Underwater sonar image target detection method based on improved YOLOv3-tiny

Abstract

The invention discloses an underwater sonar image target detection method based on improved YOLOv3-tiny. The method comprises the steps: firstly employing basic graphic transformation to enhance sonar image data, and obtaining a sufficient number of sonar image data sets; then, an improved model of a YOLOv3-tiny network fusing shallow-level features and high-level features is constructed, and detection is carried out through multi-scale feature fusion; on this basis, a network structure is optimized according to a feature map of a YOLOv3-tiny prediction layer, prediction branches with poor feature expression ability are deleted, the detection speed of the improved YOLOv3-tiny is increased, and the real-time performance of detection is ensured; and finally, adopting a non-maximum suppression method for all predicted target categories and position results, and outputting the predicted target category and position with the maximum confidence coefficient. The method is high in accuracy and reduces missing detection of the target.

Description

technical field [0001] The invention belongs to the technical field of image processing, and in particular relates to an underwater sonar image target detection method. Background technique [0002] Due to the particularity of the underwater environment, the loss of electromagnetic waves propagating on the seabed is much greater than that on land, and the distance of traditional optical detection methods is limited underwater. However, sound waves can travel long distances in the deep sea, and acoustic detection methods have incomparable advantages over optical detection methods. Imaging sonar can obtain more complete, rich and detailed underwater data by collecting reflected echoes from the surface of objects for real-time imaging, and is an important tool for underwater environment perception. [0003] At present, research methods for underwater target detection can be divided into two categories: one is traditional target detection algorithms, most of which are improved ...

AI Technical Summary

Problems solved by technology

[0004] However, in the existing underwater target detection technology, there are problems such as poor quality of sonar images, small quantity, and no public data sets, resulting in low accuracy of detection algorithms, insufficient application of deep learning data, false detection and missed detection due to noise interference, etc.

Images