Research on the Application of Adaptive Parameter Adjustment Strategy in RBF Kernel SVM Image Classification

Support Vector Machine (SVM) is a powerful machine learning model with extensive applications in fields such as image classification. The reasonable setting of kernel parameters is crucial for the performance of RBF kernel functions in SVMs. Traditional methods require manually specifying parameters or obtaining parameters through grid search, which results in high computational costs and unsatisfactory results. Therefore, we propose an adaptive parameter adjustment strategy that dynamically adjusts the RBF kernel parameters based on the characteristics of training data, continuously optimizing the performance of SVM classifiers. We evaluated the proposed strategy on the autonomous vehicle dataset and found that compared with the default setting parameters or grid search parameters, this adaptive strategy can significantly improve the classification accuracy, reaching the best accuracy of 0.992 on the autonomous vehicle dataset. Its computational time is 2.443 seconds and better than grid search. This strategy has the advantages of no manual intervention and high computational efficiency, providing an effective parameter tuning scheme for the application of RBF kernel SVMs in image classification and other fields.