For typical intrusion events of knocking, shaking, climbing, and walking in perimeter security of power system, we propose an intrusion recognition method based on AsymConv-DenseNet for \(\phi \) -OTDR events. This method uses AsymConv-DenseNet to identify these typical intrusion events with high recognition performance and few underreporting. AsymConv-DenseNet uses the differentiated feature extraction of AsymConvBlock and dense links in DenseNet. Through experiments on the open dataset, the maximum accuracy, average precision, average recall and average f1 score of AsymConv-DenseNet are 96.52%, 96.53%, 96.53%, and 96.53%. Compared with DenseNet, SE-DenseNet and CBAM-DenseNet, the value of these evaluate metrics of AsymConv-DenseNet is the highest. AsymConv-DenseNet recognizes the recalls obtained by knocking, shaking, climbing, and walking, which are 99.84%, 90.15%, 90.87%, and 98.33%. The recall of these typical intrusion events of AsymConv-DenseNet is higher than that of DenseNet, SE-DenseNet, CBAM-DenseNet, ResNet-18, GoogLeNet and Swin Transformer. In conclusion, intrusion recognition method based on AsymConv-DenseNet for \(\phi \) -OTDR events that is suitable for perimeter security of power system.