Comparison of CNN and SVM for Ship Detection in Satellite Imagery Nur Jati Lantang Marfu’ah Department of Informatics Faculty of Industrial Technology Yogyakarta Islamic University of Indonesia nurjatilantang@gmail.com Arrie Kurniawardhani Department of Informatics Faculty of Industrial Technology Yogyakarta Islamic University of Indonesia arrie.kurniawardhani@uii.ac.id Abstract— . Satellites with optical sensors generate images of the Earth over relatively large areas. Optical satellite image provides unique insights into various markets, including agriculture, defense and intelligence, and energy. Ship detection using satellite images is crucial because it can help manage marine traffic services, defense and intelligence, and fisheries management. In this study, optical satellite images are used for training models for detecting the ship. Machine Learning (ML) algorithms such as deep learning and Support Vector Machine (SVM) have been applied to detect objects in previous studies. Convolution Neural Network (CNN)-based deep learning technology outperformed many algorithms that have existed to some extent [1]. CNN has proven to be able to outperform SVM to detect ships with an average training accuracy is 0,9912 or 99.12% and the validation accuracy is 0,9798 or 97,89%. While SVM gets an accuracy of 0,9438 or 94,38%. Keywords— Optical Satellite Imagery; Object Detection; Machine Learning; Convolution Neural-Network; Support-Vector-Machine; I. I NTRODUCTION Detection of ships in satellite images has been widely applied in maritime security and sea traffic control [2]. Ship detection using satellite images is very important because it can help manage marine traffic services, defense and intelligence, and fisheries management. Remote sensing has a very important role in monitoring ships because of its long operating distance and wide monitoring range[3]. Optical satellite images have a higher image resolution and more content can be displayed than other remote sensing images, which are more suitable for ship detection. However, optical satellite images usually have two main issues: 1) Weather conditions like clouds, fog, and sea waves produce more pseudo targets for ship detection. 2) Optical satellite images with higher resolution naturally produce a more significant amount of data than other remote sensing images [2]. Today, research in the field of computer vision is very popular. In computer vision contained operations starting from capturing object images by a camera system, processing image objects into a more concise and simple form but still representing objects, until the most important is analyzing to determine the type of object[4] Machine learning algorithms such as deep learning and support vector machine (SVM) have been applied to detect objects in previous studies [2][5][6][7][8][9][10][11][12]. Recently CNN-based deep learning technology outperformed many algorithms that have existed to some extent[1]. SVM was chosen because it is considered as one of the best and uncomplicated initial classification algorithms[5]. Although the Machine Learning algorithm has been widely applied to image classification, an algorithm is not always suitable for every data type or image type. Therefore this study aims to compare which algorithm is better for detecting ship objects between CNN and SVM. II. L ITERATURE R EVIEW A. Optical Satellite Imagery Satellite imagery became publically available in 1972 and led to the founding of NPA Satellite Mapping (NPA) as a consultancy in the same year. Since then, the evolution in the capabilities of both optical satellites and data processing has been staggering. Satellites with optical sensors generate images of the Earth over relatively large areas. Recently interest in remote sensing systems using satellite images was growing, such as in maritime security, traffic control, fisheries surveillance, illegal disposal of oil waste, and marine pollution [13]. Optical satellite images provide unique insights into various markets, including agriculture, defense and intelligence, and energy [1] [13] [14][15][16][17][18]. B. Support-Vector-Machine-(SVM) SVM is one of the best classification algorithms and is not as complicated as Deep Learning [5]. Support vector machine aims to find the hyperplane that maximizes distances between the hyperplane and the support vectors (the closest data points)[19]. In other words, there is labeled training data (supervised learning), the algorithm produces an optimal hyperplane that categorizes new examples. In the two dimensional spaces, this hyperplane is a line separating an airplane into two parts where in each class are located on both sides.