基于间隔损失神经网络的异常翻栏检测方法

Anomaly Fence-Climbing Detection Method Based on Margin Loss Neural Network

摘要: 针对异常翻栏行为检测中的非线性和不均衡性问题，本文提出了一种基于间隔损失的神经网络模型，称为MarginNet模型。首先，采用神经网络来捕捉数据中的非线性特征，以更好地适应数据的复杂性并提高异常行为的识别准确性。其次，引入间隔损失函数作为模型的损失函数，通过最大化正常样本与异常样本之间的间隔来增强模型的泛化能力和鲁棒性。同时，为解决异常行为数据集中的不均衡性问题，使用不均衡因子来调整样本的权重，使得模型更加关注少数类的异常行为，从而提高对异常行为的识别能力。在此基础上，构建了一个用于实时监测和识别翻栏行为中异常情况的异常行为检测系统。最后，实验结果验证了本文所提出的MarginNet模型的有效性。

Abstract: In this paper, we propose a neural network model called MarginNet based on the concept of margin loss to address the issues of nonlinearity and class imbalance in anomaly detection. Firstly, we uti-lize a neural network to capture the nonlinear features in the data, enabling better adaptation to the complexity of the dataset and improving the accuracy of anomaly detection. Secondly, we introduce the margin loss function as the model’s objective, maximizing the margin between normal and anomaly samples to enhance the model’s generalization and robustness. Additionally, we employ an imbalance factor to address the class imbalance problem in the anomaly behavior dataset, assigning higher weights to minority class samples and improving the recognition capability of anomaly behaviors. Based on these contributions, we develop an anomaly behavior detection system for re-al-time monitoring and recognition of anomalies in fence-climbing behavior. Finally, experimental results demonstrate the effectiveness of the proposed MarginNet model in anomaly detection.