An Optimized CNN-SVM Algorithm for UAV Anomaly Detection
Xiaowei Wang · 2023
In recent years, drones have been widely used in various fields and have achieved remarkable results. However, due to the particularity of the tasks performed by drones, the staff cannot even take effective repair measures when the drone fails. To solve this problem, this paper proposes an optimized UAV (Unmanned Aerial Vehicle) anomaly detection algorithm based on CNN-SVM (Convolutional Neural Network-Support Vector Machine). Using the public drone anomaly detection data set to simplify the feature engineering construction process and model calculation time, we use the CNN model to organize the feature data of the drone failure into a time-series form and input it to the model convolution layer and pooling Floor. Meanwhile, due to the complex working environment, the fault characteristic data of drones is often not obvious. To further improve the fault identification ability, this paper proposes to add SVM for algorithm optimization. The current difference when the drone fails is considered, and the characteristic sequence of the drone when it is not faulty is input into the SVM model to assist in the judgment of abnormal data. To further verify performance, AUC (Area Under Curve) is used for model performance testing. The experimental results prove that the accuracy of the algorithm proposed in this paper is higher than other common deep learning algorithms.