Detection method of variable-scale infrared small targets based on fast automatic density clustering

Abstract

A variable-scale infrared small target detection method based on fast automatic density clustering, involving image analysis and image understanding. The SURF operator is used to solve the feature extraction of scale-variable and small targets. Solve the problem of the sensitivity of features to the environment; the fast automatic density clustering algorithm, for the problem domain, directly completes the segmentation and extraction of trajectories from the airspace, overcoming the exponential complexity problem caused by data fusion in the original sequence detection problem. It solves the over-segmentation and under-segmentation problems in the clustering process, and ensures the integrity and independence of trajectory extraction and the automatic selection of cluster centers. In the later trajectory extraction, the backtracking algorithm is used to find the optimal solution, and the smooth and invariant constraints of the trajectory are integrated into the design of the pruning function to quickly cut off irrelevant clutter branches and speed up the search for solutions. Combining the robust feature detection operator and backtracking strategy, it solves the detection problem of variable-scale infrared small targets and improves the real-time and robustness of the algorithm.

Description

technical field [0001] The invention relates to image analysis and image comprehension, in particular to a detection method for variable-scale infrared small targets based on fast automatic density clustering. Background technique [0002] A series of problems such as detection, tracking, and recognition of small infrared targets mainly come from the Infrared Search and Track (IRST) system. Its main task is to effectively detect, identify and track interesting targets such as aircraft and ships in complex scenes such as sky and sea. Due to its long distance, the target image in the image plane is generally only a few pixels in size, lacking information such as shape and texture. In the environment of Signal-to-Clutter Ratio (SCR), it is a typical weak target. For a long time, it has been the difficulty and focus of the field of automatic target recognition (Automatic Target Recognition, ATR). [0003] The Track Before Detection (TBD) method based on sequential images becam...

AI Technical Summary

Problems solved by technology

[0008] The main defect of the density clustering algorithm is that it cannot automatically determine the cluster center (it still needs to be manually selected) and cannot overcome two types of errors caused by cluster segmentation: over-segmentation and under-segmentation.

These two kinds of wrong clustering will lead to low detection probability and high false detection rate (False detection)

Images