Weak and small target detection and tracking method

Abstract

The invention relates to the technical field of computer vision and image processing, in particular to a weak and small target detection and tracking method, which comprises the following steps: S1, constructing an image data training set; s2, optimizing an inquiry type attention library; s3, adaptive filtering is carried out; and S4, carrying out progressive association. Firstly, weak and small target image information is obtained from a public data set image and a manual marking image, and a corresponding anchor-free label is generated so as to construct a weak and small target image database; and constructing a baseline deep learning model by using a residual structure, a feature fusion structure and a coding and decoding architecture. And the historical coordinate pipeline is subjected to diameter self-adaption, so that noise coordinates in the current coordinate pipeline can be filtered. And finally, carrying out association mining updating and adaptive filtering on the candidate coordinate pipelines with low-level thresholds step by step by using the reference coordinate pipelines with higher thresholds, and carrying out continuous iteration to obtain a final coordinate set. Compared with the prior art, the method has great advantages in recall rate, accuracy and comprehensive evaluation index aspects.

Description

technical field [0001] The invention relates to the technical fields of computer vision and image processing, in particular to a weak and small target detection and tracking method. Background technique [0002] Target detection and tracking is still an important method at present because of its low cost and anti-interference. Especially in anti-missile systems, sea surface monitoring, field detection, etc., all potential suspected targets should be discovered as soon as possible, so the detection of weak and small targets is an important step in target detection and tracking. Due to the lack of texture and color information of weak and small targets, the geometric features that can be captured but inconsistent, and the rapidly changing scale, these make the detection of weak and small targets the most challenging task in target detection and tracking systems. [0003] Objects to be inspected often travel between dense clouds and waves in various harsh environments and comp...

AI Technical Summary

Problems solved by technology

Then, the current deep learning methods either pursue the ultimate performance in precision with huge parameters and calculations, or introduce the concept of deep learning conceptually but lack the generalization and robustness of practical applications.

Images