Inshore Ship Detection in Remote Sensing Images via Weighted Pose Voting

Abstract

Inshore ship detection from high-resolution satellite images is a useful yet challenging task in remote surveillance and military reconnaissance. It is difficult to detect the inshore ships with high precision because various interferences are present in the harbor scene. An inshore ship detection method based on the weighted voting and rotation–scale-invariant pose is proposed to improve the detection performance. The proposed method defines the rotation angle pose and the scaling factor of the detected ship to detect the ship with different directions and different sizes. For each pixel on the ship template, the possible poses of a detection window are estimated according to all possible pose-related pixels. To improve robustness to the shape-similar distractor and various interferences, the score of the detection window is obtained by designing a pose weighted voting method. Moreover, the values of some parameters such as similarity threshold and the weight of “V” are investigated.

Existing System

Iterative Censoring Scheme, Spatial Sparse Coding Bag-Of-Words Model, Robust Regularization Path Algorithm.

Proposed System

To improve the detection precision of the inshore ship, this paper proposes a new detection method based on pose weighted voting. To tolerate the variance of the direction and scale, the pose of a detected ship is defined as a 2-tuple vector (α, λ), where α is the rotation angle and λ is the scaling factor related to the target template. For each pixel on the ship template, all possible pose-related pixels in a detection window are firstly searched based on the similarity between their RGA and used to estimate the possible poses of a detection window. And then the score of a detection window is obtained by designing a pose weighted voting method. The weight coefficients of the proposed pose weighted voting method are designed based on two parts of the “V”-shaped structure of the fore and outline continuity factor and discussed. Since the pose estimation and pose weighted voting are independent to each edge pixel in the detection window and the ship template, the proposed detection method is not only rotation–scale invariant but also robust to the shape-similar distractor and the contour noise caused by various interferences such as shadow, wave, and side-by-side. Deep learning technique such as Zou and Shi’s SVDNet can be introduced to further improve the robustness. Moreover, the target pose of the detected ship can also be obtained as a byproduct of the ship detection.

Architecture

BLOCK DIAGRAM