Comparative performance analysis of convolutional neural network and support vector machine for automatic object recognition based on computer vision

Abstract

Object recognition is one of the fundamental applications of computer vision and plays an important role in various intelligent systems. This study aims to implement and compare the performance of Convolutional Neural Network (CNN) and Support Vector Machine (SVM) for automatic object recognition using the CIFAR-10 and ImageNet (subset) datasets. The novelty of this study lies in providing a comprehensive comparative evaluation of CNN and SVM across datasets with different levels of image complexity while simultaneously analyzing classification performance and computational efficiency using multiple evaluation metrics. The CNN model was implemented using the PyTorch framework with the Adam optimizer, whereas the SVM model employed the Radial Basis Function (RBF) kernel. Performance was evaluated using accuracy, precision, recall, F1-score, confusion matrix, and computational time. Experimental results show that CNN consistently outperformed SVM on both datasets. On CIFAR-10, CNN achieved an accuracy of 91.8%, compared with 83.6% for SVM. On the ImageNet subset, CNN obtained an accuracy of 79.6%, whereas SVM achieved 65.4%. CNN also demonstrated higher precision, recall, and F1-score, indicating superior feature extraction and generalization capabilities for complex image classification tasks. In contrast, SVM required lower computational resources and shorter training time, making it suitable for resource-constrained environments. These findings demonstrate that CNN is the preferred approach for high-accuracy object recognition, while SVM remains a practical alternative when computational efficiency is prioritized. The results provide practical guidelines for selecting appropriate classification methods based on dataset complexity and available computational resources.