Ship Detection from Optical Satellite Images Using Convolutional Neural Networks

Neslihan Toprak; Yıldıray Yalman

Makale detayları

Ship Detection from Optical Satellite Images Using Convolutional Neural Networks

Yazarlar:

Neslihan Toprak | Yıldıray Yalman

Yayıncı:

Murat YAKAR

Ana Başlıklar:

Computer Software

Dergi:

Turkish Journal of Engineering Vol.9 Issue 2, pp.342-353

Belge Tipi:

Makale

Yayın Tarihi:

6/2025

Belge Dili:

İngilizce

Document ID (Org No.):

2001339334

pISSN:

2587-1366

Anahtar kelimeler:

Özet:

 Since most of the world is covered with oceans and seas, seas and oceans have aroused peoples curiosity throughout history. Humans have used oceans and seas in versatile ways. The seas are critical areas for trade, transportation, fishing, tourism, energy resources, border security, defense, and intelligence operations. Today, the increasing use of maritime routes creates problems in terms of maritime security, maritime traffic, and management. It has become necessary to look for alternatives to solve such problems in the maritime industry, and deep learning techniques have been used to solve these problems. This paper presents ship detection method from optical satellite images using convolutional neural networks. The motivation of this paper is to produce solutions to the issues of detecting possible dangers in areas with heavy maritime traffic, preventing illegal fishing, preventing pirate attacks, human smuggling, country defense, security and tracking of maritime trade routes with ship detection systems. The convolutional neural network models used in the paper are based on YOLOv8 and YOLOv9 and include different packages of these models. The dataset used in the paper was created using the FGSCR-42 dataset. The dataset used in the paper includes 12 classes. The performance of the model results was compared, and the results are presented in this paper. The mAP50 value of our YOLOv8l model, which we use as a new approach to ship detection studies in the literature, is 98.9. Compared to similar studies in the literature, our model obtains a higher mAP value.

 Since most of the world is covered with oceans and seas, seas and oceans have aroused peoples curiosity throughout history. Humans have used oceans and seas in versatile ways. The seas are critical areas for trade, transportation, fishing, tourism, energy resources, border security, defense, and intelligence operations. Today, the increasing use of maritime routes creates problems in terms of maritime security, maritime traffic, and management. It has become necessary to look for alternatives to solve such problems in the maritime industry, and deep learning techniques have been used to solve these problems. This paper presents ship detection method from optical satellite images using convolutional neural networks. The motivation of this paper is to produce solutions to the issues of detecting possible dangers in areas with heavy maritime traffic, preventing illegal fishing, preventing pirate attacks, human smuggling, country defense, security and tracking of maritime trade routes with ship detection systems. The convolutional neural network models used in the paper are based on YOLOv8 and YOLOv9 and include different packages of these models. The dataset used in the paper was created using the FGSCR-42 dataset. The dataset used in the paper includes 12 classes. The performance of the model results was compared, and the results are presented in this paper. The mAP50 value of our YOLOv8l model, which we use as a new approach to ship detection studies in the literature, is 98.9. Compared to similar studies in the literature, our model obtains a higher mAP value.

 Since most of the world is covered with oceans and seas, seas and oceans have aroused peoples curiosity throughout history. Humans have used oceans and seas in versatile ways. The seas are critical areas for trade, transportation, fishing, tourism, energy resources, border security, defense, and intelligence operations. Today, the increasing use of maritime routes creates problems in terms of maritime security, maritime traffic, and management. It has become necessary to look for alternatives to solve such problems in the maritime industry, and deep learning techniques have been used to solve these problems. This paper presents ship detection method from optical satellite images using convolutional neural networks. The motivation of this paper is to produce solutions to the issues of detecting possible dangers in areas with heavy maritime traffic, preventing illegal fishing, preventing pirate attacks, human smuggling, country defense, security and tracking of maritime trade routes with ship detection systems. The convolutional neural network models used in the paper are based on YOLOv8 and YOLOv9 and include different packages of these models. The dataset used in the paper was created using the FGSCR-42 dataset. The dataset used in the paper includes 12 classes. The performance of the model results was compared, and the results are presented in this paper. The mAP50 value of our YOLOv8l model, which we use as a new approach to ship detection studies in the literature, is 98.9. Compared to similar studies in the literature, our model obtains a higher mAP value.

 Since most of the world is covered with oceans and seas, seas and oceans have aroused peoples curiosity throughout history. Humans have used oceans and seas in versatile ways. The seas are critical areas for trade, transportation, fishing, tourism, energy resources, border security, defense, and intelligence operations. Today, the increasing use of maritime routes creates problems in terms of maritime security, maritime traffic, and management. It has become necessary to look for alternatives to solve such problems in the maritime industry, and deep learning techniques have been used to solve these problems. This paper presents ship detection method from optical satellite images using convolutional neural networks. The motivation of this paper is to produce solutions to the issues of detecting possible dangers in areas with heavy maritime traffic, preventing illegal fishing, preventing pirate attacks, human smuggling, country defense, security and tracking of maritime trade routes with ship detection systems. The convolutional neural network models used in the paper are based on YOLOv8 and YOLOv9 and include different packages of these models. The dataset used in the paper was created using the FGSCR-42 dataset. The dataset used in the paper includes 12 classes. The performance of the model results was compared, and the results are presented in this paper. The mAP50 value of our YOLOv8l model, which we use as a new approach to ship detection studies in the literature, is 98.9. Compared to similar studies in the literature, our model obtains a higher mAP value.

Makale detayları

Ship Detection from Optical Satellite Images Using Convolutional Neural Networks

Yazarlar: Neslihan Toprak | Yıldıray Yalman

Yayıncı: Murat YAKAR

Ana Başlıklar: Computer Software

Dergi: Turkish Journal of Engineering Vol.9 Issue 2, pp.342-353

Belge Tipi: Makale

Yayın Tarihi: 6/2025

Belge Dili: İngilizce

Document ID (Org No.): 2001339334

pISSN: 2587-1366

Convolutional Neural Networks

Ship Detection

Optical Satellite Image

YOLOv8

YOLOv9

Sayı

Yazarlar:

Neslihan Toprak | Yıldıray Yalman

Yayıncı:

Murat YAKAR

Ana Başlıklar:

Computer Software

Dergi:

Turkish Journal of Engineering Vol.9 Issue 2, pp.342-353

Belge Tipi:

Makale

Yayın Tarihi:

6/2025

Belge Dili:

İngilizce

Document ID (Org No.):

2001339334

pISSN:

2587-1366