Classification of aircraft images

The aim of this study is to classify military aircraft belonging to a specific group, to evaluate the classification performance and to improve the results. Aircraft classification is important for detecting potential threats from other nations, ensuring the security of national borders, and facilitating effective coordination between commercial flights and military air traffic. The classification algorithms used in this study include the Convolutional Neural Network (CNN), an artificial neural network, and the Support Vector Machine (SVM), a machine learning algorithm. CNN is designed to automatically learn data features by processing different types of input, such as images, through filters or kernels. SVM, on the other hand, is a machine learning algorithm used for classification and regression tasks. It performs classification by identifying the hyperplane that maximises the distance between data points in the feature space. In addition, by using kernel methods, SVM can also provide solutions to non-linear problems. The effectiveness of such classification methods is highly dependent on the size and diversity of the data sets used. In this research, a dataset consisting of 4,990 aircraft images of eight different aircraft types created by Shayan Khos was used. Classification was performed on this dataset using both CNN and CNN-SVM models. To evaluate the performance of these models, various hyperparameters were adjusted and the resulting results were compared. First, the dataset was applied directly to the CNN model and a comparison was made by modifying internal parameters of the CNN, such as the learning option, the initial learning rate and the number of epochs. Subsequently, the CNN architecture was used exclusively for feature extraction and these features were classified using the SVM algorithm and the results were analysed. Data augmentation was applied separately to both the CNN and CNN-SVM models, and the differences in accuracy were observed. The results showed that the scenario where the augmented dataset was fed directly into the CNN model, trained for 80 epochs with an initial learning rate of 0.001 using the sgdm algorithm, achieved the highest performance with a test accuracy of 96.52%.