A Bagging-SVM field-road trajectory classification model based on feature enhancement

Field -road trajectory classification, in which the semantic labels of unknown trajectory points are predicted by learning the features of agricultural machinery trajectories, has recently received considerable attention in agriculture. At present, most agricultural machinery trajectory samples have imbalanced data distribution problems, and existing field -road trajectory classification methods, which usually use a small set of trajectory features, cannot fully mine the potential information of trajectories, resulting in a low classification accuracy. To address these problems, this paper proposes a Bagging-SVM field -road trajectory classification model based on feature enhancement. First, we use oversampling and undersampling methods to obtain balanced agricultural machinery trajectory data. Second, local features and global features are defined to perform indepth mining of the spatiotemporal information. Specifically, the local features are derived by calculating the attributes of the dataset, and then the global features are calculated from the local features using the statistical magnitude. Next, principal component analysis (PCA) is employed to identify significant features and decrease the dimensionality. Finally, we introduce the bagging integrated learning method in support vector machines to construct Bagging-SVM as a classifier for field -road trajectory classification. To verify the effectiveness of the proposed method, we conduct experiments using 20 trajectory samples with high -frequency sampling frequencies and 20 trajectory samples with low-frequency sampling frequencies, and the classification F1 score of this method reach 97.01% and 98.71% on two trajectory dataset, respectively. The experimental results show that our method outperforms existing field -road trajectory classification methods.