Remote sensing image anchored ship detection false alarm elimination method based on DPM algorithm

Abstract

The invention provides a remote sensing image anchored ship detection false alarm elimination method based on a DPM algorithm. The effective anchored ship target false alarm elimination method is constructed for a ship target in the port in an optical remote sensing image through combination of an existing detection algorithm based on morphological property candidate areas. Firstly a ship target training database is established offline and a ship model is trained and initialized; a feature image pyramid is constructed through a HOG algorithm; then root model response scores are calculated on the images of the l-layer resolution of the feature pyramid; the feature images of twice resolution corresponding to the feature images of the l-layer are found, and part model response scores are calculated on the images and a high-score area is extended; and finally the integrated scores of all layers of the feature image pyramid are obtained according to the root model and part model scores, and the position of the ship target on the images of the candidate areas are obtained through iterative loop optimization.

Description

technical field [0001] The invention relates to a processing method of remote sensing images, in particular to a method for eliminating false alarms of target detection of berthing ships in the automatic processing of optical satellite remote sensing images. Background technique [0002] In recent years, with the rapid development of aerospace platforms and sensor technologies, the demand for ship detection in port areas using optical remote sensing images is increasing. This technology can realize the dynamics of shipping safety in large-scale port areas with low labor costs. monitor. The main technical approaches currently used include: optimal entropy automatic threshold method (KSW), dual-parameter constant false alarm (CFAR) algorithm, multi-polarization detection method, etc. [0003] In the existing research, the research on ship detection based on optical remote sensing images mainly focuses on ship detection in ocean scenes. These targets are located in the graysca...

AI Technical Summary

Problems solved by technology

Most of the detection methods for berthing ships in the existing research directly use the extracted target candidate area as the detection result, or assume some simple geometric features for preliminary false alarm elimination. However, there is a complex composition between the berthing ship and the connecting dock. relationship, these existing features are difficult to remove false alarm areas

, leading to the inability to obtain accurate targets of berthing ships. Recently, some improved detection methods of berthing ships, although adopting the idea based on the confirmation of local features of the ship, such as the judgment method of the morphological characteristics of the sharp bow of the ship

However, in practical applications, there are various types of ship targets, and many ship types do not have sharp bow features, such as figure 1 , and there are also facilities with sharper angles in the port

In addition, optical remote sensing images are greatly affected by factors such as equipment platform imaging and weather, and the performance of such methods is difficult to guarantee, which affects the efficiency and accuracy of removing false alarms for ships calling at ports.

Finally, a large number of false alarms generated in the target candidate area detection stage are difficult to eliminate

Images