition accuracy of 99.54% when classifying six types of perimeter intrusion events. Additionally, the test duration for a single test dataset is 34.36 milliseconds. This method improves testing accuracy and reduces testing time.