Performance evaluation of CNN and SVM in computer vision-based animal image classification
Keywords:
Convolutional Neural Network, Support Vector Machine, comparative analysis, image classification, background removalAbstract
Manual animal image classification has limitations in terms of efficiency and consistency, particularly when dealing with large-scale image datasets. This study aims to compare the performance of Convolutional Neural Network (CNN) and Support Vector Machine (SVM) with the Radial Basis Function (RBF) kernel for animal image classification in a computer vision system. The experiment was conducted using the Animals-10 dataset, which consists of 26,179 images, with an 80:20 ratio for training and testing data. Two experimental scenarios were evaluated, namely the original dataset and the dataset processed using background removal, to examine the impact of image preprocessing on the performance of both algorithms. The results show that CNN achieved the highest classification accuracy of 70.68%, while SVM with the RBF kernel achieved 49.50% on the background-removed dataset. Background removal improved the accuracy of SVM by 25.21%, whereas CNN showed only a 0.21% improvement. These findings indicate that background removal is more beneficial for improving SVM performance than CNN, suggesting that image preprocessing techniques should be selected according to the characteristics of the classification algorithm.
