Optical remote sensing image ship detection method based on depth learning single-step detector

Abstract

The invention discloses an optical remote sensing image ship detection method based on a depth learning single-step detector and mainly solves problems of the limited ability of artificially-designedcharacteristic representation in traditional ship detection methods, insufficient interference resistance capability against the complex background and tedious and slow detection process. The method comprises steps that 1), a remote sensing image data set is established; 2), VGG-16 is utilized as a characteristic extraction portion, a multipath residual block is constructed and is taken as a prediction portion, and a new loss function is constructed; 3), the characteristic extraction portion, the prediction portion and the loss function form a single-step detector; 4), the established data setis utilized to train the single-step detector; and 5), ship detection is carried out thorugh utilizing the trained single-step detector. The method is advantaged in that the method is simple for realization, a ship target in the complex background can be rapidly and accurately detected, and the method is applicable to multiple types of real-time ship detection systems.

Description

technical field [0001] The invention belongs to the technical field of optical remote sensing image processing, and mainly relates to an optical remote sensing image ship detection method, which can be used for target recognition under various sea conditions and complex backgrounds of ports and coasts. Background technique [0002] In recent years, with the rapid development of remote sensing observation technology, abundant optical remote sensing image data resources have promoted the progress of ship target detection and recognition technology. Accurate and fast ship detection technology is useful in monitoring port shipping traffic, assisting in the rescue of ships in distress, cooperating with supervision and cracking down on illegal activities such as illegal fishing, illegal dumping of oil pollution, and smuggling, monitoring enemy port deployment and dynamics, and obtaining maritime combat intelligence. Wide range of applications. [0003] For the ship detection meth...

AI Technical Summary

Problems solved by technology

[0008] 1. Since the actual optical remote sensing image is affected by factors such as illumination, weather, and imaging conditions, different ships have various features in the image, so it is difficult to accurately represent ship information with traditional artificially designed features;

[0009] 2. Ship targets usually have a symmetrical long strip structure. In complex backgrounds such as ports and rivers and seas, artificial buildings such as docks, houses, and container arrays on land, as well as small islands and huge waves in the sea are likely to interfere with the detection. ;

[0010] 3. The traditional ship target detection technology usually consists of three parts: sea and land segmentation, candidate region extraction, and target classification. The processing steps for information-rich optical remote sensing images are cumbersome and slow

Although CNN can enhance the feature representation of the target and solve the above two problems to a certain extent, the input of CNN is the small target area in the original image, ignoring the role of background information in the target detection process; and CNN cannot complete Positioning tasks, so this method cannot perform end-to-end detection, and still cannot solve the third problem described

Images