ABSTRACT As the volume of marine transportation increases, it becomes increasingly important to monitor ships for efficient coastal monitoring and management. To this end, high-resolution satellite images can be utilized to surveil oceanic environments synoptically. In this study, high-resolution optical satellite image was used to detect ships and estimate the size of each ship in the Korean coastal region. All the pixels in an image were first classified into ship, ship shadow, wake, sea, and land by applying a maximum likelihood classifier. The positions corresponding to the boundary of the ship were obtained from the magnitude of the 2-dimensional gradient on the classified ship pixels, and then the length and width of the ship were estimated by applying an ellipse fitting method to the ship periphery. This method resulted, in slight overestimations of the sizes of the ships. In order to improve the accuracy of the estimated ship sizes, a correction formula was developed by investigating the errors of the estimated values and their potential relationships to the variables representing the spatial shape of the vessels, such as eccentricity, kurtosis. Applying the suggested formulation for ship size estimation improved accuracy by 54.41% compared to the estimated sizes obtained through ellipse fitting. We anticipate that our method of estimating the lengths of the vessels will contribute to identifying missing ships using high-resolution satellite images.

中文翻译：

---

利用船舶外围椭圆特征从卫星光学图像估算船舶尺寸

摘要随着海上运输量的增加，对船舶进行有效的沿海监测和管理变得越来越重要。为此，可以利用高分辨率卫星图像对海洋环境进行天气监测。在这项研究中，高分辨率光学卫星图像被用于检测船只并估计韩国沿海地区每艘船的大小。首先通过应用最大似然分类器将图像中的所有像素分类为船舶、船舶阴影、尾流、海洋和陆地。根据分类后的船舶像素上的二维梯度大小得到船舶边界对应的位置，然后对船舶外围应用椭圆拟合方法估计船舶的长宽。这种方法导致，略微高估了船只的尺寸。为了提高估计船舶尺寸的准确性，通过调查估计值的误差及其与代表船舶空间形状的变量（如偏心率、峰度）的潜在关系，制定了修正公式。与通过椭圆拟合获得的估计尺寸相比，应用建议的船舶尺寸估计公式将准确度提高了 54.41%。我们预计，我们估算船只长度的方法将有助于使用高分辨率卫星图像识别失踪船只。通过调查估计值的误差及其与代表血管空间形状的变量（如偏心率、峰度）的潜在关系，制定了修正公式。与通过椭圆拟合获得的估计尺寸相比，应用建议的船舶尺寸估计公式将准确度提高了 54.41%。我们预计，我们估算船只长度的方法将有助于使用高分辨率卫星图像识别失踪船只。通过调查估计值的误差及其与代表血管空间形状的变量（如偏心率、峰度）的潜在关系，制定了修正公式。与通过椭圆拟合获得的估计尺寸相比，应用建议的船舶尺寸估计公式将准确度提高了 54.41%。我们预计，我们估算船只长度的方法将有助于使用高分辨率卫星图像识别失踪船只。